<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Apryse - Medium]]></title>
        <description><![CDATA[Apryse provides SDKs for viewing, editing, and extracting data from documents, helping developers build smarter document workflows across web, server and mobile platforms. - Medium]]></description>
        <link>https://medium.com/apryse?source=rss----f72464c16b46---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Apryse - Medium</title>
            <link>https://medium.com/apryse?source=rss----f72464c16b46---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 12:53:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/apryse" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How to Extract Barcodes from Incoming PDFs into a Database With C#]]></title>
            <link>https://medium.com/apryse/how-to-extract-barcodes-from-incoming-pdfs-into-a-database-with-c-dd7661b5a603?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/dd7661b5a603</guid>
            <category><![CDATA[database]]></category>
            <category><![CDATA[csharp]]></category>
            <category><![CDATA[barcode]]></category>
            <dc:creator><![CDATA[Garry Klooesterman]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 20:28:08 GMT</pubDate>
            <atom:updated>2026-11-18T20:28:08.534Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-7WpUb-8-JjMRD_IXJN8OQ.jpeg" /></figure><p>To send barcode data from incoming PDFs to a database, run the Apryse Server SDK Barcode Extraction Module in a C# service on your own Windows or Linux server. For each PDF, call BarcodeModule.ExtractBarcodesAsString to get every barcode as JSON, then write each value and its type to your table. To handle volume, initialize the SDK once and process one document per thread.</p><h3>Why Barcodes from Incoming PDFs Belong in a Server-Side Pipeline</h3><p>It is six in the morning at a freight forwarder’s dock office, and the first scanner run of the day is already done. Every bill of lading is now a multi-page PDF, and the barcodes on its pages name the shipment, the container, and the seal. Until a person keys those values into the shipment table, dispatch cannot match the load, and every typo sends a pallet to the wrong lane.</p><p>Bills of lading, patient forms, lab reports, and invoices all arrive as PDFs with barcodes that tie each page to a record. A backend barcode recognition pipeline reads those codes and lands them in a table, so matching and routing start without anyone keying values by hand. Teams working with <a href="https://apryse.com/blog/barcode-sdk-healthcare">barcodes in patient records and claims</a> build the same intake service.</p><p>The naive build starts with an open-source decoder. Those libraries decode images, not PDFs, so you add a separate renderer, rasterize every page, and keep both dependencies aligned on Windows and Linux. The <a href="https://apryse.com/capabilities/barcode">Barcode Extraction Module</a> removes that stack. It is an add-on module for the Apryse Server SDK that detects, extracts, and decodes barcodes directly in PDFs, and it ships with the <a href="https://apryse.com/capabilities/smart-data-extraction">Smart Data Extraction</a> add-on license as a separate module download. It runs on your own servers, so documents are processed locally with no cloud API call and no document content going to Apryse.</p><p>Compared with an open-source decoder, the choice is a license cost against engineering and maintenance time. What you build once is the pipeline: the loop that reads one bill of lading is the loop that reads a folder of them, because you initialize the SDK once and give each document its own thread. Camera capture is a separate job, so this pipeline starts after the PDF exists.</p><p>This post covers Apryse Server SDK integration using C# on Windows or Linux, reading barcodes from incoming PDFs into a database. It does not cover camera scanning, which belongs to the <a href="https://apryse.com/scanbot">Scanbot SDK</a>, or reading barcodes from a single PDF, which the <a href="https://apryse.com/blog/extract-barcodes-from-pdf-csharp">C# how-to for extracting barcodes from a PDF</a> walks through. For other languages or platforms, see the <a href="https://docs.apryse.com/core/guides/get-started">Server SDK get-started guide</a>.</p><h3>Extract Barcodes from PDFs into a Database with C# and the Apryse Barcode Extraction Module</h3><p>The Apryse Barcode Extraction Module returns every barcode in a PDF as one JSON string, so your C# service only has to parse it and insert rows. The module runs on Windows and Linux x64 only.</p><p>This feature requires the Barcode Extraction Module add-on, which is a separate download from the base Server SDK. See the <a href="https://docs.apryse.com/core/guides/info/modules">modules documentation</a> for download and installation instructions.</p><p><strong>License keys are confidential.</strong> Do not commit them to public repositories. Use environment variables or a secrets manager to inject your key at runtime. The code below reads the key from APRYSE_LICENSE_KEY.</p><p>The service below walks a folder of incoming bills of lading, gives each PDF its own thread with standard .NET Parallel.ForEach, reads the barcodes, and saves one row per barcode.</p><pre>// Read barcodes from each incoming bill of lading on its own thread and write them to the database<br>using pdftron;<br>using pdftron.Common;<br>using pdftron.PDF;<br>using System.Text.Json;<br><br>string licenseKey = Environment.GetEnvironmentVariable(&quot;APRYSE_LICENSE_KEY&quot;)<br>    ?? throw new InvalidOperationException(&quot;Set the APRYSE_LICENSE_KEY environment variable.&quot;);<br><br>PDFNet.Initialize(licenseKey);<br>PDFNet.AddResourceSearchPath(&quot;/opt/apryse/PDFNetC/Lib/&quot;);<br><br>var jsonOptions = new JsonSerializerOptions { PropertyNameCaseInsensitive = true };<br>var parallelOptions = new ParallelOptions { MaxDegreeOfParallelism = Environment.ProcessorCount };<br><br>try<br>{<br>    if (!BarcodeModule.IsModuleAvailable())<br>        throw new InvalidOperationException(&quot;Barcode Extraction Module not found. Check the AddResourceSearchPath folder.&quot;);<br><br>    Parallel.ForEach(Directory.EnumerateFiles(&quot;/data/incoming&quot;, &quot;*.pdf&quot;), parallelOptions, billOfLadingPath =&gt;<br>    {<br>        try<br>        {<br>            using PDFDoc billOfLading = new PDFDoc(billOfLadingPath);<br><br>            var extractionOptions = new BarcodeOptions();<br>            extractionOptions.SetBarcodeSearchTypes(BarcodeOptions.BarcodeTypeGroup.e_linear);<br><br>            string extractionJson = BarcodeModule.ExtractBarcodesAsString(billOfLading, extractionOptions);<br>            var extraction = JsonSerializer.Deserialize&lt;ExtractionResult&gt;(extractionJson, jsonOptions);<br><br>            foreach (var page in extraction?.Pages ?? new List&lt;PageBarcodes&gt;())<br>                foreach (var barcode in page.Barcodes ?? new List&lt;BarcodeEntry&gt;())<br>                    IntakeRepository.SaveBarcode(billOfLadingPath, barcode.Type, barcode.Value);<br>        }<br>        catch (PDFNetException ex)<br>        {<br>            // Send the file path and ex.Message to your own error handling<br>            Console.WriteLine($&quot;Failed to process {billOfLadingPath}: {ex.Message}&quot;);<br>        }<br>    });<br>}<br>finally<br>{<br>    PDFNet.Terminate();<br>}<br><br>// Verify these property names against real BarcodeTest output before publishing.<br>// If the decoded value comes back under &quot;text&quot;, use [JsonPropertyName(&quot;text&quot;)] on Value.<br>record ExtractionResult(List&lt;PageBarcodes&gt; Pages);<br>record PageBarcodes(List&lt;BarcodeEntry&gt; Barcodes);<br>record BarcodeEntry(string Type, string Value);<br><br>// Placeholder: replace with your own data access code<br>static class IntakeRepository<br>{<br>    public static void SaveBarcode(string documentPath, string barcodeType, string barcodeValue)<br>        =&gt; Console.WriteLine($&quot;{documentPath} | {barcodeType} | {barcodeValue}&quot;);<br>}</pre><p>PDFNet.Initialize and PDFNet.AddResourceSearchPath run once at startup. The search path points at the parent Lib folder that holds the installed modules, not at the Windows or Linux Barcode folder inside it. PDFNet.Terminate runs in the finally block at shutdown. Replace the Lib path and the incoming folder with your own paths.</p><p>Each bill of lading opens in its own PDFDoc, so nothing needs locking. ExtractBarcodesAsString reads every page in one call and returns JSON, and each entry carries the barcode type, position, rotation, and value. IntakeRepository.SaveBarcode stands in for your own data access call that writes the document, type, and value to your table.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*Mr1ik458W3UdcIEo" /><figcaption>Figure 1: Sample JSON output. ​​​​‌ ‍ ​‍​‍‌‍ ‌ ​‍‌‍‍‌‌‍‌ ‌‍‍‌‌‍ ‍​‍​‍​ ‍‍​‍​‍‌ ​ ‌‍​‌‌‍ ‍‌‍‍‌‌ ‌​‌ ‍‌​‍ ‍‌‍‍‌‌‍ ​‍​‍​‍ ​​‍​‍‌‍‍​‌ ​‍‌‍‌‌‌‍‌‍​‍​‍​ ‍‍​‍​‍​‍ ‌ ​ ‌ ‌​‌ ‌‌‌‍‌​‌‍‍‌‌‍ ​‍ ‌‍‍‌‌‍ ‍‌ ‌​‌‍‌‌‌‍ ‍‌ ‌​​‍ ‌‍‌‌‌‍‌​‌‍‍‌‌ ‌​​‍ ‌‍ ‌‌‍ ‌‍‌​‌‍‌‌​ ‌‌ ​​‌ ​‍‌‍‌‌‌ ​ ‌‍‌‌‌‍ ‍‌ ‌​‌‍​‌‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​ ‍ ‌‍‍‌‌‍‌​​ ‌​ ‌ ‌‍‌​‌‍​‌‌‍​ ‌‍‌​​ ‌‍‌‍​‌‌‍‌​​‍ ‌‌‍‌‍‌‍‌​‌‍‌​‌‍​ ​‍ ‌​ ‌​‌‍​ ​ ​​​ ​‍​‍ ‌​ ‍‌‌‍‌‍​ ‍​‌‍​‍​‍ ‌‌‍‌​​ ‍​​ ‌‍‌‍​‌‌‍‌​​ ‌ ​ ​‌​ ‌‌‌‍‌​​ ​​‌‍‌‌‌‍​ ​ ‍ ‌ ‌​‌ ‍‌‌ ​​‌‍‌‌​ ‌‌‍​‍‌‍ ​‌‍ ‌‍‌ ‌‌​​‌‍ ‌ ​ ‌ ‌​​ ‍ ‌ ​​‌‍​‌‌ ‌​‌‍‍​​ ‌‌‍​‍‌‍ ‌‍‌​‌ ‍‌​‍‌‌​ ‌‌‌​​‍‌‌ ‌‍‍ ‌‍‌‌‌ ‍‌​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​‍​ ​‍​ ​‍​ ​‌​ ​ ‌‍‌​​ ​ ​ ‍‌‌‍​‌​ ​ ‌‍​‍​ ​ ​ ‍​‌‍‌​​‍‌‌​ ​‍​ ​‍​‍‌‌​ ‌‌‌​‌​​‍ ‍‌‍​ ‌‍​‌‌ ​​‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​ ‌‍​‍‌‍​‌‌ ​ ‌‍‌‌‌‌‌‌‌ ​‍‌‍ ​​ ‌​‍‌‌​ ​‍‌​‌‍‌ ​ ‌ ‌​‌ ‌‌‌‍‌​‌‍‍‌‌‍ ​‍‌‍‌‍‍‌‌‍‌​​ ‌​ ‌ ‌‍‌​‌‍​‌‌‍​ ‌‍‌​​ ‌‍‌‍​‌‌‍‌​​‍ ‌‌‍‌‍‌‍‌​‌‍‌​‌‍​ ​‍ ‌​ ‌​‌‍​ ​ ​​​ ​‍​‍ ‌​ ‍‌‌‍‌‍​ ‍​‌‍​‍​‍ ‌‌‍‌​​ ‍​​ ‌‍‌‍​‌‌‍‌​​ ‌ ​ ​‌​ ‌‌‌‍‌​​ ​​‌‍‌‌‌‍​ ​‍‌‍‌ ‌​‌ ‍‌‌ ​​‌‍‌‌​ ‌‌‍​‍‌‍ ​‌‍ ‌‍‌ ‌‌​​‌‍ ‌ ​ ‌ ‌​​‍‌‍‌ ​​‌‍​‌‌ ‌​‌‍‍​​ ‌‌‍​‍‌‍ ‌‍‌​‌ ‍‌​‍‌‌​ ‌‌‌​​‍‌‌ ‌‍‍ ‌‍‌‌‌ ‍‌​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​‍​ ​‍​ ​‍​ ​‌​ ​ ‌‍‌​​ ​ ​ ‍‌‌‍​‌​ ​ ‌‍​‍​ ​ ​ ‍​‌‍‌​​‍‌‌​ ​‍​ ​‍​‍‌‌​ ‌‌‌​‌​​‍ ‍‌‍​ ‌‍​‌‌ ​​‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​‍​‍‌ ‌</figcaption></figure><p>The options object narrows the work. SetBarcodeSearchTypes limits the search to basic 1D types such as Code 128, which cuts processing time, so widen it if your documents also carry 2D codes. Add SetPages with a range such as “1–3,5,10-” to skip pages that never carry barcodes. The profile setting trades speed for tolerance: the default profile assumes a moderately high quality source and reads fastest, the low quality source profile is slightly slower, and the small barcodes and natural picture profiles are significantly slower.</p><p>Run the service with dotnet run. The native library must sit in your build output folder, and the search path in the code must point at the Lib folder that holds the module. The <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a> ships in nine languages and shows page ranges and type filters, and the <a href="https://docs.apryse.com/core/samples/multithreadedtest">MultiThreadedTest sample</a> shows the locking model in C#. For the full list of options, see the <a href="https://docs.apryse.com/core/smart-data-extraction/barcode/workflow">Barcode workflow docs</a>.</p><h3>Running a PDF Barcode Extraction Pipeline in Production: Threads, Streams, and Scanned TIFFs</h3><p>A PDF barcode extraction pipeline on the Apryse Server SDK comes down to four decisions: how documents map to threads, where the bytes come from, which format scans arrive in, and what you store when a barcode does not decode.</p><p><strong>Give each document its own thread:</strong> Initialize the SDK once at startup and leave it initialized for the life of the worker. If each thread opens its own PDFDoc, as the loop above does, nothing needs locking. If two threads share one PDFDoc, use the documented Lock call for writes and LockRead for read-only work, because sharing a document without locking gives undefined behavior, including crashes. The <a href="https://docs.apryse.com/core/guides/features/basics/lock">multi-threaded document locking guide</a> covers both patterns. How many threads run, how documents reach them, and what happens on a retry are your code.</p><p><strong>Open PDFs from a stream when they come from a queue or object store:</strong> Construct the PDFDoc from the in-memory bytes instead of a file path, and the document never touches disk. The SDK does not fetch URLs, so your code downloads the bytes first. A path-based PDFDoc and a stream-based PDFDoc feed the same extraction call.</p><p><strong>Convert scanned TIFFs to PDF first:</strong> The Barcode Extraction Module reads PDFs, so a TIFF from your scanner needs converting before extraction. The Server SDK converts TIF and TIFF files to PDF as part of its base <a href="https://docs.apryse.com/core/guides/features/conversion">image-to-PDF conversion</a>. If converted scans still return missed barcodes, see <a href="https://apryse.com/blog/efficient-barcode-extraction">fixing missed barcodes on scanned PDFs</a>.</p><p><strong>Store Base64 payloads from the data field as-is:</strong> When the module cannot decode a barcode, it returns the raw bytes as a Base64-encoded string in the data field instead of the value field. Write that string to its own column without decoding it and decode it later with your own logic. Keep the row, because a barcode that was read but not decoded still tells your matching step that the document carries a code.</p><h3>How to Check Barcode Values Extracted from PDFs Before They Reach Your Database</h3><p>Barcode values extracted from PDFs can arrive as valid JSON and still be wrong for your table, so check each document before the insert. Run these four checks in the service, between parsing and saving.</p><ul><li><strong>Count:</strong> Compare the number of barcodes returned per document with the number you expect. If every bill of lading in your intake carries one barcode per page, count the entries across all pages and compare the total with the page count. Log both numbers on the document row and route any mismatch to review.</li><li><strong>Type:</strong> Check that each barcode type matches the document, for example Code 128 for a bill of lading number. Keep a map from document type to allowed barcode types, and compare the type field of each entry against it. A QR code on a document that should carry only Code 128 points to a misread or a misfiled page.</li><li><strong>Data field:</strong> Flag every entry that returns the data field in place of a value. Test for the data field on each entry before the insert and save those rows with a needs-review status.</li><li><strong>Your own format rules:</strong> Validate each decoded value against the pattern your documents follow. Match the value against a regular expression per document type, such as the bill of lading number format your carriers use, and apply the check digit rule where the barcode type defines one. Reject the row on a mismatch and keep the raw JSON for the review queue.</li></ul><p>Run all four on the worked example. A three-page bill of lading that returns two entries sends one row to review before dispatch ever sees it.</p><h3>Get Started with the Apryse Barcode Extraction Module in C#</h3><p>Download the Apryse Server SDK for .NET 8 or 9 (PDFNetC64.zip) or .NET Framework (PDFNetDotNet4.zip), then download the Barcode Extraction Module and extract it into your SDK directory. Follow the <a href="https://docs.apryse.com/core/guides/get-started/dotnetcore">C# get-started guide</a> for project setup and use the <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a> and the <a href="https://docs.apryse.com/core/samples/multithreadedtest">MultiThreadedTest sample</a> as your references.</p><p>Try it on your own documents: get a trial key at <a href="https://dev.apryse.com/">dev.apryse.com</a>, run the <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a> against one of your own incoming PDFs, and check the type, position and value of each barcode in the JSON. Then join the <a href="https://community.apryse.com/">Apryse Community</a> to share your results, get advice, and connect with developers.</p><p>That is the whole shape of it: one PDF in, JSON out, and rows in your table. The same loop that reads your first test file reads your whole intake folder, so the work you do today is the work you keep. Start with the messiest bill of lading in your folder, because the document that breaks your assumptions teaches you more than the clean ones do.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd7661b5a603" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-to-extract-barcodes-from-incoming-pdfs-into-a-database-with-c-dd7661b5a603">How to Extract Barcodes from Incoming PDFs into a Database With C#</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Run Barcode Extraction From PDFs in Docker]]></title>
            <link>https://medium.com/apryse/how-to-run-barcode-extraction-from-pdfs-in-docker-0f50c0ed579a?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/0f50c0ed579a</guid>
            <category><![CDATA[barcode]]></category>
            <category><![CDATA[docker]]></category>
            <dc:creator><![CDATA[Garry Klooesterman]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 20:21:58 GMT</pubDate>
            <atom:updated>2026-11-18T20:21:58.378Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x6jyQlg8BcvWFHlhEtkheQ.jpeg" /></figure><p>To run barcode extraction from PDFs in Docker, build a Linux x64 image that installs the Apryse Server SDK for Python and the Barcode Extraction Module, then run a small service that initializes the SDK once and extracts the barcodes from each PDF it receives. To scale, run more containers; Kubernetes handles the orchestration.</p><h3>Why Run PDF Barcode Extraction in Your Own Container Instead of a Cloud API</h3><p>It’s 7:45 on a Monday morning at a freight broker, and the intake queue is already deep. Bills of lading keep landing as PDFs, and each one carries a shipment barcode your routing code needs before anything else can happen. Nobody is watching the queue; your service is.</p><p>Teams that need barcodes out of PDFs usually weigh two paths first. A cloud document API sends every PDF to a third-party server, which is a data residency problem for logistics, finance, insurance, and healthcare teams. An open-source decoder keeps the data home, but it takes images, not PDFs, so your team owns the step that turns each page into an image.</p><p>Here’s where the SDK fits. The <a href="https://apryse.com/capabilities/barcode">Barcode Extraction Module</a> is an add-on module for the Apryse Server SDK that works as a barcode API for backend services, detecting, extracting, and decoding barcodes inside PDFs. It runs inside your own container, not as a cloud API, and document content never leaves your environment. By default, the SDK sends limited usage metadata, such as API call names, call counts, and page counts, and Apryse issues a no-tracking license key, or a key configured for fully offline operation, on request. Camera-based scanning on mobile is a different job and belongs to the Scanbot SDK.</p><p>When the queue doubles, add replicas. When it clears, remove them. Kubernetes handles the scheduling, and you control the scale. The Barcode Extraction Module is part of the <a href="https://apryse.com/capabilities/smart-data-extraction">Smart Data Extraction</a> add-on license and installs as a separate module download. Against an open-source decoder, the choice is a license cost versus engineering and maintenance time.</p><p>This post covers Apryse Server SDK integration using Python on Linux x64, in a Docker container and on AWS Lambda. It does not cover other serverless platforms or any language other than Python. For other languages or platforms, see the <a href="https://docs.apryse.com/core/guides/get-started">Server SDK get-started guide</a>.</p><h3>Build a Docker Image for Barcode Extraction from PDFs with Python and the Apryse Barcode Extraction Module</h3><p>Building a Docker image for barcode extraction from PDFs with Python and the Apryse Barcode Extraction Module takes three layers: the Apryse Server SDK for Python, the Barcode Extraction Module, and your own service script. A barcode detection API integration in Docker starts with the Dockerfile below, which stacks those layers on a Debian-based Python image.</p><p>This feature requires the Barcode Extraction Module add-on, which is a separate download from the base Server SDK. See the <a href="https://docs.apryse.com/core/guides/info/modules">modules documentation</a> for download and installation instructions.</p><pre># Build a Linux x64 image that runs the barcode extraction service<br>FROM --platform=linux/amd64 python:3.12-slim-bookworm<br>WORKDIR /opt/barcode-extraction<br><br># Install the Apryse Server SDK for Python from the Apryse package index<br>RUN pip install --no-cache-dir apryse-sdk --extra-index-url=https://pypi.apryse.com<br><br># Extract the Barcode Extraction Module archive into /opt/apryse, which creates /opt/apryse/Lib<br>COPY BarcodeModuleLinux.tar.gz /tmp/<br>RUN mkdir -p /opt/apryse \<br>    &amp;&amp; tar -xzf /tmp/BarcodeModuleLinux.tar.gz -C /opt/apryse \<br>    &amp;&amp; rm /tmp/BarcodeModuleLinux.tar.gz<br><br># Copy the service script that initializes the SDK once per process<br>COPY barcode_extraction_service.py .<br><br># The script reads APRYSE_LICENSE_KEY at runtime. Never bake the key into the image.<br>CMD [&quot;python&quot;, &quot;barcode_extraction_service.py&quot;]</pre><p>The pip line pulls apryse-sdk from pypi.apryse.com rather than public PyPI, because the legacy PDFNetPython3 package on public PyPI is no longer maintained. The Barcode Extraction Module installs by extracting BarcodeModuleLinux.tar.gz, which you download into the build folder before running docker build. The archive extracts into /opt/apryse and creates a Lib/Linux folder holding BarcodeModule.so.</p><p>Extraction alone does not make the module visible to the SDK. Your service script calls PDFNet.AddResourceSearchPath(&quot;/opt/apryse/Lib/&quot;) with the Lib folder, not Lib/Linux, so a single search path covers every installed module.</p><p>The module supports Windows and Linux on x64, which is why the image pins the linux/amd64 platform. Pin the Python version to one your SDK release supports. The extraction code itself lives in a separate walkthrough on how to <a href="https://apryse.com/blog/extract-barcodes-from-pdf-python">extract barcodes from PDFs in Python</a>. The image only has to run it.</p><p><strong>License keys are confidential.</strong> Do not commit them to public repositories. Use environment variables or a secrets manager to inject your key at runtime.</p><h3>Running Barcode Extraction Containers in Production with Kubernetes</h3><p>Barcode extraction containers on Kubernetes scale horizontally because each operation is stateless. The SDK has nothing to coordinate across documents, so you choose the topology that fits your memory and throughput limits. Five rules keep the queue moving.</p><p><strong>Initialize once per process and verify the module at startup:</strong> Call PDFNet.Initialize() at service startup, then PDFNet.AddResourceSearchPath(), and leave the SDK initialized for the life of the worker. Repeated calls have no extra effect, but the first one carries the setup cost, so pay it at startup rather than inside a request handler. Check BarcodeModule.IsModuleAvailable() in the same startup path and fail your Kubernetes readiness check when it returns false. A pod with a wrong search path then never receives a PDF.</p><p><strong>Give each worker its own document:</strong> Each worker opens its own PDFDoc, extracts the barcodes, and releases the document. Sharing one PDFDoc across threads without locking gives undefined behavior, including crashes. If your design shares a document, use the documented Lock and LockRead APIs.</p><p><strong>Inject the license key as a Kubernetes Secret:</strong> Expose the key to the container as the APRYSE_LICENSE_KEY environment variable from a Secret, never from the image or from a manifest in source control. A key baked into an image travels with every copy of that image.</p><p><strong>Match input and options to your documents:</strong> Convert scanned TIFFs to PDF first. The Barcode Extraction Module reads barcodes in PDFs, and Image to PDF in the base SDK handles TIF, TIFF, JPG, PNG, GIF, and BMP files. For very poor scans, the Low Quality Source profile is slightly slower than the default High Quality Source profile, which is the fastest. The Small Barcodes and Natural Picture profiles are significantly slower, so reserve them for unusually small barcodes and images. To read only part of a document, set the Pages option with a range such as “1–3,5,10-” through SetPages. If barcodes still go missing, see the guide to <a href="https://apryse.com/blog/efficient-barcode-extraction">fixing missed barcodes on scanned PDFs</a>.</p><p><strong>Own the queue and the retries:</strong> Queueing, retries, and job orchestration are your code. Your service decides when to requeue a malformed PDF, and your pipeline decides where results go. To <a href="https://apryse.com/blog/barcode-extraction-pipeline-csharp">send barcode values to a database</a>, follow this pipeline walkthrough.</p><p>Back at the freight broker, Monday looks different. The queue still doubles, and the replicas follow. Your platform owns the queue, the scaling, and the retries. The Barcode Extraction Module reads the barcodes inside each PDF, and no PDF leaves your environment.</p><h3>Running PDF Barcode Extraction on AWS Lambda with Python</h3><p>Running PDF barcode extraction on AWS Lambda with Python uses the same Apryse Server SDK and the same Barcode Extraction Module, packaged as a zip file instead of a Docker image.</p><p>Install apryse-sdk into your function folder with the pip --target option, as the Apryse Lambda guide shows, then copy the extracted Barcode Extraction Module Lib folder into the same package and zip the folder. Your handler calls PDFNet.AddResourceSearchPath() with the packaged Lib folder, initializes the SDK with the key from APRYSE_LICENSE_KEY, and checks BarcodeModule.IsModuleAvailable() before it reads a PDF. Repeated calls to PDFNet.Initialize() are safe, so initializing inside the handler works.</p><p>The Barcode Extraction Module supports x64 only, so create the function with the x86_64 architecture. The Apryse Lambda sample writes the incoming file to /tmp before it opens the document, and each invocation processes one document. Triggers, queues, and retries belong to your AWS configuration, not to the SDK.</p><p>For the full setup, follow the AWS Lambda tab of the <a href="https://docs.apryse.com/core/get-started/languages/python3">Python get-started guide</a>.</p><h3>Get Started with the Apryse Barcode Extraction Module in Docker</h3><p>To get started with the Apryse Barcode Extraction Module in Docker, install the SDK with pip install apryse-sdk --extra-index-url=https://pypi.apryse.com and download the module from the <a href="https://docs.apryse.com/core/guides/info/modules#barcode-module">modules documentation</a>. Follow the <a href="https://docs.apryse.com/core/guides/get-started/python3">Python get-started guide</a> for environment setup and the <a href="https://docs.apryse.com/core/smart-data-extraction/barcode/workflow">Barcode workflow docs</a> for the extraction steps.</p><p>Try it on your own documents: get a trial key at <a href="https://dev.apryse.com/">dev.apryse.com</a>, run the <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a> against one of your own PDFs inside the container, and check the decoded content of each barcode in the JSON output. The <a href="https://community.apryse.com/">Apryse Community</a> is a great place to get answers, share what you’re learning, and meet developers working on similar problems.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0f50c0ed579a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-to-run-barcode-extraction-from-pdfs-in-docker-0f50c0ed579a">How to Run Barcode Extraction From PDFs in Docker</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Extract Barcodes from PDFs in C#]]></title>
            <link>https://medium.com/apryse/how-to-extract-barcodes-from-pdfs-in-c-a8b470ec35f3?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/a8b470ec35f3</guid>
            <category><![CDATA[csharp]]></category>
            <category><![CDATA[barcode]]></category>
            <dc:creator><![CDATA[Garry Klooesterman]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 20:12:12 GMT</pubDate>
            <atom:updated>2026-11-18T20:12:12.702Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xMf-4eHdB1VeCGK48a_PFQ.jpeg" /></figure><p>To extract barcodes from PDFs in C#, add the Apryse Server SDK for .NET to your project, download the Barcode Extraction Module into the SDK’s Lib folder, and call BarcodeModule.ExtractBarcodesAsString on your PDF. You get every barcode as JSON with its type, position, rotation and value, from born-digital or scanned PDFs, on Windows or Linux servers.</p><h3>Why Read Barcodes from PDFs in C# with an SDK That Opens the PDF Directly</h3><p>Picture a freight-forwarding service that receives bills of lading as PDF packets. Page one has a clean Code 128 label, and page nine is a driver’s scan with a coffee ring across the pallet barcode. An open-source decoder takes images, not PDFs, so your code renders each page to a bitmap first and picks a resolution for it. Pick low and small barcodes lose detail. Pick high and memory climbs on every page of every packet.</p><p>One approach is a C# barcode extraction SDK that opens the PDF itself. The <a href="https://apryse.com/capabilities/barcode">Barcode Extraction Module</a> detects and decodes barcodes in PDFs, covers 100+ barcode types, and returns the results as JSON. It runs on Windows and Linux x64 inside your own infrastructure, so no document data leaves your environment. It is not a camera scanner: the Scanbot SDK captures barcodes through a device camera, while the Barcode Extraction Module reads barcodes already inside PDFs.</p><p>The Barcode Extraction Module falls under the <a href="https://apryse.com/capabilities/smart-data-extraction">Smart Data Extraction</a> add-on license and installs as a separate module download. Compared with an open-source decoder, the choice is a license cost against engineering and maintenance time.</p><p>This post covers Apryse Server SDK integration using C# on Windows and Linux, and shows how to extract barcodes from PDFs to JSON. It does not cover <a href="https://apryse.com/blog/extract-qr-codes-from-pdf-files">QR codes from PDFs in C#</a>, how to <a href="https://apryse.com/blog/barcode-extraction-pipeline-csharp">send barcode values to a database</a>, or camera scanning, which belongs to the Scanbot SDK. For other languages or platforms, see the <a href="https://docs.apryse.com/core/guides/get-started">Server SDK get-started guide</a>.</p><h3>Extract Barcodes from a PDF to JSON in C# with the Apryse Barcode Extraction Module</h3><p>Extracting barcodes from a PDF to JSON in C# with the Apryse Barcode Extraction Module takes one call after setup. Start with the packet from the story above: a bill of lading with one born-digital page and one scanned page.</p><p><strong><em>This feature requires the Barcode Extraction Module add-on, which is a separate download from the base Server SDK. See the </em></strong><a href="https://docs.apryse.com/core/guides/info/modules"><strong><em>modules documentation</em></strong></a><strong><em> for download and installation instructions.</em></strong></p><p><strong><em>License keys are confidential. Do not commit them to public repositories. Use environment variables or a secrets manager to inject your key at runtime.</em></strong></p><p>The code initializes PDFNet, points the SDK at the Lib folder, checks that the module loaded, extracts the barcodes, and terminates.</p><pre>// Extract every barcode from a bill of lading packet and print the JSON<br>using System;<br>using pdftron;<br>using pdftron.PDF;<br>class BillOfLadingBarcodeService<br>{<br>    static void Main()<br>    {<br>        PDFNet.Initialize(&quot;YOUR_APRYSE_LICENSE_KEY&quot;); // Get your key at https://dev.apryse.com<br>        try<br>        {<br>            PDFNet.AddResourceSearchPath(&quot;../../../Lib/&quot;); // Parent Lib folder, not Lib/Windows/Barcode<br>            if (!BarcodeModule.IsModuleAvailable())<br>                throw new InvalidOperationException(&quot;Barcode Extraction Module not found. Check the Lib path.&quot;);<br>            string barcodeJson = BarcodeModule.ExtractBarcodesAsString(&quot;bill-of-lading-packet.pdf&quot;, &quot;bill-of-lading-barcodes.json&quot;);<br>            Console.WriteLine(barcodeJson);<br>        }<br>        finally { PDFNet.Terminate(); }<br>    }<br>}</pre><p>Run it from the project folder on Windows or Linux:</p><pre># Run from the project folder. The Lib location is set in code through AddResourceSearchPath<br>dotnet run​​​​</pre><p>PDFNet.Initialize registers the license key. AddResourceSearchPath tells the SDK where the Lib folder sits, relative to your executable, and one path covers every installed module. IsModuleAvailable returns false when the module is missing or the path is wrong, so the code stops with a clear message instead of returning nothing. ExtractBarcodesAsString reads the PDF directly and returns the detected barcodes as a JSON string, which the code prints. The input and output file names follow the form on the Barcode workflow page. One call covers the whole packet.</p><p>.NET Framework projects reference PDFNetLoader.dll and PDFNet.dll, and keep the PDFNetLoader.Instance() line shown in the <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a>.</p><p>Here is the JSON from the BarcodeTest sample, run against its barcodes.pdf:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*_O6UxlXkjP47bQ2G" /><figcaption>Figure 1: Sample JSON output. ​​​​‌ ‍ ​‍​‍‌‍ ‌ ​‍‌‍‍‌‌‍‌ ‌‍‍‌‌‍ ‍​‍​‍​ ‍‍​‍​‍‌ ​ ‌‍​‌‌‍ ‍‌‍‍‌‌ ‌​‌ ‍‌​‍ ‍‌‍‍‌‌‍ ​‍​‍​‍ ​​‍​‍‌‍‍​‌ ​‍‌‍‌‌‌‍‌‍​‍​‍​ ‍‍​‍​‍​‍ ‌ ​ ‌ ‌​‌ ‌‌‌‍‌​‌‍‍‌‌‍ ​‍ ‌‍‍‌‌‍ ‍‌ ‌​‌‍‌‌‌‍ ‍‌ ‌​​‍ ‌‍‌‌‌‍‌​‌‍‍‌‌ ‌​​‍ ‌‍ ‌‌‍ ‌‍‌​‌‍‌‌​ ‌‌ ​​‌ ​‍‌‍‌‌‌ ​ ‌‍‌‌‌‍ ‍‌ ‌​‌‍​‌‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​ ‍ ‌‍‍‌‌‍‌​​ ‌‌‍​ ​ ‌‌​ ‌ ‌‍‌‍​ ‌​​ ​ ‌‍‌‌​ ‌‌​‍ ‌​ ‌​​ ​‌‌‍‌‍​ ‍​​‍ ‌​ ‌​‌‍​ ‌‍​ ​ ‌ ​‍ ‌​ ‍​​ ‍​​ ‍‌​ ‍‌​‍ ‌‌‍​‍‌‍‌‌‌‍‌‍​ ‍​‌‍​ ‌‍‌‍‌‍​‌‌‍‌‍‌‍​ ‌‍‌​​ ​‍​ ‌ ​ ‍ ‌ ‌​‌ ‍‌‌ ​​‌‍‌‌​ ‌‌‍​‍‌‍ ​‌‍ ‌‍‌ ‌‌​​‌‍ ‌ ​ ‌ ‌​​ ‍ ‌ ​​‌‍​‌‌ ‌​‌‍‍​​ ‌‌‍​‍‌‍ ‌‍‌​‌ ‍‌​‍‌‌​ ‌‌‌​​‍‌‌ ‌‍‍ ‌‍‌‌‌ ‍‌​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​‍​ ​‍​ ‌ ​ ‌‌‌‍‌‌‌‍‌‌‌‍​ ​ ​‌‌‍​‌‌‍​‍​ ​‍‌‍‌​​ ​ ‌‍​‍​‍‌‌​ ​‍​ ​‍​‍‌‌​ ‌‌‌​‌​​‍ ‍‌‍​ ‌‍​‌‌ ​​‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​ ‌‍​‍‌‍​‌‌ ​ ‌‍‌‌‌‌‌‌‌ ​‍‌‍ ​​ ‌​‍‌‌​ ​‍‌​‌‍‌ ​ ‌ ‌​‌ ‌‌‌‍‌​‌‍‍‌‌‍ ​‍‌‍‌‍‍‌‌‍‌​​ ‌‌‍​ ​ ‌‌​ ‌ ‌‍‌‍​ ‌​​ ​ ‌‍‌‌​ ‌‌​‍ ‌​ ‌​​ ​‌‌‍‌‍​ ‍​​‍ ‌​ ‌​‌‍​ ‌‍​ ​ ‌ ​‍ ‌​ ‍​​ ‍​​ ‍‌​ ‍‌​‍ ‌‌‍​‍‌‍‌‌‌‍‌‍​ ‍​‌‍​ ‌‍‌‍‌‍​‌‌‍‌‍‌‍​ ‌‍‌​​ ​‍​ ‌ ​‍‌‍‌ ‌​‌ ‍‌‌ ​​‌‍‌‌​ ‌‌‍​‍‌‍ ​‌‍ ‌‍‌ ‌‌​​‌‍ ‌ ​ ‌ ‌​​‍‌‍‌ ​​‌‍​‌‌ ‌​‌‍‍​​ ‌‌‍​‍‌‍ ‌‍‌​‌ ‍‌​‍‌‌​ ‌‌‌​​‍‌‌ ‌‍‍ ‌‍‌‌‌ ‍‌​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​‍​ ​‍​ ‌ ​ ‌‌‌‍‌‌‌‍‌‌‌‍​ ​ ​‌‌‍​‌‌‍​‍​ ​‍‌‍‌​​ ​ ‌‍​‍​‍‌‌​ ​‍​ ​‍​‍‌‌​ ‌‌‌​‌​​‍ ‍‌‍​ ‌‍​‌‌ ​​‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​‍​‍‌ ‌</figcaption></figure><p>Every option used later in this post is documented in the <a href="https://docs.apryse.com/core/smart-data-extraction/barcode/workflow">Barcode workflow docs</a>.</p><h3>Running C# Barcode Extraction on Windows and Linux Servers in Production</h3><p>Running C# barcode extraction on Windows and Linux servers in production comes down to five habits.</p><p><strong>Initialize once per process:</strong> The sample above initializes and terminates around a single run. In a long-running .NET service, call PDFNet.Initialize at startup, leave it initialized for the life of the worker, and call PDFNet.Terminate at shutdown. Repeated calls to Initialize have no extra effect, but the first call carries the setup cost.</p><p><strong>Process one document per thread:</strong> Give each worker thread its own PDF and its own extraction call. Sharing one PDFDoc across threads without locking gives undefined behavior, including crashes. The queue that feeds those workers is yours to build, because the Server SDK does not include batch orchestration.</p><p><strong>Convert scanned TIFFs to PDF first:</strong> The Barcode Extraction Module reads PDFs, so a TIFF from a scanner goes through the Server SDK’s image-to-PDF conversion before extraction. That conversion is part of the base Server SDK license and needs no extra module. The <a href="https://docs.apryse.com/core/guides/features/conversion">conversion guide</a> covers the call.</p><p><strong>Read payloads from the right JSON field:</strong> When the module decodes a barcode, the value lands in the “text” field. When it cannot decode one, the raw bytes land in the “data” field as a Base64 string. If your downstream code needs one consistent field, call SetDataOutputFormat with BarcodeOptions.OutputFormat.e_binary and decode the Base64 yourself.</p><p><strong>Narrow the search, then tune the profile:</strong> By default the module searches every supported barcode type in every orientation, which is the slowest setting. Limit it to the types you expect with SetBarcodeSearchTypes, and limit pages with SetPages, which takes ranges such as “1–3,5,10-”. Switch profiles only for documents that miss barcodes: Low Quality Source is slightly slower than the default High Quality Source, while Small Barcodes and Natural Picture are significantly slower. For missed reads on scans, see the guide to <a href="https://apryse.com/blog/efficient-barcode-extraction">fixing missed barcodes on scanned PDFs</a>.</p><h3>How to Check Barcode Values Extracted from PDFs in C#</h3><p>Barcode values extracted from PDFs in C# arrive as valid JSON even when a barcode is missing or misread, so check four things before you trust a run. Build your fixture from the bill of lading packet: the clean page-one label and the coffee-ring scan.</p><ul><li><strong>Count:</strong> Compare the number of barcode entries in the JSON with the number you count on the page. Keep a fixture PDF with a known count, run it one page at a time with SetPages(“1”), and assert the count in a unit test.</li><li><strong>Type:</strong> Confirm each entry’s type matches the symbology you expect. Put one barcode of each type your workflow handles in the fixture, then assert the reported type for each.</li><li><strong>Decoded value:</strong> Compare each value in the text field with the human-readable line printed under the barcode, where one exists. A transposed digit on a 1D label shows up immediately.</li><li><strong>Undecoded fallbacks:</strong> Count the entries that land in data instead of text. A rise in that count on a batch of scans points to poor source quality, so rerun those pages with the Low Quality Source profile and compare the counts.</li></ul><h3>Get Started with the Apryse Barcode Extraction Module in C#</h3><p>Download the Apryse Server SDK for <a href="https://docs.apryse.com/core/guides/download">.NET 8 or 9 (PDFNetC64.zip) or .NET Framework (PDFNetDotNet4.zip)</a>, then reference it from your project as described in the <a href="https://docs.apryse.com/core/get-started/frameworks/dotnetcore">C# get-started guide</a>. Download the Barcode Extraction Module for <a href="https://docs.apryse.com/downloads/BarcodeModuleWindows.zip">Windows (BarcodeModuleWindows.zip)</a> or <a href="https://docs.apryse.com/downloads/BarcodeModuleLinux.tar.gz">Linux (BarcodeModuleLinux.tar.gz)</a> and extract it into the PDFNetC directory of the SDK, which places the module files under Lib. The <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a> includes runnable C# for page ranges and type filters.</p><p>Try it on your own documents: get a trial key at <a href="https://dev.apryse.com/">dev.apryse.com</a>, run the BarcodeTest sample against one of your own PDFs, and check the type, position, and value of each barcode in the JSON. Then bring your questions and discoveries to the <a href="https://community.apryse.com/">Apryse Community</a>, where other developers are working through similar challenges.</p><p>Back in the freight office, the packet that used to wait for someone to squint at page nine now goes through one call, and your service hands the values to your tracking system. You still own the queue, the routing, and what happens when a barcode comes back undecoded. The Barcode Extraction Module reads the PDF. Everything after that is your app.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a8b470ec35f3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-to-extract-barcodes-from-pdfs-in-c-a8b470ec35f3">How to Extract Barcodes from PDFs in C#</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Extract Barcodes From PDFs in Python​​​​‌‍‌‍‌‌‌‍]]></title>
            <link>https://medium.com/apryse/how-to-extract-barcodes-from-pdfs-in-python-d7ae64b146e5?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/d7ae64b146e5</guid>
            <category><![CDATA[barcode]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Garry Klooesterman]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 20:09:02 GMT</pubDate>
            <atom:updated>2026-11-18T20:09:02.099Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gotj_JhsC7-11vhG6ZrZyg.jpeg" /></figure><p>To extract barcodes from PDFs in Python on a server, install the Apryse SDK from Apryse’s Python package index, add the Barcode Extraction Module to the SDK’s Lib folder, and call BarcodeModule.ExtractBarcodes with your PDF and an output file name. The module reads 100+ barcode types from born-digital and scanned PDFs on Windows or Linux and writes each barcode’s type, position, rotation and value to a JSON file.</p><h3>Why Read Barcodes from PDFs with an SDK Instead of Rendering Pages for a Python Decoder</h3><p>It is 6 a.m. at a freight broker, and the overnight batch has landed: a stack of scanned bills of lading, each with a tracking barcode that decides where a shipment goes next. A Python worker has to read every one of those barcodes before anyone downstream opens a file. The barcode is small. Getting it out of the PDF and into your system is the whole job, and the same job shows up in invoice routing, claims intake, and patient record indexing.</p><p>Most teams start by rendering each PDF page to an image and passing the image to a decoder. Open-source decoders take images, not PDFs, so you add a page renderer, choose a resolution, and maintain the glue between the two. The <a href="https://apryse.com/capabilities/barcode">Barcode Extraction Module</a> reads the PDF directly, so there is no page-rendering step in your code. It reads barcodes inside PDF files, which is a different job from camera scanning, where the Scanbot SDK is the right tool.</p><p>Compared with an open-source decoder, the choice is a license cost against engineering and maintenance time. The Barcode Extraction Module is an add-on module that ships as a separate download, and it belongs to the <a href="https://apryse.com/capabilities/smart-data-extraction">Smart Data Extraction</a> add-on license. Everything runs on your own servers, so no document data leaves your environment.</p><p>This post covers Apryse Server SDK integration using Python on Windows and Linux (x64). It shows how to extract barcodes from PDFs, scanned pages included, and save the results as JSON. Camera scanning belongs to the Scanbot SDK, and other languages are covered in the <a href="https://docs.apryse.com/core/guides/get-started">Server SDK get-started guide</a>.</p><h3>Extract Barcodes from a PDF to JSON in Python with the Apryse Barcode Extraction Module</h3><p>Extracting barcodes from a PDF to JSON in Python takes one call to BarcodeModule.ExtractBarcodes. That class is the Python barcode extraction API in this post: a local call into the SDK on your own server, not a request to a hosted service. Install the SDK with pip install apryse-sdk --extra-index-url=https://pypi.apryse.com, then extract the Barcode Extraction Module archive into the folder you run the script from, so a Lib folder sits next to your script.</p><p>This feature requires the Barcode Extraction Module, an add-on module that is a separate download from the base Server SDK. See the <a href="https://docs.apryse.com/core/guides/info/modules">modules documentation</a> for download and installation instructions.</p><p>License keys are confidential. Do not commit them to public repositories. Use environment variables or a secrets manager to inject your key at runtime.</p><pre># Read every barcode in a shipping manifest PDF and save the results as JSON<br>import os<br>from apryse_sdk import *<br><br>def extract_manifest_barcodes(manifest_path, results_path):<br>    PDFNet.Initialize(os.environ[&quot;APRYSE_LICENSE_KEY&quot;]) # Key from dev.apryse.com<br>    # Point the SDK at the parent Lib folder that holds the Barcode Extraction Module<br>    PDFNet.AddResourceSearchPath(&quot;./Lib&quot;)<br>    try:<br>        if not BarcodeModule.IsModuleAvailable():<br>            raise RuntimeError(&quot;Barcode Extraction Module not found on search path&quot;)<br>        manifest_document = PDFDoc(manifest_path)<br>        BarcodeModule.ExtractBarcodes(manifest_document, results_path)<br>        manifest_document.Close()<br>    finally:<br>        PDFNet.Terminate()<br><br>if __name__ == &quot;__main__&quot;:<br>    extract_manifest_barcodes(&quot;manifest.pdf&quot;, &quot;manifest_barcodes.json&quot;)</pre><p>Save the script as extract_manifest_barcodes.py, set the APRYSE_LICENSE_KEY environment variable, and run it with python3 -u extract_manifest_barcodes.py on Linux or python.exe -u extract_manifest_barcodes.py on Windows. The SDK installs as a precompiled Python package, so the run command needs no library-path flag. The AddResourceSearchPath call in the script is what tells the SDK where the module lives.</p><p>PDFNet.Initialize registers your license key, and PDFNet.Terminate in the finally block releases the SDK even when extraction fails. AddResourceSearchPath takes the parent Lib folder, not the Windows or Linux Barcode folder inside it. IsModuleAvailable returns false when that path is wrong, so the script fails with a clear message. ExtractBarcodes scans every page and converts each barcode it finds to JSON in the output file. Scanned pages go through the same call, because the default profile assumes vector graphics or an average to high quality scan.</p><p>Each entry in the output file reports the barcode type, position, rotation, and value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*H8NO_QsZ6FTo6hQ3" /><figcaption>Figure 1: Sample JSON output. ​​​​‌ ‍ ​‍​‍‌‍ ‌ ​‍‌‍‍‌‌‍‌ ‌‍‍‌‌‍ ‍​‍​‍​ ‍‍​‍​‍‌ ​ ‌‍​‌‌‍ ‍‌‍‍‌‌ ‌​‌ ‍‌​‍ ‍‌‍‍‌‌‍ ​‍​‍​‍ ​​‍​‍‌‍‍​‌ ​‍‌‍‌‌‌‍‌‍​‍​‍​ ‍‍​‍​‍​‍ ‌ ​ ‌ ‌​‌ ‌‌‌‍‌​‌‍‍‌‌‍ ​‍ ‌‍‍‌‌‍ ‍‌ ‌​‌‍‌‌‌‍ ‍‌ ‌​​‍ ‌‍‌‌‌‍‌​‌‍‍‌‌ ‌​​‍ ‌‍ ‌‌‍ ‌‍‌​‌‍‌‌​ ‌‌ ​​‌ ​‍‌‍‌‌‌ ​ ‌‍‌‌‌‍ ‍‌ ‌​‌‍​‌‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​ ‍ ‌‍‍‌‌‍‌​​ ‌​ ‍‌‌‍‌​​ ​‍​ ‍‌‌‍​‌‌‍‌‍‌‍‌‌‌‍​‍​‍ ‌‌‍‌‍​ ​​‌‍​‍​ ‌‍​‍ ‌​ ‌​​ ‌​​ ‍​​ ​‍​‍ ‌‌‍​‌​ ‌‍​ ​‍​ ‌‍​‍ ‌‌‍‌‍​ ‍‌‌‍‌‍​ ‌‍​ ​‌​ ‌‍‌‍‌‌‌‍​ ‌‍‌‍​ ‌​​ ‍​​ ‍‌​ ‍ ‌ ‌​‌ ‍‌‌ ​​‌‍‌‌​ ‌‌‍​‍‌‍ ​‌‍ ‌‍‌ ‌‌​​‌‍ ‌ ​ ‌ ‌​​ ‍ ‌ ​​‌‍​‌‌ ‌​‌‍‍​​ ‌‌‍​‍‌‍ ‌‍‌​‌ ‍‌​‍‌‌​ ‌‌‌​​‍‌‌ ‌‍‍ ‌‍‌‌‌ ‍‌​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​‍​ ​‍​ ‍​​ ​ ​ ​ ​ ​ ‌‍​‍‌‍‌​‌‍​ ​ ‌‍​ ​ ​ ​​​ ‌‌‌‍‌‍​‍‌‌​ ​‍​ ​‍​‍‌‌​ ‌‌‌​‌​​‍ ‍‌‍​ ‌‍​‌‌ ​​‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​ ‌‍​‍‌‍​‌‌ ​ ‌‍‌‌‌‌‌‌‌ ​‍‌‍ ​​ ‌​‍‌‌​ ​‍‌​‌‍‌ ​ ‌ ‌​‌ ‌‌‌‍‌​‌‍‍‌‌‍ ​‍‌‍‌‍‍‌‌‍‌​​ ‌​ ‍‌‌‍‌​​ ​‍​ ‍‌‌‍​‌‌‍‌‍‌‍‌‌‌‍​‍​‍ ‌‌‍‌‍​ ​​‌‍​‍​ ‌‍​‍ ‌​ ‌​​ ‌​​ ‍​​ ​‍​‍ ‌‌‍​‌​ ‌‍​ ​‍​ ‌‍​‍ ‌‌‍‌‍​ ‍‌‌‍‌‍​ ‌‍​ ​‌​ ‌‍‌‍‌‌‌‍​ ‌‍‌‍​ ‌​​ ‍​​ ‍‌​‍‌‍‌ ‌​‌ ‍‌‌ ​​‌‍‌‌​ ‌‌‍​‍‌‍ ​‌‍ ‌‍‌ ‌‌​​‌‍ ‌ ​ ‌ ‌​​‍‌‍‌ ​​‌‍​‌‌ ‌​‌‍‍​​ ‌‌‍​‍‌‍ ‌‍‌​‌ ‍‌​‍‌‌​ ‌‌‌​​‍‌‌ ‌‍‍ ‌‍‌‌‌ ‍‌​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​ ‌​‌​​‍‌‌​ ​‍​ ​‍​ ‍​​ ​ ​ ​ ​ ​ ‌‍​‍‌‍‌​‌‍​ ​ ‌‍​ ​ ​ ​​​ ‌‌‌‍‌‍​‍‌‌​ ​‍​ ​‍​‍‌‌​ ‌‌‌​‌​​‍ ‍‌‍​ ‌‍​‌‌ ​​‌ ‌​‌‍‍‌‌‍ ‌‍ ‍​‍​‍‌ ‌</figcaption></figure><p>For the full list of extraction options, see the <a href="https://docs.apryse.com/core/smart-data-extraction/barcode/workflow">Barcode workflow docs</a>.</p><h3>Running Python Barcode Extraction on a Server in Production</h3><p>Running Python barcode extraction on a server in production with the Apryse Barcode Extraction Module adds five constraints that a single-file test never shows.</p><p><strong>Initialize once per process:</strong> PDFNet.Initialize is a once-per-process call. In a long-running worker, call it at startup, keep per-document work between startup and shutdown, and call PDFNet.Terminate when the worker exits. The example above wraps both calls around one document so it runs as a standalone script.</p><p><strong>Convert scanned TIFFs to PDF first:</strong> Pass the module a PDF. If your scanner writes TIFF files, convert each one with the base Server SDK’s <a href="https://docs.apryse.com/core/guides/features/conversion">Image to PDF conversion</a>, which covers TIF and TIFF input, then run extraction on the converted file.</p><p><strong>Handle payloads that land in the data field:</strong> When the module decodes a barcode, it reports the decoded value. When it cannot decode one, it stores the payload as a Base64-encoded string in the data field. Decode those entries in your own code or set the binary output format on BarcodeOptions to receive every payload as Base64 in data.</p><p><strong>Narrow the search to cut processing time:</strong> By default, the module searches for every supported barcode type in every orientation, which is its slowest setting. If your documents carry only shipping labels, restrict the search to the types and orientations you expect, and limit pages with a range such as “1–3,5,10-”. Profiles trade speed for source quality: the default profile is fastest, Low Quality Source is slightly slower, and Small Barcodes and Natural Picture are significantly slower. Read about <a href="https://apryse.com/blog/efficient-barcode-extraction">fixing missed barcodes on scanned PDFs</a> before you change profiles and pass the BarcodeOptions object as the last argument to ExtractBarcodes.</p><p><strong>Deploy on Windows or Linux x64:</strong> Server-side barcode detection with this module runs on Windows and Linux x64 only, so build container images for Linux x64, not ARM or 32-bit. For image setup, see how to <a href="https://apryse.com/blog/barcode-extraction-docker">run it in a container</a>.</p><h3>How to Check Barcode Values Extracted from PDFs in Python</h3><p>Checking barcode values extracted from PDFs in Python catches output that is valid JSON and still wrong.</p><p>Back at the freight broker’s dock, the check that matters is the count: a manifest with four labels yields four entries, or the batch goes to a person.</p><ul><li><strong>Count:</strong> Compare the number of barcodes in the JSON with the number you expect. Keep a test PDF with a known barcode count, load the output with Python’s JSON module, and assert that the entry count matches. A result file with fewer entries than you expect is still valid JSON, so the check has to live in code.</li><li><strong>Type:</strong> Check each entry’s type against what the document should contain. Build a test set with one PDF per barcode type you accept, assert that every reported type is on your allowed list, and log any other type for review.</li><li><strong>Data field:</strong> Flag every entry whose payload sits in the data field. Scan the entries for that field, route them to a review queue, and decode the Base64 string in your code to confirm the content.</li><li><strong>Scanned copy:</strong> Compare a scanned copy of a test document with its born-digital original. Run both through the same script and diff the two counts, which shows whether your scans need a different profile.</li></ul><h3>Get Started with the Apryse Barcode Extraction Module in Python</h3><p>Get started with the Apryse Barcode Extraction Module in Python by installing the SDK with pip install apryse-sdk --extra-index-url=https://pypi.apryse.com and downloading the Barcode Extraction Module from the <a href="https://docs.apryse.com/core/guides/info/modules#barcode-module">modules documentation</a>. Virtual environment and Ubuntu setup steps are in the <a href="https://apryse.com/blog/apryse-sdk-python-linux-pdf-barcode-guide">Python-on-Linux setup guide</a> and the <a href="https://apryse.com/blog/apryse-sdk-python-linux-pdf-barcode-guide">Python get-started page</a>. The full working example is the <a href="https://docs.apryse.com/core/get-started/samples/barcodetest">BarcodeTest sample</a>.</p><p>Try it on your own documents: get a trial key at <a href="https://dev.apryse.com/">dev.apryse.com</a>, run the BarcodeTest sample against one of your own PDFs, and check the type, position and value of each barcode in the JSON. If you have any questions, the <a href="https://community.apryse.com/">Apryse Community</a> is a great place to get answers.</p><p>That is the full path: one install, one module download, and one call that turns a PDF into JSON. What your app does with each barcode is up to you, whether that is matching a tracking number, filing an invoice, or flagging a claim for review. Start with one of your own documents and see what comes back.</p><p><a href="https://dev.apryse.com/">Start a Free Trial</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7ae64b146e5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-to-extract-barcodes-from-pdfs-in-python-d7ae64b146e5">How to Extract Barcodes From PDFs in Python​​​​‌‍‌‍‌‌‌‍</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Convert a PDF to Structured JSON in C#]]></title>
            <link>https://medium.com/apryse/how-to-convert-a-pdf-to-structured-json-in-c-bd0218a1ccd1?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/bd0218a1ccd1</guid>
            <category><![CDATA[data-extraction]]></category>
            <dc:creator><![CDATA[Garry Klooesterman]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 19:25:28 GMT</pubDate>
            <atom:updated>2026-11-16T19:25:30.091Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8n-V3WcfRivY1dYk5fyLnA.png" /></figure><p>The pipeline reported zero errors, and the number was still wrong.</p><p>Dana found it on a Monday morning, ten minutes into her review queue and not yet through her first coffee. The statement on her left monitor said Q3 fuel cost 412,300. The analytics store on her right said 398,150. She checked the PDF twice, because the job had run overnight, because nothing had thrown an error, and because nobody enjoys telling engineering the data is bad. Then she saw it. 398,150 was the Q2 number, sitting in the Q3 slot like it had always lived there.</p><p>Run a two-column quarterly statement through a plain text extractor and the footer lands in the middle of a paragraph. The table with the merged Operating Expenses header collapses into a single line of numbers. Nothing produces an error, and since the output looks like text, the bug ships.</p><p>The cause is how PDFs store content. A PDF stores glyphs at coordinates. It keeps no record of which glyphs form a heading, a footer, or a table cell, so a text extractor returns characters in the order the file draws them, not the order anyone reads them. Template parsers patch over that until a vendor moves a column. A model fed the flat text inherits the scrambled order and reasons over it anyway.</p><p>Dana’s team runs a finance application that loads vendor statements into an analytics store and a search index. Your app needs the structure recovered from the page itself, with a page number attached so a reviewer can find the source. This post builds that step in C# on .NET, on Windows x64 and Linux x64, using the Apryse Server SDK. Smart Data Extraction, the add-on that does the work, does not run on macOS, ARM, or 32-bit platforms.</p><p>This post covers document structure, tables, and Markdown. For plain text extraction with the base SDK, see <a href="https://apryse.com/capabilities/extraction/text-extraction">Text Extraction</a>.</p><p><strong>Recovering Structure from Layout, Not Text Order</strong></p><p>Smart Data Extraction recovers structure from layout. Its Document Structure engine is deterministic and layout-aware. It reads position, spacing, and indentation, then separates paragraphs, lists, headers, footers, and images into distinct blocks. It also tells section columns apart from table columns, so a two-column narrative stops interleaving. The same file produces the same structure on every run.</p><p>The JSON carries page numbers and bounding boxes. That pair is what lets your app draw a highlight on the source page when a reviewer asks where a number came from.</p><p><strong>Keeping Tables Intact: Merged Cells, Multi-Row Headers, and Page Breaks</strong></p><p>Tables get their own engine because they fail differently. The Tabular engine rebuilds rows and columns, including merged cells and multi-row headers, and follows a table across pages. Your Operating Expenses header stays one cell that spans its columns, not a stray label above a run of numbers.</p><p>The tradeoff is validation. Table results carry no confidence score, so your code checks them, for example by summing a column against the printed total.</p><p>The Tabular engine converts all content in the PDF, so paragraphs and other non-table content appear in its output next to the tables. Remove what you do not need after extraction.</p><p><strong>Choosing the Output: JSON, Excel, or Markdown</strong></p><p>Smart Data Extraction returns the same document in several shapes. Pick by what reads the output.</p><ul><li><strong>Structure JSON</strong> fits code that needs headings, paragraphs, lists, headers, and footers with page numbers and bounding boxes. You parse the schema yourself.</li><li><strong>Tables JSON</strong> fits analytics code that needs rows and cells. Table results carry no confidence score, and the output includes non-table content that you filter out after extraction.</li><li><strong>Excel</strong> comes from ExtractToXLSX, which writes the tabular result to a spreadsheet for analysts. Like the tables JSON, it includes non-table content, so remove what you do not need.</li><li><strong>Markdown </strong>output was added in SDK 12.2 and fits search indexes and LLM pipelines. It preserves reading order, heading levels 1 to 6, nested lists, and tables with merged cells. The output is data for a pipeline, not a visual copy of the page.</li><li><strong>Word, Excel, PowerPoint, or HTML</strong> come from the separate Structured Output Module, which reproduces the whole document for people to view or edit. It is a different add-on from Smart Data Extraction and does not output Markdown.</li></ul><p><strong>Extracting Structure, Tables, and Markdown in C#</strong></p><p>The example below runs three extractions against one statement: document structure as JSON, tables as JSON, and Markdown. Put it in the Program.cs of a .NET 8 console project, and place quarterly-statement.pdf in the working directory or replace the path.</p><p>This feature requires the Smart Data Extraction add-on module, which is a separate download from the base Server SDK. See the <a href="https://docs.apryse.com/core/guides/info/modules">modules documentation</a> for download and installation instructions.</p><p>License keys are confidential. Do not commit them to public repositories. Use environment variables or a secrets manager to inject your key at runtime.</p><pre>// Extract document structure, tables, and Markdown from a quarterly statement PDF <br>using System; <br>using pdftron; <br>using pdftron.Common; <br>using pdftron.PDF; <br>  <br>string licenseKey = Environment.GetEnvironmentVariable(&quot;APRYSE_LICENSE_KEY&quot;);  <br>if (string.IsNullOrEmpty(licenseKey))  <br>{  <br>    Console.Error.WriteLine(&quot;Set the APRYSE_LICENSE_KEY environment variable to your trial key.&quot;);  <br>    Environment.ExitCode = 1; return;  <br>}  <br>PDFNet.Initialize(licenseKey); <br>try <br>{ <br>    // The Lib folder next to your built application, where you expanded the module <br>    PDFNet.AddResourceSearchPath(&quot;./Lib/&quot;); <br>    if (!DataExtractionModule.IsModuleAvailable(DataExtractionModule.DataExtractionEngine.e_doc_structure) <br>        || !DataExtractionModule.IsModuleAvailable(DataExtractionModule.DataExtractionEngine.e_tabular)) <br>    { <br>        Console.Error.WriteLine(&quot;Data Extraction module not found. Check the resource search path.&quot;); <br> Environment.ExitCode = 1; <br>        return; <br>    } <br>    string statementPath = &quot;quarterly-statement.pdf&quot;; <br>    string structureJson = DataExtractionModule.ExtractData(statementPath, DataExtractionModule.DataExtractionEngine.e_doc_structure); <br>    string tablesJson = DataExtractionModule.ExtractData(statementPath, DataExtractionModule.DataExtractionEngine.e_tabular); <br>    System.IO.File.WriteAllText (&quot;quarterly-statement.structure.json&quot;, structureJson); <br>    System.IO.File.WriteAllText(&quot;quarterly-statement.tables.json&quot;, tablesJson); <br>    DataExtractionModule.ExtractToMarkdown(statementPath, &quot;quarterly-statement.md&quot;); <br>} <br>catch (PDFNetException ex) { Console.Error.WriteLine(ex.Message); Environment.ExitCode = 1; } <br>finally { PDFNet.Terminate(); } </pre><p>ExtractData takes a file path and an engine, and returns the JSON as a string. ExtractToMarkdown writes the Markdown file directly. Extraction runs in your own process, so no document leaves your environment. IsModuleAvailable returns false when the resource search path is wrong, which makes the check fail fast with a readable message. Pass the Lib folder itself, not the Windows or Linux subfolder inside it. A missing module sets a non-zero exit code, so a scheduler sees the failure. Each call is a separate extraction, so run only the outputs your pipeline reads.</p><p><strong>What the JSON Looks Like</strong></p><p>The shapes below are illustrative, not captured output. Property names in your output follow the <a href="https://sdk.apryse.com/api/structurejson/index.html">JSON specification for tabular data and document structure</a>, so treat the specification as the reference, including the coordinate origin, before you draw boxes on a page.</p><pre>{ <br>  &quot;pages&quot;: [ <br>    { <br>      &quot;pageNumber&quot;: 3, <br>      &quot;elements&quot;: [ <br>        { <br>          &quot;type&quot;: &quot;heading&quot;, <br>          &quot;text&quot;: &quot;Operating Results&quot;, <br>          &quot;boundingBox&quot;: { &quot;x&quot;: 72.0, &quot;y&quot;: 96.4, &quot;width&quot;: 210.5, &quot;height&quot;: 22.0 } <br>        }, <br>        { <br>          &quot;type&quot;: &quot;paragraph&quot;, <br>          &quot;text&quot;: &quot;Freight revenue grew in the third quarter as volume recovered.&quot;, <br>          &quot;boundingBox&quot;: { &quot;x&quot;: 72.0, &quot;y&quot;: 130.2, &quot;width&quot;: 220.0, &quot;height&quot;: 64.8 } <br>        }, <br>        { <br>          &quot;type&quot;: &quot;footer&quot;, <br>          &quot;text&quot;: &quot;Page 3 of 12&quot;, <br>          &quot;boundingBox&quot;: { &quot;x&quot;: 72.0, &quot;y&quot;: 744.0, &quot;width&quot;: 60.0, &quot;height&quot;: 12.0 } <br>        } <br>      ] <br>    } <br>  ] <br>} </pre><p>This excerpt of the tables file, trimmed to the fields that matter, shows the merged Operating Expenses header as one cell spanning three columns, above the column labels and the first data row.</p><pre>{ <br>  &quot;pageNumber&quot;: 3, <br>  &quot;tables&quot;: [ <br>    { <br>      &quot;boundingBox&quot;: { &quot;x&quot;: 72.0, &quot;y&quot;: 310.5, &quot;width&quot;: 468.0, &quot;height&quot;: 120.0 }, <br>      &quot;rows&quot;: [ <br>        { &quot;cells&quot;: [ { &quot;text&quot;: &quot;Operating Expenses&quot;, &quot;columnSpan&quot;: 3 } ] }, <br>        { &quot;cells&quot;: [ { &quot;text&quot;: &quot;Category&quot; }, { &quot;text&quot;: &quot;Q3&quot; }, { &quot;text&quot;: &quot;Q2&quot; } ] }, <br>        { &quot;cells&quot;: [ { &quot;text&quot;: &quot;Fuel&quot; }, { &quot;text&quot;: &quot;412,300&quot; }, { &quot;text&quot;: &quot;398,150&quot; } ] } <br>      ] <br>    } <br>  ] <br>} </pre><p><strong>Production Considerations for PDF to JSON Extraction</strong></p><ul><li><strong>Trial output is not a test fixture:</strong> Trial output is limited to 100 pages per operation, inserts a random watermark page, and can add an evaluation message to the JSON. Do not build schema assertions from it.</li><li><strong>Check the module once, at startup:</strong> A wrong Lib path surfaces as IsModuleAvailable returning false. Fail the process there, and call PDFNet.Initialize once per process rather than per document. In a long-running worker, call PDFNet.Terminate at shutdown.</li><li><strong>Set the OCR language when you use Paddle:</strong> If you choose the Paddle OCR engine for scanned statements, set the OCR language explicitly. Unlike the default engine, Paddle does not detect language automatically. Password-protected PDFs and page ranges have their own options in the <a href="https://docs.apryse.com/core/smart-data-extraction/workflow">workflow guide</a>.</li><li><strong>Validate tables with your own checks:</strong> Because tables carry no confidence score, sum each column and compare it to the printed total. Route mismatches to a review queue in your app.</li><li><strong>Match dev machines to the platform:</strong> Extraction runs on x64 Windows and Linux only. A developer on an ARM or macOS machine cannot run it locally, so give the team a Linux container.</li></ul><p><strong>What to Test Before You Trust the JSON</strong></p><ul><li><strong>Reading order:</strong> Open the PDF beside the Markdown file and read the Markdown top to bottom on a two-column page. Confirm no footer text appears mid-paragraph.</li><li><strong>Merged cells and multi-row headers:</strong> Run 10 real statements with merged headers. Render each page, draw the table bounding boxes from the JSON on it, and compare every cell value against the page.</li><li><strong>Cross-page tables:</strong> Pick a table that spans two pages. Count its rows in the PDF and in the tables JSON, confirm every continuation row appears, and check in the JSON specification how the two parts link.</li><li><strong>Totals:</strong> Recompute each column total from the extracted cells and compare it to the printed total.</li><li><strong>Native versus scanned:</strong> Run a native statement and a scanned copy of it. Diff the two outputs and inspect every mismatch by hand.</li></ul><p><strong>Where Extraction Sits Before the Model</strong></p><p>Extraction is the first step in your ingestion path. Your app validates table totals, stores the JSON, and decides what reaches a model. Send the Markdown to your index or LLM so headings and tables arrive in reading order. Keep the structure JSON when you need page numbers and bounding boxes to highlight a value on the source page.</p><p><strong>Setup Walkthroughs for Other Languages</strong></p><p>C# is one of several Server SDK bindings, and the <a href="https://docs.apryse.com/core/guides/smart-data-extraction">Smart Data Extraction guides</a> show each call in C#, C++, Go, Java, JavaScript, PHP, Python, Ruby, and VB. Companion posts cover install and a runnable sample in four languages, so this post skips that setup:</p><ul><li><strong>Node.js:</strong> <a href="https://apryse.com/blog/pdf-data-extraction-with-nodejs">Extract PDF Data in Node.js: Tables, Fields and JSON</a> extracts form fields from scanned and digital PDFs and shows short excerpts for tables and document structure.</li><li><strong>Python:</strong> <a href="https://apryse.com/blog/pdf-data-extraction-with-python">PDF Data Extraction in Python: Tables, Key-Value Pairs, and Form Fields</a> covers key-value pairs, tables, document structure, form fields, and XLSX export.</li><li><strong>Java:</strong> <a href="https://apryse.com/blog/data-extraction-using-java">How to Extract Data from PDF in Java with Apryse Smart Data Extraction</a> extracts form fields from scanned and digital PDFs.</li><li><strong>C#:</strong> <a href="https://apryse.com/blog/extract-data-from-pdf-with-c-sharp">Extract PDF Data in C# and .NET: Tables, Fields, JSON</a> covers form field key-value extraction as JSON in C#, with the NuGet install and module setup. It leaves out tables and document structure, which this post covers.</li></ul><p>For PHP, Ruby, Go, and C++, start from the <a href="https://docs.apryse.com/core/guides/get-started">Server SDK get-started guide</a>.</p><p><strong>Getting Started</strong></p><p>Install the base SDK for .NET through NuGet. The <a href="https://apryse.com/blog/extract-data-from-pdf-with-c-sharp">C# companion post</a> shows the package and the Linux command. Download the Data Extraction module from the <a href="https://docs.apryse.com/core/guides/info/modules#data-extraction-module">modules page</a> and expand it into a Lib folder next to your built application.</p><p>The full DataExtractionTest sample is on the <a href="https://docs.apryse.com/core/guides/smart-data-extraction/samples">Smart Data Extraction samples page</a>. Get your free trial key at dev.apryse.com before running the code.</p><p>Go back to that Monday morning.</p><p>Dana opens her queue and the same question lands: where did 412,300 come from? Her team’s app stored each value with its table, its page, and its bounding box, so the answer is one click away. She opens the PDF to page 3, finds the Q3 column, and there is the number, sitting inside its box right where the statement printed it. She closes the ticket and finishes her coffee while it is still warm.</p><p>The overnight job did not get smarter. It started from structure instead of a stream of characters, and every number kept its address. Dana never had to wonder.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd0218a1ccd1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-to-convert-a-pdf-to-structured-json-in-c-bd0218a1ccd1">How to Convert a PDF to Structured JSON in C#</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Redact PDFs in C# with the Apryse Server SDK]]></title>
            <link>https://medium.com/apryse/how-to-redact-pdfs-in-c-with-the-apryse-server-sdk-e6d875cc4779?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/e6d875cc4779</guid>
            <category><![CDATA[csharp]]></category>
            <category><![CDATA[document-redaction]]></category>
            <category><![CDATA[pdf-sdk]]></category>
            <dc:creator><![CDATA[Isaac Maw]]></dc:creator>
            <pubDate>Fri, 23 Oct 2026 17:00:09 GMT</pubDate>
            <atom:updated>2026-10-23T17:00:08.487Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g-15UfY97hvmfzL0djxxFQ.png" /></figure><p>A black rectangle over an account number looks like redaction, but select the area, copy it, or run text extraction, and the original string comes right back. The account number is <a href="https://www.americanbar.org/groups/judicial/resources/judges-journal/archive/embarrassing-redaction-failures/">still recoverable.</a></p><p>This tutorial demonstrates <a href="https://apryse.com/capabilities/redaction/csharp">coordinate-based PDF redaction in C#</a> with the Apryse Server SDK. The example marks a region, removes its content and saves a separate output file. It also covers how to identify regions through text search and what to check before releasing the result.</p><p>The SDK runs in your own backend on Windows, Linux or macOS. This example targets C# on .NET 8 and 9. For setup instructions, see the <a href="https://docs.apryse.com/core/guides/get-started/dotnet">C# get-started guide</a> in the Apryse documentation.</p><h3>Why a Filled Rectangle Leaves Sensitive Data Behind</h3><p>Drawing a black box over text changes the page’s appearance, but leaves the underlying content in the PDF. Someone who extracts the text may still recover it.</p><p>True redaction removes content from the file. With the Apryse Server SDK, the portions of text, images and vector graphics inside a redaction region are destroyed when the redaction is applied. The overlay shows where content was removed.</p><p>Hidden information needs separate attention. Metadata, embedded scripts, attachments and form data can contain sensitive information beyond the page content you redact. Include <a href="https://apryse.com/capabilities/pdfsanitization">sanitization</a> in your release process when those also need to be removed.</p><h3><strong>Using the Apryse Server SDK</strong></h3><p>The Apryse Server SDK is a self-hosted library you embed in a backend application. Its C# API uses the pdftron namespace, with PDFNet as the initialization entry point. Document processing runs inside your service, so you do not need to upload PDFs to Apryse for redaction.</p><p>Redaction requires an add-on license. It ships in the base binary, so there is no separate module to download, but your production key needs the Redaction Package. Trial keys include all packages.</p><p>Remember to keep your license key out of source control. Load it from an environment variable or a secrets manager and pass it to your application at runtime.</p><p><strong>Apply a Redaction in C#</strong></p><p>Here’s how to use Apryse redaction:</p><p>Identify the regions before applying the redaction. Depending on the workflow, those regions can come from:</p><ul><li>Redact annotations placed by a reviewer.</li><li>Matches found with pdftron.PDF.TextSearch, including regular expressions.</li><li>Locations identified through pdftron.PDF.TextExtractor.</li><li>Known coordinates, such as a fixed field on a statement.</li></ul><p>An unapplied mark does not remove content. Your application can review or adjust the regions before calling Redactor.Redact.</p><p>The following example removes content from one rectangle on page 1 and saves the result to a new file.</p><pre>using System; <br>using System.Collections; <br>using pdftron; <br>using pdftron.PDF; <br>using pdftron.SDF; <br> <br>public class StatementRedactionService <br>{ <br>    public void RedactAccountRegion( <br>        string licenseKey, <br>        string statementPath, <br>        string redactedPath) <br>    { <br>        PDFNet.Initialize(licenseKey); <br> <br>        try <br>        { <br>            using PDFDoc statementDocument = new PDFDoc(statementPath); <br>            statementDocument.InitSecurityHandler(); <br> <br>            ArrayList redactionRegions = new ArrayList <br>            { <br>                new Redactor.Redaction( <br>                    1, <br>                    new Rect(100, 100, 550, 600), <br>                    false, <br>                    &quot;REDACTED&quot;) <br>            }; <br> <br>            Redactor.Appearance overlayStyle = new Redactor.Appearance <br>            { <br>                RedactionOverlay = true <br>            }; <br> <br>            Redactor.Redact( <br>                statementDocument, <br>                redactionRegions, <br>                overlayStyle, <br>                ext_neg_mode: false, <br>                page_coord_sys: true); <br> <br>            statementDocument.Save( <br>                redactedPath, <br>                SDFDoc.SaveOptions.e_linearized); <br>        } <br>        catch (PDFNetException e) <br>        { <br>            Console.WriteLine(e.Message); <br>            throw; <br>        } <br>        finally <br>        { <br>            PDFNet.Terminate(); <br>        } <br>    } <br>} </pre><p>Replace the example rectangle with coordinates that match your document. It covers a large area and is not intended to locate an account number automatically.</p><p>The <strong>Redactor.Redaction</strong> constructor takes the page number, rectangle, a negative-redaction flag and overlay text. Setting the negative flag to false selects content inside the rectangle for removal.</p><p><strong>Redactor.Appearance</strong> controls the overlay. Here, <strong>RedactionOverlay = true</strong> enables it, and the region supplies the label “REDACTED.” Changing the overlay’s appearance does not change whether the underlying content is removed.</p><p>The two named Boolean arguments in <strong>Redactor.Redact </strong>control scope and coordinates:</p><ul><li><strong>ext_neg_mode:</strong> false keeps negative redactions localized to their specified page. The example uses an ordinary redaction, so it does not rely on negative-redaction behavior.</li><li><strong>page_coord_sys:</strong> true interprets the rectangle relative to the page’s lower-left corner. Check this coordinate convention when translating selections from a viewer.</li></ul><p>The Apryse <a href="https://sdk.apryse.com/api/PDFTronSDK/dotnetcore/pdftron.PDF.Redactor.html">Redactor API reference</a> documents these parameters.</p><h3>Find Regions with Text Search</h3><p>Fixed coordinates work for consistent layouts. If an account number can appear anywhere on a page, use text search to locate it before creating the redaction regions.</p><p>With TextSearch, your code searches for a string or regular expression, collects the matching locations and passes those regions to the apply step. Review the matches before removing content.</p><p>For example, a pattern intended to find Social Security numbers may also match reference numbers with the same digit grouping. Test patterns against representative documents and check both missed values and unwanted matches.</p><p>Search also depends on the text available in the PDF. For scanned pages without a text layer, include OCR in your identification workflow.</p><h3><strong>Handle Initialization and Failures in a Service</strong></h3><p>The example initializes and terminates PDFNet within one method to show the complete lifecycle. In a running service, initialize it once at startup and terminate it at shutdown.</p><p>For a service processing multiple PDFs:</p><ul><li><strong>Use separate document instances.</strong> Open a separate PDFDoc for each worker rather than sharing one document across threads.</li><li><strong>Keep the original during review.</strong> Save to a different path and release the output only after verification passes.</li><li><strong>Handle failures per document.</strong> Catch PDFNetException within the batch loop, log a document identifier and quarantine failed inputs.</li><li><strong>Control concurrency and retries.</strong> Your application owns the queue, scheduling and retry behavior.</li><li><strong>Make approval explicit.</strong> If a reviewer must approve regions, complete that approval before calling Redactor.Redact.</li></ul><p>Avoid logging the sensitive strings you are trying to remove.</p><h3><strong>Verify the Saved PDF</strong></h3><p>A rectangle placed a line too low can remove the wrong content and still produce a valid PDF. To double-check, reopen the saved output and check both the removed content and the surrounding page.</p><p><strong>Extract text from the redacted pages.</strong> Use TextExtractor to check whether the targeted strings remain. This catches cases where text was merely covered.</p><p><strong>Repeat the search.</strong> Run the identification patterns against the output and investigate remaining matches. Expect zero matches only when every occurrence of that pattern was intended for removal.</p><p><strong>Inspect the page visually.</strong> Confirm that the overlays cover the intended regions and that nearby content remains readable. Text extraction alone cannot verify redaction of scanned images or vector graphics.</p><p><strong>Check hidden information separately.</strong> Apply sanitization after redaction when required, then verify the final file that will be distributed. Reviewing document properties in a viewer can help, but it does not provide a complete inventory of hidden data.</p><h3><strong>Getting Started</strong></h3><p>That’s all there is to it. Follow the <a href="https://docs.apryse.com/core/guides/get-started/dotnet">C# get-started guide</a> to install the Server SDK and configure your project. Get a trial key at <a href="https://dev.apryse.com/">dev.apryse.com</a>, then run the PDFRedact sample against a copy of one of your documents.</p><p>Here’s a tip: start with a known region so you can verify the removal. Once that works, add the search or review workflow your application needs, and scale it up.</p><h3><strong>FAQ</strong></h3><p><strong>Can I undo an applied redaction?</strong></p><p>The removed content cannot be recovered from the redacted output. Keep the original separately if your workflow requires it.</p><p><strong>Can I change the overlay’s appearance?</strong></p><p>Yes. Redactor.Appearance provides options for the overlay, including its color and text styling.</p><p><strong>Do I need a separate redaction module?</strong></p><p>No separate download is required. Redaction ships in the Server SDK binary, but production use requires a license key with the Redaction Package.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6d875cc4779" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-to-redact-pdfs-in-c-with-the-apryse-server-sdk-e6d875cc4779">How to Redact PDFs in C# with the Apryse Server SDK</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Fast Is Apryse OCR v12.2? Benchmarks Against PaddleOCR and v12.0]]></title>
            <link>https://medium.com/apryse/how-fast-is-apryse-ocr-v12-2-benchmarks-against-paddleocr-and-v12-0-1ae6dc5c4fbf?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/1ae6dc5c4fbf</guid>
            <category><![CDATA[apryse]]></category>
            <category><![CDATA[ocr-performance]]></category>
            <category><![CDATA[ocr-memory-usage]]></category>
            <category><![CDATA[apryse-ocr]]></category>
            <category><![CDATA[server-side-ocr]]></category>
            <dc:creator><![CDATA[Tabinda Khan]]></dc:creator>
            <pubDate>Mon, 19 Oct 2026 13:01:03 GMT</pubDate>
            <atom:updated>2026-10-19T13:01:02.888Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b0Ucs8aHbZHQIxDVAmpIuw.png" /></figure><p><em>Apryse OCR v12.2 runs up to 4.27× faster with up to 78% lower peak memory in internal benchmarks, so your workers clear the queue sooner.</em></p><h3><strong>Why Does OCR Speed Matter in a Document Pipeline?</strong></h3><p>Every scanned document waits on OCR. Now it waits less.</p><p>Trace one scanned invoice through an ingestion pipeline. It lands in a queue as an image. It waits for an OCR worker. The worker turns pixels into a text layer, and only then can search index it, extraction parse it, or a reviewer find the total.</p><p>Search, extraction, and review all wait on that one step.</p><p>Now multiply it by a morning’s intake of bank statements, receipts, cheques, and passports. The queue grows when OCR is slow. It also grows when OCR is memory-hungry, because each worker’s peak memory caps how many workers fit on a node. Push past that limit, and the container is stopped mid-batch.</p><p>OCR rarely gets attention in architecture reviews. It sets the pace anyway.</p><h3><strong>What Is New in Apryse OCR v12.2?</strong></h3><p>Apryse OCR v12.2, the OCR module for the Apryse Server SDK, makes that step shorter and lighter.</p><p>In internal benchmarks, v12.2 finished 2.59× faster than v12.0, cutting runtime by 61.4%. Against stock PaddleOCR V6 Server, it ran up to 4.27× faster and used up to 78% less peak memory.</p><p>For a pipeline, that changes two numbers at once: how long a batch takes, and how many workers fit on the same hardware.</p><h3>Why the OCR Step Was Costing You Capacity</h3><p><strong>Runtime sets your worker count.</strong> If a batch has to clear by a deadline, a slower OCR step means more workers running in parallel. Each one is compute you pay for.</p><p><strong>Peak memory sets your density.</strong> Containers are sized for a worker’s worst moment, not its average one. A worker that spikes on a large scan forces a higher memory limit, and a higher limit means fewer workers per node.</p><p><strong>Image size changes the math.</strong> A 2 MB scan and a 33 MB scan do not cost the same to process. A pipeline tuned on small, clean samples can behave very differently on real intake.</p><p><strong>Speed alone is not the goal.</strong> A run that misreads words still produces a searchable PDF that looks valid. A faster engine only helps if its output holds up.</p><h3>What the Benchmarks Show</h3><p>Apryse measured v12.2 against two baselines: the previous release, v12.0, and stock PaddleOCR V6 Server.</p><ul><li><strong>vs. v12.0, runtime:</strong> 3:53.823 down to 1:30.155. That is 2.59× faster, with 61.4% less time.</li><li><strong>vs. v12.0, peak memory:</strong> 20.1% to 44.5% lower across tested image sizes, with a 1.61× lower average footprint.</li><li><strong>vs. stock PaddleOCR V6 Server, runtime:</strong> 26:00.060 down to 6:05.676. That is 4.27× faster, with 76.6% less time.</li><li><strong>vs. stock PaddleOCR V6 Server, peak memory:</strong> 56.9% to 78.0% lower, with a 3.58× lower average footprint. The largest reduction came on 8 MB images.</li></ul><p>Memory tests used small, medium, and large images of about 2 MB, 8 MB, and 33 MB. Each comparison stands on its own, because the two runtimes come from separate benchmark runs.</p><p>These are internal benchmarks. Results vary with document type, image size, language, hardware, runtime configuration, batch size, and workload.</p><h3><strong>Does Apryse OCR v12.2 Change OCR Accuracy?</strong></h3><p>No. Apryse OCR v12.2 is a speed and memory release, and recognition accuracy is unchanged from v12.0. The accuracy gains came in v12.0, which replaced the v11 OCR engine with a deep-learning model. Compared with v11, Apryse OCR v12.0 and v12.2 deliver:</p><ul><li><strong>About 16% better word recognition accuracy</strong></li><li><strong>Better results on low-quality scans</strong> under 300 DPI</li><li><strong>Better handling of skewed pages</strong> and complex, multidirectional layouts</li><li><strong>Wider language coverage</strong>, adding Chinese, Japanese, Korean, Portuguese, and more Latin-alphabet languages, for 80+ languages in total</li></ul><h3><strong>Where Does Apryse OCR Run?</strong></h3><p>Apryse OCR runs in-process inside the Apryse Server SDK, on infrastructure you control: physical servers, virtual machines, private cloud, or containers. No document content goes to Apryse or any third party for processing, and the engine needs no GPU.</p><p>Since SDK 12.0, the module has used a deep-learning engine that handles low-quality scans, skewed pages, and multidirectional layouts across more than 80 languages. Version 12.2 makes that engine cheaper to run. Apryse validated its performance and behavior across 18 complete regression runs.</p><p>The SDK keeps no state between documents, so you choose the topology: sequential, threaded, multi-process, or scaled out across nodes. Lower peak memory per worker gives you more room in whichever one you run.</p><h3>Getting Started</h3><ol><li>Get a free trial key at dev.apryse.com. Trial keys include unlimited access to add-on modules, including OCR.</li><li>Install the Server SDK and the OCR module. For Node.js on Windows or Linux, run npm i @pdftron/pdfnet-node and npm install @pdftron/ocr. For other languages, see docs.apryse.com/core/guides/info/modules#ocr-module.</li><li>Follow the OCR workflow guide at docs.apryse.com/core/guides/ocr/workflow, then benchmark v12.2 against your own scans.</li></ol><p>Every scanned document still waits on OCR. With Apryse OCR v12.2, it waits less</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ae6dc5c4fbf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/how-fast-is-apryse-ocr-v12-2-benchmarks-against-paddleocr-and-v12-0-1ae6dc5c4fbf">How Fast Is Apryse OCR v12.2? Benchmarks Against PaddleOCR and v12.0</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Scanbot Fall 2026 Release: Document Cleanup, Barcode Reconstruction & Landscape Orientation Support]]></title>
            <link>https://medium.com/apryse/https-apryse-com-blog-scanbot-barcode-reconstruction-document-cleanup-8a43e47153b6?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/8a43e47153b6</guid>
            <category><![CDATA[barcode]]></category>
            <category><![CDATA[barcode-scanner]]></category>
            <category><![CDATA[document-capture]]></category>
            <category><![CDATA[barcode-scanning-api]]></category>
            <dc:creator><![CDATA[Laura Lee Massingham]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 15:53:41 GMT</pubDate>
            <atom:updated>2026-10-07T15:53:39.418Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E-5uKgTYsMpEajANGeiSAg.png" /></figure><p>A warehouse worker scans a barcode on a box that’s been dragged across a loading dock. The corner’s torn, the ink’s smudged, and the scanner just fails. So, they type the code in by hand, or the box goes to a rescan bin, and someone deals with it later. Multiply that by a few hundred packages a day and “later” becomes a real cost.</p><p>This is the gap between how capture works in a demo and how it works in the field. Labels get scuffed. Documents get creased, held open by a thumb, photographed at an angle on a warehouse floor instead of scanned flat on a desk. Most capture SDKs are built and tested against clean conditions, which is rarely what shows up in the field.</p><h3>Barcode Reconstruction (Barcode Scanner SDK)</h3><p>Picture a shipping label with a handwritten routing note scrawled across part of the barcode, or a box that got crushed, torn, or scuffed somewhere in transit. Traditional 1D barcode decoding, including dedicated laser scanners, needs one continuous scan line with the full sequence of bars and spaces intact. When no single line survives undamaged, the scan fails, no matter how good the scanner is.</p><p>Barcode Reconstruction works differently. It uses computer vision to pull together multiple partially readable scan lines and assemble a complete sequence from the fragments, even when no single line was ever intact on its own. That’s how it recovers barcodes damaged badly enough to defeat a dedicated laser scanner.</p><p>It runs inside the <a href="https://scanbot.io/barcode-scanner-sdk/">Barcode Scanner SDK</a>, as part of the normal scan. No separate repair step, no round trip to a server. A barcode that would have failed outright, and forced a manual entry or a rescan, now just reads.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*blZBUET4wVSi2hHcdK28pw.gif" /></figure><h3>Document Cleanup (Document Scanner SDK)</h3><p>Documents fail differently. The content is usually fine. What’s wrong is everything around it, a finger in frame, a hole punch, a stray mark from the handling of the page before it hit the scanner.</p><p>Document Cleanup, in the Document Scanner SDK, starts with the user marking the area to clean up. From there, computer vision takes over: it detects the surrounding text so cleanup doesn’t damage real content, then fills in the background underneath the artifact. The result is a copy closer to the original, not just an erased patch. The feature is available across all supported platforms, so the cleanup step is consistent no matter where the scan happened.</p><p>Both features are built on the same idea: most real-world capture failures are partial, not total, and the fix is using what survived instead of throwing out the whole scan.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*1MMxFykIUrOnKzV6FimQoQ.gif" /></figure><h3>Landscape Orientation Support</h3><p>Field capture isn’t confined to phones. This release adds support for landscape and large-screen capture flows, tablets, kiosks, unfolded foldables, and ChromeOS. That’s not optional with the June release of Android 17. If your app captures documents or barcodes on Android and you haven’t accounted for large-screen layouts, this closes that gap.</p><h3>Same outcome, less friction</h3><p>None of these capabilities change what your app does with a scan once it’s captured. It changes what happens when the scan isn’t clean. A damaged barcode recovers automatically, inside the SDK, before it ever reaches your user. A messy document gets a manual fix instead of a failed capture.</p><p><a href="https://apryse.com/blog/apryse-fall-2026-release">Read the full Fall Release details</a> on the Apryse website</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8a43e47153b6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/https-apryse-com-blog-scanbot-barcode-reconstruction-document-cleanup-8a43e47153b6">Scanbot Fall 2026 Release: Document Cleanup, Barcode Reconstruction &amp; Landscape Orientation Support</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Apryse Fall Release: Markdown Output, Highlighting, and Faster Document Editing]]></title>
            <link>https://medium.com/apryse/https-apryse-com-blog-apryse-fall-2026-release-d5cdfde79159?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/d5cdfde79159</guid>
            <category><![CDATA[apryse]]></category>
            <category><![CDATA[microsoft-office]]></category>
            <category><![CDATA[pdf]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Laura Lee Massingham]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 15:22:35 GMT</pubDate>
            <atom:updated>2026-10-07T15:22:33.957Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KxmzxPR0ePKzwJf97swBKg.png" /></figure><p><em>Across Web, Server, and Scanbot SDKs, this release trades custom workarounds for built-in capabilities.</em></p><p>This fall’s release is a handful of features and a few fewer workarounds you’ll need to build yourself.</p><p>Across the <a href="https://apryse.com/products/webviewer">Web</a> and <a href="https://apryse.com/products/core-sdk">Server</a> SDKs, things that used to require per-editor workarounds, downstream conversions, or low-level drawing code are now handled inside the SDK itself. Highlighting is now available in DOCX Editor, and comments extend to Spreadsheet Editor, so you’re not building a separate review workflow for every file type. Content Edit in WebViewer and search and replace in DOCX Editor both get faster. AEC workflows get easier, with vector stamp support and better stamp management. On the Server side, full document generation and Markdown extraction come straight out of the SDK, so you’re not converting or hand-coding what comes out the other end.</p><p>The <a href="https://scanbot.io/">Scanbot SDKs</a> make real-world capture hold up better, too. Barcode Reconstruction recovers codes that are scuffed or torn. <a href="https://apryse.com/blog/scanbot-barcode-reconstruction-document-cleanup">Document Cleanup </a>removes fingers, marks, and hole punches from scans, the stuff that used to fail outright or get kicked back for manual entry. Android 17 now requires landscape support for large-screen layouts, tablets, kiosks, foldables, and ChromeOS, and this release adds it.</p><h3>Web SDK: More Consistent Review Across Editors</h3><p>Commenting now shares the same Notes Panel UI across WebViewer, <a href="https://apryse.com/capabilities/docx-editor">DOCX Editor</a>, and <a href="https://apryse.com/capabilities/spreadsheet-editor">Spreadsheet Editor</a>. Highlighting reaches DOCX Editor too, using the same interaction pattern as WebViewer.</p><ul><li><a href="https://apryse.com/blog/docx-editor-highlight-tool"><strong>DOCX Editor gets a highlight tool</strong></a><strong>.</strong> A lightweight way to flag text without opening a full comment thread or tracked change, the same quick markup already available in WebViewer.</li><li><a href="https://apryse.com/blog/spreadsheet-editor-threaded-comments"><strong>Spreadsheet Editor adds comments</strong></a><strong>.</strong> Same Notes Panel as WebViewer and DOCX Editor: threaded, cell-level, and available through both UI and API. Legacy Excel Notes convert automatically, so comment history isn’t stranded when a workbook moves into Spreadsheet Editor. (Note: .xls files are the exception, those don’t carry over.)</li><li><a href="https://apryse.com/capabilities/pdf-editing">Content Edit</a> in WebViewer and search and replace in DOCX Editor are also faster. Search and replace no longer runs on a manual frontend implementation, which was the actual bottleneck on long documents.</li><li>Field navigation for signing workflows gets easier too, if you’re building a custom e-signature experience on WebViewer.</li><li>And finally, <a href="https://apryse.com/industries/aec">AEC workflows</a> get three upgrades. PDF content can now be imported directly as vector stamps, so custom symbol libraries hold their fidelity instead of rasterizing when they leave WebViewer for another PDF editor. New stamp management replaces the single unwieldy list for teams with growing stamp libraries. And custom line and fill rendering, previously available only through code, is now built into the UI to support engineering workflows with recurring line types.</li></ul><p>Read the full details of the Web SDK release:</p><ul><li><a href="https://apryse.com/blog/webviewer-vector-stamps-content-edit">WebViewer Update: Vector Stamps, Faster Content Edit, and Easier Signing</a></li><li><a href="https://apryse.com/blog/docx-editor-highlight-tool">Mark Up Documents Faster in DOCX Editor</a></li><li><a href="https://apryse.com/blog/spreadsheet-editor-threaded-comments">Spreadsheet Editor Adds Threaded Comments: One Review Experience Across PDF, DOCX, and Excel</a></li></ul><h3>Server SDK: From Parsing to Generation, Built In</h3><p>This release brings two additions on the Server side, pointing opposite ways: structure coming out of documents, and documents built from structure.</p><ul><li><a href="https://apryse.com/blog/smart-data-extraction-markdown-output"><strong>Markdown is now an output option for Smart Data Extraction</strong></a><strong>.</strong> It’s built on Apryse’s document-structure intelligence, so hierarchy and context survive instead of getting flattened into plain text, which matters if that output feeds ingestion or retrieval. Processing stays server-side, which matters when the documents are contracts, claims files, or patient records that can’t leave your infrastructure.</li><li><a href="https://apryse.com/blog/document-creation-api-pdf-generation"><strong>The Document Creation API now covers the whole document.</strong></a> Paragraphs, tables, and lists were already there. This release adds hyperlinks, page numbers, headers and footers, images, shapes, floating elements, charts, and metadata. So, a branded multi-page invoice with a logo, a line-item table, and a numbered footer is a high-level API call rather than a tedious coordinate system you manage yourself.</li></ul><p>Read more details of the Server SDK release:</p><ul><li><a href="https://apryse.com/blog/document-creation-api-pdf-generation">Programmatic PDF Generation Without the Coordinate Math</a></li><li><a href="https://apryse.com/blog/smart-data-extraction-markdown-output">What Gets Lost Between the PDF and the Prompt: Markdown Output for Smart Data Extraction</a></li></ul><h3>Scanbot SDKs: Capture That Survives Real Conditions</h3><p>Field capture is messy. Damaged source documents, someone holding a page open with their thumb, a barcode that has been marked up or torn. We’ve added two new features that improve results in these conditions:</p><ul><li><a href="https://apryse.com/blog/scanbot-barcode-reconstruction-document-cleanup"><strong>Barcode Reconstruction</strong></a> reads what’s still intact on a damaged barcode and rebuilds the rest, inside the SDK, before it ever becomes a fallback your end user has to deal with.</li><li><a href="https://apryse.com/blog/scanbot-barcode-reconstruction-document-cleanup"><strong>Document Cleanup</strong></a> lets users manually remove fingers, marks, and hole punches from a scan after the fact, producing a copy closer to the original.</li><li>Android 17, released in June 2026, requires landscape support for large-screen layouts. This release satisfies this requirement: tablets, kiosks, unfolded foldables, and ChromeOS are all covered.</li></ul><p>Ready to get started? Build with what’s new this fall. <a href="https://docs.apryse.com/">Check out the docs</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d5cdfde79159" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/https-apryse-com-blog-apryse-fall-2026-release-d5cdfde79159">Apryse Fall Release: Markdown Output, Highlighting, and Faster Document Editing</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Spreadsheet Editor Adds Threaded Comments: One Review Experience Across PDF, DOCX, and Excel]]></title>
            <link>https://medium.com/apryse/spreadsheet-editor-adds-threaded-comments-one-review-experience-across-pdf-docx-and-excel-39a506bd9d58?source=rss----f72464c16b46---4</link>
            <guid isPermaLink="false">https://medium.com/p/39a506bd9d58</guid>
            <category><![CDATA[collaboration]]></category>
            <category><![CDATA[spreadsheets]]></category>
            <category><![CDATA[sdk]]></category>
            <category><![CDATA[spreadsheet-editor]]></category>
            <category><![CDATA[web-sdk]]></category>
            <dc:creator><![CDATA[Mia Perese]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 15:13:49 GMT</pubDate>
            <atom:updated>2026-10-07T15:13:48.247Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xtmxdcIavSt-adf9SniuxA.png" /></figure><p><a href="https://apryse.com/capabilities/spreadsheet-editor">Spreadsheet Editor</a> now supports threaded comments and replies, using the same Notes Panel and API already live in <a href="https://apryse.com/products/webviewer">WebViewer</a> and <a href="https://apryse.com/capabilities/docx-editor">DOCX Editor</a>. Teams reviewing PDF, DOCX, and Excel files now get one consistent way to comment, reply, and collaborate across every format.</p><h3>How Does Comment Support in Spreadsheet Editor Work?</h3><p>With Spreadsheet Editor’s threaded comments and replies now available through both the UI and a public API, users no longer have to leave the app to provide feedback. A reviewer flagging a formula or specific cell can leave a note and return to editing in the same session.</p><p>When opening a file in Spreadsheet Editor from Excel or another spreadsheet tool, existing comments or legacy notes carry over automatically, so review history remains intact.</p><p>Legacy Notes in .xls files won’t carry over, since .xls files convert to .xlsx on load.</p><h3>Comments in Spreadsheet Editor: User Experience and Functionality</h3><p>Comments are linked to specific cells and allow threaded replies, enabling reviewers to leave notes or respond directly. Each comment displays the message, author, date, and cell location in the Notes Panel, and clicking a comment takes you directly to that cell.</p><h3>One Review Experience Across PDF, DOCX, and Excel</h3><p>Spreadsheet Editor’s comments run on the same Notes Panel and comment pattern already built for WebViewer and DOCX Editor, so a workbook’s comments look and behave the way comments already do everywhere else in the SDK. With the Apryse Web SDK, a team redlining a contract in PDF, a proposal in DOCX, and a budget in Excel gets one consistent way to create and respond to comments.</p><h3>How to Start Building</h3><p>Get started: <a href="https://dev.apryse.com/">https://dev.apryse.com/</a></p><p>Docs: <a href="https://docs.apryse.com/web/spreadsheet-editor/spreadsheet-editor">https://docs.apryse.com/web/spreadsheet-editor/spreadsheet-editor</a></p><p>Release notes: <a href="https://docs.apryse.com/web/release-notes/release-notes">https://docs.apryse.com/web/release-notes/release-notes</a></p><h3>More from Spreadsheet Editor’s Fall Release</h3><p>Spreadsheet Editor’s new comments capability ships alongside two more releases in the same Fall Web SDK cycle. WebViewer delivers faster and more reliable Content Edit performance and improves stamp management for AEC teams. DOCX Editor adds a highlight tool for flagging text without a full comment thread, along with faster search and replace on long documents.</p><p>Read the WebViewer release blog: <a href="https://apryse.com/blog/webviewer-vector-stamps-content-edit">https://apryse.com/blog/webviewer-vector-stamps-content-edit</a></p><p>Read the DOCX Editor release blog: <a href="https://apryse.com/blog/docx-editor-highlight-tool">https://apryse.com/blog/docx-editor-highlight-tool</a></p><p>See everything shipping this fall: <a href="https://apryse.com/blog/apryse-fall-2026-release">https://apryse.com/blog/apryse-fall-2026-release</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=39a506bd9d58" width="1" height="1" alt=""><hr><p><a href="https://medium.com/apryse/spreadsheet-editor-adds-threaded-comments-one-review-experience-across-pdf-docx-and-excel-39a506bd9d58">Spreadsheet Editor Adds Threaded Comments: One Review Experience Across PDF, DOCX, and Excel</a> was originally published in <a href="https://medium.com/apryse">Apryse</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>