<?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[Stories by Anuj Shah on Medium]]></title>
        <description><![CDATA[Stories by Anuj Shah on Medium]]></description>
        <link>https://medium.com/@chelmanuhazard?source=rss-7fdca73aba87------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*ilr_KnJ2xnF2lq6DH1HOGw.jpeg</url>
            <title>Stories by Anuj Shah on Medium</title>
            <link>https://medium.com/@chelmanuhazard?source=rss-7fdca73aba87------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 04:20:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@chelmanuhazard/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Improving Customer Buying Decision Using ExtractiveText Summarization(TextRank Algorithm)]]></title>
            <link>https://medium.com/analytics-vidhya/improving-customer-buying-decision-using-extractivetext-summarization-textrank-algorithm-37e823419b5c?source=rss-7fdca73aba87------2</link>
            <guid isPermaLink="false">https://medium.com/p/37e823419b5c</guid>
            <category><![CDATA[textrank]]></category>
            <category><![CDATA[graph]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Anuj Shah]]></dc:creator>
            <pubDate>Fri, 15 Nov 2019 16:32:04 GMT</pubDate>
            <atom:updated>2019-12-28T13:55:08.271Z</atom:updated>
            <content:encoded><![CDATA[<h3>Improving Customer Buying Decision Using Extractive Text Summarizatio<strong>n(TextRank Algorithm)</strong></h3><p>With the growing need of the internet for fetching information and availability of huge amounts of data, it becomes important to filter the content and convert the abundant information into a summary capturing the key components which can be helpful to analyze and make decisions. In the age of technology, people are informed about any piece of news by reading a short limited version of the news.</p><p>Manually analyzing each segment of a text and generating summaries can become both time consuming as well as backbreaking. Therein, comes the need of an automatic text summarization technique representing the most relevant and voted responses can help the audience absorb the information and saves time from going through redundant data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/260/1*al__sM7udm40z_jpGPTLYg.jpeg" /></figure><p>Online Buying has become an everyday part of human life. Summarization of reviews of a particular product by different consumers can help a potential buyer identify the product usefulness and its ability to satisfy consumer needs. Summarization, using intelligent computer based algorithms can help capture the essential syntactical and semantic representations of the text. The use of such a system can help consumers make effective decisions about buying a product.</p><h3>RELATED WORK</h3><p><strong><em>Hyper Induced Topic Search(HITS)</em></strong></p><p>Hyper Induced Topic Search algorithm was developed by Jon Kleinberg used to rank web pages for a search query. The same algorithm can be applied to text summarization where, instead of the web pages, we rank the sentences in the text. The words are represented Term Frequency-Inverse document frequency values and the sentence similarity is calculated using cosine similarity. A sentence can be considered as pointing to other sentence if the similarity measure between the given pair exceeds a threshold value. The two scores used in the algorithm are Authority Value and Hub Value.</p><p><strong>Authority Value:</strong> For a given sentence, it’s authority score is the sum of the hub value of the sentences that points to it. A sentence is said to possess authority if it is pointed by other sentences[3].</p><p><strong>Hub Value:</strong> The hub score is the sum of all the authority value of the sentences it points to[3].</p><p>The two values imply that a sentence has a higher score if it’s been pointed by higher hub valued sentences. Initially, the hub values for sentences are 1 and authority update rule and hub update rule is applied until convergence.</p><p><strong><em>PageRank Algorithm</em></strong></p><p>PageRank algorithm is used by Google to determine a webpage’s relevance for a query. Webpages are said to be connected when one contains the link to others. PageRank deals with the probability of jumping to a particular webpage[4]. A similar system can be imitated for using PageRank as TextRank. In textrank, the tokenized sentences will be considered as the nodes for our graph and weights represent the sentence similarity score. Use of an iterative formula until score converge below a threshold value.</p><p><em>PR(A) = (1-d) + d (PR(T1)/C(T1) + … + PR(Tn)/C(Tn))</em></p><p><strong><em>Shortest Path Algorithm</em></strong></p><p>In case of ranking, the selected sentences belong to different parts of the text and there is a possibility of sudden topic shifts while generating the summary. In this approach, the sentences form the nodes and weights are assigned based on number of common words. Further the sentences in the original text, the higher the cost.</p><p><em>Sim(Si , Sj ) = |{wk|wk ∈ Si&amp;wk ∈ Sj}| log(|Si |) + log(|Sj|)</em></p><p>where, wk represent the words in the sentence.</p><p>The goal is to start from the first sentence of the original text and reach to the last sentence choosing the cheapest path discarding the loop paths.</p><h3><strong>PROPOSED WORK</strong></h3><p>focuses on extractive Graph based summarization technique. The objective is to examine the effect of summarization in case of reviews/feedback of products. The procedure and results are discussed below. Extractive graph based Summarization is a 5 step process :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/566/1*dLt1BEeFl5UxJB6wRj8HEA.png" /></figure><ol><li>Text Processing.</li><li>Sentence Scoring.</li><li>Create intermediate representation of sentences in form of a graph.</li><li>Finding Sentence Rankings.</li><li>Generating Summary[5].</li></ol><h3>DATASET SELECTION</h3><p>In our proposed network, we would be working with the Amazon Product reviews dataset which has reviews of a total of 5000 customers for different 23 products. The customer reviews per product for 5 products is shown in Fig. below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/524/1*hucFSXXp4LI_ikC5m9P2gA.png" /><figcaption>Fig. 1: Reviews per product(Only 5 products)</figcaption></figure><h3><strong>TEXT PROCESSING AND REPRESENTATION</strong></h3><p><strong><em>Grouping Reviews</em></strong></p><p>The first step is to concatenate the reviews of all customers by grouping them according to the product they bought. These would give us a large text representation of reviews which can be used to generate the summary. We can use pandas group by and concatenate function to do so.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6N-3qn3zzyX9jXpGUg7wRg.png" /><figcaption>Fig. 2 : Snapshot of Concatenating reviews of same product.</figcaption></figure><p><strong><em>Sentence Splitting</em></strong></p><p>For the generated text , we split the text into sentences. Sentence splitting is done to store each of the review and indexing them so that after further processing and textrank, we can extract the original representation which is syntactically meaningful. Store the sentences in a csv format.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gEmj5zuo6iq0GxZJ-TF_Ww.png" /><figcaption>Fig. 3: Sentence Splitting</figcaption></figure><p><strong><em>Decontraction</em></strong></p><p>Reviews are usually written using short forms of the words or combining two words. For example, some consumers may write “The product hasn’t been upto my expectation” and others may write “The product has not been upto my expectation”. In order to avoid the same sentences being identified as different sentences, all the contracted words are converted into their expanded forms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/594/1*QrZo98nzNNg0ToRCvtjrtw.png" /><figcaption>Fig. 4: List of decontracted words</figcaption></figure><p><strong><em>Unwanted Words Removal</em></strong></p><p>The inclusion of punctuation marks, unwanted breaks, brackets and multiple spaces is common among a large set of reviews. These extra special characters can be due to human mistake or serve as a purpose to bolster the idea. They do not contribute anything to summary formation and therefore, are removed.</p><p><strong><em>Spelling Corrector</em></strong></p><p>Humans seem to make spelling mistakes often while writing anything, whether it be reviews of a large piece of text. The paper makes use of the Peter-Norvig’s spelling corrector logic.</p><p>The first step is to edit the word by removing a letter, replacing a letter, swapping letters or inserting a letter. We will create a list of all the edits. The probability of each word in the edited list is calculated as relative frequency to frequency of total words in a large document. From the probability values, we produce the candidates that are likely to replace the word in order of priority.</p><ol><li>The original word, if is known.</li><li>The list of words at distance one away.</li><li>The list of words at distance two away.</li><li>The original word, even though it is not known[6].</li></ol><h4><strong>TEXT CLEANING</strong></h4><p>The list of processes mentioned above are applied to the sentences stores in the csv format as seen in Fig. (3). The output is a cleaned and accurate representation of original sentences as shown below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/926/1*AMg2HgupRgj4OvZG-LbFmA.png" /><figcaption>Fig. 5: Sentences after processing</figcaption></figure><h3>SENTENCE SIMILARITY</h3><p><strong><em>Synsets</em></strong></p><p>Before calculating similarity of all sentences in the generated text, it becomes necessary to find the synonyms of the word. In different sentences, same description can be represented using different words. For example, “plant” can be named as “flora” or “plant_life”. To gain better results while generating the summary, we also have to consider the synonyms of words while calculating sentence similarity. These synonyms are called as synsets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1009/1*CjdjWRQSuzQf4HePTGjRzQ.png" /><figcaption>Fig. 6: Synset examples</figcaption></figure><p><strong><em>Similarity Calculation</em></strong></p><p>For calculating between two sentences s1 and s2, first generate individual list of all words in both sentences. Find the synsets of all words and add it to the two different lists synset1 and synset2. To calculate the similarity between every 2 sentences, the algorithm is mentioned below</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/569/1*gLvqtbndAX7z_cpCggVSbQ.png" /><figcaption>Fig. 7: Calculating Similarity</figcaption></figure><p>If we have n sentences, then the similarity matrix will have n x n dimensions. Store the similarity matrix in a .csv file.</p><h3>PageRank as TextRank</h3><p>In textrank, the vertices are the tokenized sentences. Weights are assigned to the edges connecting the node using similarity value. With arbitrary initial values, the iterative formula is run for each node until it converges below a specified threshold value[9]. The textrank is calculated considering the edge weight between two vertices and weights of all the edges between sentences and it’s connected nodes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/693/1*ShvsTwuPOpYD7Yu_zvbmnQ.png" /><figcaption>Fig. 8: PageRank representation.</figcaption></figure><p>The textrank method takes three input parameters, the <strong>similarity matrix</strong>, the value of <strong>epsilon</strong>, and the <strong>damping factor(d)</strong>. Each sentence is considered to be a dangling web page. Therefore, the probability of going from one sentence to any other sentence is equal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/538/1*cmpvzhZYm4pqXuNNkApTgw.png" /><figcaption>Fig. 9: Applying PageRank as TextRank</figcaption></figure><p>Damping factor — When visiting a web page, it contains some reference links. The probability that the person surfing the page will link on the reference links is equal to d. The probability that the user will stop at a particular webpage is (1-d). The same logic has been applied to sentences.</p><p><strong><em>Summary Generation</em></strong></p><p>The output of textrank algorithm is the ranking of the sentences. Depending upon our requirements we will select the first n sentences. We will get the original indices of the sentences and use the original sentences in our summary. Concatenating these sentences will generate the summary.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h8cwAhtDm9wiVIY3Uxdx3w.png" /><figcaption>Fig. 10: Summary generated for reviews as shown in Fig. 2</figcaption></figure><p>The GitHub implementation for the project is available at <a href="https://github.com/anuj200199/Autotext-Summarization-for-improving-customer-buying-decision">https://github.com/anuj200199/Autotext-Summarization-for-improving-customer-buying-decision</a></p><h3>REFERENCES</h3><ol><li>Changjian Fang, Dejun Mu, Zhenghong Deng, Zhiang Wu “Word-Sentence co ranking for automatic extractive text summarization”.</li><li>Khusboo S Thakkar, Dr. R. V. Dharaskar, M.B. Chandak , “Graph Based Algorithm For Text Summarization”, page 1–3.</li><li>Fadhlil Hadyan, Shaufiah and Moch. Arif Bijaksana “Comparison of Document Index Graph Using TextRank and HITS Weighting Method in Automatic Text Summarization.”</li><li>Albi Dode, Silvester Hasani “PageRank Algorithm”.</li><li>Jai Prakash Verma and Atul patel “Evaluation of Unsupervised Learning Based Extractive Text Summarization Technique For Large Scale Review and Feedback Data.”</li><li>Harshit Pande “Effective search space reduction for spell correction using character neural embeddings “</li><li>Rafael Ferreira, Gabriel Silva, Fred Freitas, Rinaldo Lima, Steven J. Simske, “Assessing sentence scoring techniques for extractive text summarization” from ELSEVIER.</li><li>Jai Prakash Verma “Spark Frameworks and Architecture” .</li><li>Federico Barrios, Luis Agrerich, Rosita Waxhenchauzer, “Variations of the Similarity Function of TextRank for Automated Summarization”.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=37e823419b5c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/improving-customer-buying-decision-using-extractivetext-summarization-textrank-algorithm-37e823419b5c">Improving Customer Buying Decision Using ExtractiveText Summarization(TextRank Algorithm)</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[License Plate Object Detection and Recognition using Deep Learning]]></title>
            <link>https://medium.com/analytics-vidhya/license-plate-object-detection-yolov3-and-recognition-pytesseract-d7628de1d110?source=rss-7fdca73aba87------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7628de1d110</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[license-plate-detection]]></category>
            <category><![CDATA[object-detection]]></category>
            <category><![CDATA[yolov3]]></category>
            <category><![CDATA[pytesseract]]></category>
            <dc:creator><![CDATA[Anuj Shah]]></dc:creator>
            <pubDate>Tue, 12 Nov 2019 05:13:56 GMT</pubDate>
            <atom:updated>2019-12-27T05:12:41.563Z</atom:updated>
            <content:encoded><![CDATA[<h4>Using Yolov3 and pytesseract for Humain 2019</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jL3abxoCaM2Y_ac4" /><figcaption>Photo by <a href="https://unsplash.com/@tomgruenbauer?utm_source=medium&amp;utm_medium=referral">Tom Grünbauer</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The huge blend of information developments, under different pieces of the cutting edge world, has incited the treatment of vehicles as calculated resources in information systems. Since a decision information structure has no significance without any data, there is a need to change vehicle information in the present world.</p><p>There are two approaches to make sense of this information, either with the help of human administrators or with the help of the latest technology that will help us to recognize evidence of vehicles by their license plates. Surveillance plays a vital role in almost every industry, whether it be construction sites, safety while driving for helmet detection, and also for the topic License Plate Detection which helps us to identify the unregistered plates.</p><p>Deep Learning has a promising future in the field of detection and identification through Computer Vision. Use of deep learning involves use of Convolutional Neural Network for image classification, Deep Neural Network, Recurrent Neural Network, etc. For the purpose of object detection, the possible approaches are R-CNN or Faster R-CNN and other pretrained models.</p><h3><strong>Scope of work</strong></h3><p>The project on License Plate Detection can help us identify violators of the traffic rules, especially at signals, exceeding a certain minimum speed near schools, etc. License Plate Detection in case of two wheelers, can be combined with helmet detection for possible drivers not wearing helmets while driving.However, the basic problem associated with Indian License Plate Detection problem is lack of good quality images taken from low resolution cameras.</p><p>Another problem is lack of defined or specific size of the number plates unlike foreign nations. While developing the solution of license plate detection, one of the aspects that came to my mind that a portion of the license plate, usually the portion above the character and digits, specifies some written information. For example, it specifies the designation of the person, in case of a higher ranking government official(see fig below) . We do not want to consider this portion and remove the unwanted written portion and retrieve only the numbers and characters on the license plate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/280/1*OivIIEZz8uZHkhu1aHbYYw.png" /><figcaption>NOTE: THE IMAGE USED IS SOLELY FOR EXPLANATION PURPOSE AND NOT INTENDED TO USE ANYONE PRIVATE LICENSE PLATE NUMBER OR DESIGNATION</figcaption></figure><p>Indians can also specify the digits on the license plates in any random shape instead of only machine generated digits. As an example, figure below specifies the number “214” as a sequence of characters specifying name of Lord Ram in Hindi.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/386/1*T-swkhJBdDFixRJ1OK_9yQ.png" /><figcaption>NOTE: THE IMAGE USED IS SOLELY FOR EXPLANATION PURPOSE AND NOT INTENDED TO USE ANYONE PRIVATE LICENSE PLATE NUMBER OR DESIGNATION</figcaption></figure><h3><strong>SOLUTION APPROACH</strong></h3><p><strong>ALGORITHM/LIBRARY USED:</strong></p><ol><li>Json File preprocessing: The dataset is present in a json file specifying the content , label, image width, image height, and x and y coordinates of bottom right and top left bounding box. We have to separate all the fields and store them in csv format. For this , I have used the pandas library in python.</li><li>For Data Augmentation: Darknet enables us to perform data augmentation by changing the saturation, hue, brightness of images inside the yolo-tiny-obj.cfg which contains the architecture of our model. For more information please refer [2]</li><li>For Object detection: The main part in the case of object detection is to decide which of the models to use. There is a wide pool of models available to us, with variations of each model . The broad category or the widely used can be divided into YOLO(You Only Look Once), RetinaNet and SSD5 models.</li></ol><p>From the figure below, the best models are YOLOv3 and RetinaNet.YOLOv3 has comparable Mean Average Precision(map) time with RetinaNet. Nevertheless, the inference time is smaller than RetinaNet. Hence, I have used the YOLOv3-tiny model for object detection.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/999/1*xV6b0Fl5q8TX5Lj5IJOlAg.png" /><figcaption>Comparison of various models with respect to inference time[4]</figcaption></figure><p>4. For Optical Character Recognition: After getting the required bounding box on the license plate, we have to generate the string containing characters and digits. The following task is done using the library pytesseract library. For documentation refer [4].</p><h3>IMPLEMENTATION DETAILS</h3><p><strong>PREPROCESSING STEPS</strong></p><p>The data in json file is converted into a .csv format with columns for content, label, given coordinates of bounding box, image height, image width, notes and extras. It is important to note that the coordinates are in normalised form. First all the details into content, notes, extras, top left coordinates, etc and stored in a csv file. For this, run the<strong><em> seperate.py</em></strong> file.</p><p>After getting all the details in csv file, we are scraping the image from the web link. The images are saved incrementally as image(i).jpg where i is the value of i from 0 to size of data-1 and also add another column, that is the name of each image to the csv file. Run <strong><em>image_save.py</em></strong></p><p>The top left and bottom right coordinates are stored in normalised form in the csv file. We convert them to their unnormalised form by multiplying the x coordinate with width and y coordinate with height of the image. These four columns are added to the dataset specifying the unnormalised coordinate values. This step is done as YOLO has a particular format for taking the input coordinates of the bounding box surronding the region of interest which is discussed in the next paragraph. Run <strong><em>bounding_box.py</em></strong>. To see the bounded box randomly on images, run <strong><em>draw_bounding_box.py</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/387/1*11yTd_fjqafMWFUNp45mmg.png" /><figcaption>Bounding box across license plates</figcaption></figure><p>In YOLO, we have to create a .txt file for each image and should have the same name as the image. For each image, the txt file would contain five values — the class label, which is 0 in our case, the absolute center coordinates, x_center and y_center, and the absolute width and height of the bounding box with respect to the shape of the whole image</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/429/1*Nt9CdHMZ8-GONv6Zr4MzGQ.png" /><figcaption>Orientation of annotated .txt file for YOLO</figcaption></figure><p>Run generating_annotations_test.py For example , bounding box of coordinates (582,274) and (700,321) with image width 806 and height 466 gets converted to :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/653/1*p3HyfM5NTqGGPd0PAGU1Gg.png" /></figure><h3>SETTING DARKNET YOLOv3:</h3><p>Clone the darknet YOLOv3 architecture from [2]. Execute the following commands in command prompt to setup darknet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/455/1*zhdPAp62ZgRNJypCGmNAbA.png" /><figcaption>Cloning Darknet from [2]</figcaption></figure><p>Copy images and their corresponding txt files in folder named obj under data. Create obj.names file specifying the class name. Under data folder, also create a obj.data file specifying the following:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/543/1*nCdlGtZEo_Nt0g0hjl5o1g.png" /><figcaption>Specifying details in obj.data file</figcaption></figure><p>Since we are dealing with only one class, we have to change the conf file specifying the YOLO architecture by setting filters at output YOLOv3 layer along with classes parameter. Set classes=1 and filters =(classes+5) *3 = 18 [2] in our case.</p><h3>TRAINING AND TESTING</h3><p>Separate the images into train and test sets in the format shown in fig. Run <strong><em>split_train_test.py</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/392/1*HOKGV7Xwmc2QnN5iTI2dqg.png" /><figcaption>train.txt file snapshot stored in data folder</figcaption></figure><p>Run the following command to train model</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/679/1*2d4O6kPXs0kl9oqBK25AQQ.png" /><figcaption>Training object detection model</figcaption></figure><p>For Linux or Mac, replace darknet.exe with ./darknet</p><p>To test on images and get the coordinates of the region of interest, run predict.py with command line arguments as image_name, architecture .cfg file, trained weight file and classes file(obj.names)</p><p><strong>Note : Give entire path to the weight and other file</strong>.</p><p><strong><em>predict.py — image path_to_img — config yolov3-tiny-obj.cfg — weights yolov3-tiny.conv.15 — classes data/obj.names</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/350/1*_C2GLnFeib3g-3V9J4eIVQ.jpeg" /><figcaption>Prediction of license plate.</figcaption></figure><p><strong>GETTING THE CHARACTERS(OCR):</strong></p><p>Once we know the value of the coordinated of the bounding box, crop the original image to only consist the license plate instead of the whole image using PIL library.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/116/1*ohaI5P7qZMbP9sSsw1PgGg.jpeg" /><figcaption>Cropped image capturing region of interest</figcaption></figure><p>Then use tesseract to generate text from image.</p><p><strong><em>python ocr.py trying.jpg</em></strong></p><p>trying.jpg is the name of the cropped image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/148/1*F_lxm3GtZaottRpcJMqpkw.png" /><figcaption>Output of the tesseract library</figcaption></figure><p>You can download the pretrained weights and necessary files at my github repository</p><p><a href="https://github.com/anuj200199/licenseplatedetection">https://github.com/anuj200199/licenseplatedetection</a></p><h3><strong>REFERENCES</strong></h3><p>[1] Object Detection Using Gradient-Based Learning by Yann Lecun, Patrick Haffner, Leon Bottou, Yoshua Bengio.</p><p>[2] Yolo-v3 and Yolo-v2 for Windows by AlexeyAB.</p><p>[3]YOLOv3 — You Only Look Once (Object Detection Improved YOLOv2, Comparable Performance with RetinaNet, 3.8× Faster! by Sik-Ho Tsang.</p><p>[4] Python Tesseract by Mathias A Lee.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7628de1d110" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/license-plate-object-detection-yolov3-and-recognition-pytesseract-d7628de1d110">License Plate Object Detection and Recognition using Deep Learning</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>