<?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 Harshita Kukreja on Medium]]></title>
        <description><![CDATA[Stories by Harshita Kukreja on Medium]]></description>
        <link>https://medium.com/@harshitakukreja?source=rss-dd179ae4f463------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*HUZNXBlYJkIxRRlF.</url>
            <title>Stories by Harshita Kukreja on Medium</title>
            <link>https://medium.com/@harshitakukreja?source=rss-dd179ae4f463------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 23:39:14 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@harshitakukreja/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[Biometric Privacy Protection in Facial Features using Deep Learning Systems]]></title>
            <link>https://medium.com/@harshitakukreja/biometric-privacy-protection-in-facial-features-using-deep-learning-systems-ea28a178fc1?source=rss-dd179ae4f463------2</link>
            <guid isPermaLink="false">https://medium.com/p/ea28a178fc1</guid>
            <dc:creator><![CDATA[Harshita Kukreja]]></dc:creator>
            <pubDate>Tue, 07 Dec 2021 00:51:02 GMT</pubDate>
            <atom:updated>2021-12-13T19:27:15.944Z</atom:updated>
            <content:encoded><![CDATA[<h4>Learning about protecting the privacy of facial biometric templates with the help of deep learning systems and comparing different techniques for the same</h4><h3>Introduction</h3><p>Every human being has certain biological and behavioral characteristics that are unique to that particular person only. Take, for example, fingerprints. Every person has a unique set of fingerprints that can be used to identify the said person. Such biological and behavioral characteristics which help to verify the identity of a person are known as biometric traits. Some of these traits are the iris, voice, signature, face, palmprint, and gait among others.</p><p>When humans are recognized automatically with the help of these traits, the information system is referred to as Biometric Recognition. In this system, the traits cannot be directly stored; they need to be mapped with the identity of the person involved. That is, if we are dealing with the face of a person, we need to know the name of the person to whom that face belongs as well. Hence, the traits are stored in the form of biometric templates.</p><p>Out of all available traits, facial biometric recognition is widely used due to the several features that can be worked upon like the shape of the face, the color of the iris, and different facial expressions. However, since the digital biometric templates deal with the identity of a person, they can be easily breached and the identity of the person can be compromised. Overcoming this issue also proves to be very tedious when there is a need to reissue another face template for the concerned user.</p><p>Therefore, having a secure biometric system for face templates is indispensable. This study provides an insight into the use of deep learning systems for facial biometrics by comparing the related works over the years by a few authors.</p><h3>Background</h3><p>Biometric template protection can be classified into four broad categories :</p><ol><li>Biometric Cryptosystems</li><li>Transform-based Biometrics</li><li>Hybrid System</li><li>Homomorphic Encryption</li></ol><p>The focus of this article is non-invertible transformation-based biometric systems. Ratha et al. <a href="https://ieeexplore.ieee.org/document/5386935">[1]</a> were the first to use cancelable non-invertible transformations for biometric template protection by proposing Cartesian, Functional, and Polar transforms for fingerprint templates <a href="https://dl.acm.org/doi/10.1109/TPAMI.2007.1004">[2]</a>. Cancelable biometrics, in this case, refers to the irreversibly transformed information which helps us protect the original biometric template. This is very helpful if there is ever a breach of security and the template is leaked or damaged as the cancelable template can be easily replaced with another one. Thus the user-specific sensitive information is always protected. A cancelable biometric template must have the following four properties:</p><ol><li>Unlinkability: Cancelable features used in different platforms should be different for each user to protect the system from vulnerability.</li><li>Revocability: If the biometric is compromised, it should be easy to generate a new template from the raw data.</li><li>Non-invertibility: Retrieving original biometric data with the help of a cancelable template should not be possible.</li><li>Performance: The trade-off between accuracy and security should not be present.</li></ol><p>Over the years, non-invertible cancelable biometrics have made use of robust hash functions using one-to-many transformations, Gaussian and Uniform random projections with knowledge factors, visual-thermal fused secure systems, non-linear multidimensional spectral hashing, and encryption and decryption of biometric feature templates. The table below represents some of the related studies on the same.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zYzhshKDHNH9f5sXV8aAYA.png" /><figcaption>Table 1: Studies related to Non-Invertible Cancelable Approaches for protecting biometric templates</figcaption></figure><h3>Insight into some Existing Works</h3><p>This article’s focus is the work done by Arora <a href="https://link.springer.com/chapter/10.1007/978-981-16-4149-7_3">[10]</a>, Abdellatef <a href="https://link.springer.com/article/10.1007/s00371-019-01715-5">[11]</a>, and Talreja <a href="https://arxiv.org/abs/1708.02314">[12]</a> on the privacy protection of facial biometric traits with the help of deep learning systems. This section discusses the methodology used by the three authors, the datasets they have used for the same, the performance metrics being targeted by them, and the results achieved. The section is then concluded with a comparison between the works detailing the open challenges and the future directions possible for biometric template protection with respect to deep learning systems.</p><h3>Deep Learning and Hashing</h3><h4>Methodology</h4><p>Arora et al. <a href="https://link.springer.com/chapter/10.1007/978-981-16-4149-7_3">[10]</a> proposed a one-shot enrolment method to map the images of people’s faces against the unique binary codes that are being assigned to them. They do so by binarizing the pixels of the input image and normalizing it to 0 or 1. Further, the SHA-256 algorithm is being utilized to generate hash values for every person available on the dataset as it reduces the length of the binary code generated in the previous step. This compressed hash code is inherently generated for each user, making it unique to every one of them. The figure below describes the flow of events for the methodology being employed by the authors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/968/1*a4HZcQjSVndn33inczwZyg.png" /><figcaption>Figure 1: Framework for preserving the privacy of face biometric templates</figcaption></figure><p>The facial images are then augmented and passed along with the hashed values as labels for feature extraction. The proposed deep learning architecture has been provided below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UDrBcg5TKsScKFIz9MFCFg.png" /><figcaption>Figure 2: Proposed Deep Learning Architecture</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/632/1*wO9zrS9PskUtNWpBdgN7cQ.png" /><figcaption>Table 2: Detailed structure of the Deep Learning architecture</figcaption></figure><h4>Datasets Used</h4><p>The proposed architecture has been trained and tested on two publicly available datasets, namely, Labeled Faces in the Wild and Casia Webface. LFW is designed for unconstrained face recognition and consists of more than 13000 images belonging to approximately 1680 distinct people. Casia Webface consists of 494K images of faces of more than 10,000 people and is also widely used for face verification tasks.</p><h4>Performance Metrics and Results</h4><p>The model DeepCrypt proposed has been tested with different splits, with the best results being achieved by the 60–40 split for training and testing data. The targeted metrics are False Acceptance Rate and Genuine Acceptance Rate. The table below summarizes the results achieved by the authors for different types of optimization algorithms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/646/1*KzQgTBITjCc1aup0pMTGEw.png" /><figcaption>Table 3: Results achieved for different optimizers by Arora et al.</figcaption></figure><h3>Deep Learning and Bio-convolving Encryption</h3><h4>Methodology</h4><p>Abdellatef et al. <a href="https://link.springer.com/article/10.1007/s00371-019-01715-5">[11]</a> proposed a cancelable method for uni-biometric and multi-biometric recognition of faces that exploits depth concatenation in collaboration with batch normalization. The authors deploy bio-convolving along with deep learning to extract features from the different facial regions. The proposed method is divided into four prominent stages, namely, detection, extraction, fusion, and bio-convolving. The figure below shows the detailed framework for the face recognition method.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/841/1*B7sVQodlv4uqpzSjLR0rOw.png" /><figcaption>Figure 3: Proposed Framework for Face Recognition</figcaption></figure><p>The first stage is detection where the original face images from all the datasets are accessed and the visible changes are made a note of. These include but are not limited to laughing, frowning, spectacles on eyes, and speaking. The work then uses convolutional neural networks on the facial regions that were detected to extract deep features. The table below summarizes the multiple CNN architecture used by the authors. Further, local and fusion layers have been implemented on the extracted features and the result is a more reliable descriptor of facial features. Finally, the transformed feature is convolved with a random kernel to complete the task of bioconvolving.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/517/1*qQXrClAFUC2509Q3gxAFPQ.png" /><figcaption>Table 4: Detailed structure of the Deep learning architecture</figcaption></figure><h4>Datasets Used</h4><p>The proposed architecture has been trained and tested on three publicly available datasets, namely, Facial Recognition Technology, Labeled Faces in the Wild, and Point and Shoot Face Recognition Challenge. The Facial Recognition Technology (FERET) database was created in the 1990s and consists of 14,126 images belonging to 1199 individuals along with certain duplicate images. LFW, as before, is used for its special design for unconstrained face recognition. Point and Shoot Face Recognition Challenge (PaSC) includes data belonging to 293 distinct people, containing 9376 still images and 2802 videos.</p><h4>Performance Metrics and Results</h4><p>Apart from the accuracy of the recognized images, the authors also calculate the precision, recall, F1-score, and the specificity of the deep learning model used. The training was done only for five epochs and stochastic gradient descent optimizer was used instead of the adam optimizer. To get comparative results with other works, evaluation is done first on the unimodal and multi-biometric face recognition approaches and then on the bioconvolving method proposed in the study.</p><p>The architecture detects four major features of the face: the eyes, nose, mouth, and the whole face. For each of these traits, a simple Convolutional Neural Network is trained and achieves results that are better than the state-of-the-art CNNs. The authors then proceed to fuse the four traits into a single model to get the multi-biometric model architecture and compare the results with the uni-biometric model.</p><p>To make the system even more secure, the proposed approach is encrypted using bioconvolution. There is little to no trade-off in accuracy and security that can be observed and privacy of the users data is preserved using encryption and decryption algorithms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/608/1*LzN98THXKJY3ZXxH8uJ-hA.png" /><figcaption>Table 5: Results achieved for different face recognition approaches by Abdellatef et al.</figcaption></figure><h3>Deep Learning and Error-correcting Code</h3><h4>Methodology</h4><p>Talreja et al. <a href="https://arxiv.org/abs/1708.02314">[12]</a> presented a combination of deep learning systems and error correction codes in their study to tackle the security and privacy protection of biometric templates. Deep neural networks were being implemented in the secure multibiometric system and the work was done using cancelable biometric templates. The two template blocks were cancelable and secure sketches. Cancelable template block (CTB) took into account the deep neural networks and handled feature extraction and fusion of the models, apart from the binarizing phase. The authors used VGG-19 that had been pre-trained on ImageNet to finetune their results. The proposed approach fused two CNNs — iris and the face, at the representation level using bilinear architecture (BLA) and the fully connected architecture (FCA). This resulting multimodal feature vector was then finally encoded in binary form and was further given as an input to the secure sketch template block (SSTB). The framework used Reed-Solomon (RS) decoder for forward error correction to decode the noisy codeword in the encrypted binarized results. This helped in managing the intra-class variability of the joint template and generating the template for the enrolled multibiometric sketch. Access was only granted if the hash of the generated sketch matched a threshold for the enrolled sketch. The figure below depicts the flow diagram of the proposed architecture.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/737/1*Uz0V4otI7irsPIGq0uSe1g.png" /><figcaption>Figure 4: Framework for preserving the privacy of face and iris biometric templates</figcaption></figure><h4>Datasets Used</h4><p>The proposed approach has been divided into two parts since it is a multi-biometric system, Face-CNN and Iris-CNN, so the datasets have been used accordingly. The first part uses the Casia Webface database, which has already been discussed in the above section. The second part uses a combination of Casia-Iris-Thousand and ND-Iris-0405 databases. The Casia-Iris-Thousand database consists of 22,035 images of the iris from over 700 distinct subjects and the ND-Iris-0405 database contains 64,980 images corresponding to 356 unique subjects, and 712 unique irises. Both the parts are then fine-tuned on the WVU-Multimodal database which has information on around 70 subjects each for iris and face.</p><h4>Performance Metrics and Results</h4><p>False Accept Rate (FAR) is the threshold value at which an unauthorized user can get access to the system while Genuine Accept Rate (GAR) is the threshold value at which an authorized user can get access to the system. These two metrics have been used by the authors for their proposed work. The trade-off between the performance of the recognition and the security of the presented approach has been evaluated by the GAR-Security curves for the two approaches, FCA and BLA. The table below summarizes the results of these curves for the two approaches, and it was noted that BLA provides better recognition performance than FCA. On analyzing the security aspect of the framework, it was found that the generated secure sketch of the biometric template was revocable and non-invertible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/712/1*MMD-0lcp4FpjqtB52P-GAw.png" /><figcaption>Table 6: Results achieved for different face and iris recognition approaches by Valreja et al.</figcaption></figure><h3>Comparing the Works</h3><p>The main aim of biometric systems is to ensure the privacy of the user who has to be verified. This is done by generating cancelable templates from the original biometric traits and features so as not to actually use the original traits. The above three works make use of cancelable templates with three different approaches involving deep learning, namely hashing, bio-convolving, and error-correcting. Using deep convolutional neural networks independently often does not give satisfactory results and the system ends up compromising the privacy of the enrolled users. Therefore, it is imperative that the deep CNN is assisted with some technique to achieve results comparable to the state-of-the-art methods. From the above three works, it can be gathered that secure transformed templates can be generated using hashing <a href="https://link.springer.com/chapter/10.1007/978-981-16-4149-7_3">[10]</a> when we only have the facial images of the enrolled individuals and desire to implement an unsophisticated biometric system. This can be achieved without degrading the performance of the authentication system when we are only working with one modality and the resulting system is the least secure out of the three approaches even though the genuine accept rates approximate around 99%. In case we need to make the system more secure, we need to introduce another modality along with the facial images. The other two approaches do precisely that. Bio-convolving the resulting fused features from the deep CNN <a href="https://link.springer.com/article/10.1007/s00371-019-01715-5">[11]</a> helps in achieving better recognition accuracy across different datasets. Implementing a multimodal biometric system instead of a unimodal system ensures that the feature fusion is more representative and well-grounded and gives a thorough output. To further maintain the security and robustness of the system, hybridizing cancelable biometric techniques with secure sketch schemes is also a good approach. Secure sketches have varying symbol sizes and the proposed architecture in <a href="https://arxiv.org/abs/1708.02314">[12]</a> gives the best results with symbol size 6. The framework has genuine accept rates as high as 99.7% which is infeasible to crack with plain old brute force attack.</p><h3>Open Challenges</h3><p>Biometric systems for security using facial images have become predominant in today’s age. This also means that the system needs to be robust, otherwise the leaked templates can compromise the user’s identity. Certain challenges with cancelable biometrics have been discussed further.</p><p>Cancelable biometric templates make use of the transformed templates that are generated from the original biometric. Therefore, the recognition performance of the template depends on its alignment as it focuses on the distinct features of the biometric traits. Since feature representation is another key aspect of the facial recognition system, this may prove to be a challenge in biometric template security. The article has also discussed approaches where multiple modalities have been used using feature-level fusion. This usually leads to a trade-off between the security and accuracy of the system. It is essential that the system has little to no trade-off between the two for better results, even if the number of modalities is more than two.</p><h3>Future Work</h3><p>Apart from working only on the face templates or some of the common biometric traits like fingerprints and iris, the system can be extended to other physiological traits like handwriting and the gait of the person. Cancelable biometric transforms are usually a part of non-invertible transformation approaches. This can be implemented along with biometric cryptosystems, creating a hybrid between the two approaches. Jami et al. <a href="https://ieeexplore.ieee.org/document/8661905">[13]</a> enforced the two approaches using adversarial learning to result in better matching performance while addressing the re-enrolment issue as well. Unlinkability and diversity have been tackled by Sadhya et al. <a href="https://www.semanticscholar.org/paper/Design-of-a-cancelable-biometric-template-scheme-on-Sadhya-Singh/638e4e0d3d67ee318dfc15525f4f3a389e1dd375">[14]</a> when they combined cancelable templates with a cryptographic hash. Security of the cancelable biometric templates is a very vast field and is yet to be fully explored.</p><h3>Citations</h3><p><a href="https://ieeexplore.ieee.org/document/5386935">[1] N. Ratha, J. Connell, and R. Bolle, “Enhancing security and privacy in biometrics-based authentication systems”, IBM Systems Journal, vol. 40, no. 3, pp. 614–634, 2001.</a></p><p><a href="https://dl.acm.org/doi/10.1109/TPAMI.2007.1004">[2] N. Ratha, S. Chikkerur, J. Connell and R. Bolle, “Generating Cancelable Fingerprint Templates”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 29, no. 4, pp. 561–572, 2007.</a></p><p><a href="https://ailb-web.ing.unimore.it/icpr/media/slides/10909.pdf">[3] X. Dong, K. Wong, Z. Jin, and J. Dugelay, “A Secure Visual-thermal Fused Face Recognition System Based on Non-Linear Hashing”, IEEE 21st International Workshop on Multimedia Signal Processing (MMSP), 2019.</a></p><p><a href="https://www.researchgate.net/publication/333919054_A_Cancellable_Face_Template_Scheme_Based_on_Nonlinear_Multi-Dimension_Spectral_Hashing">[4] X. Dong, K. Wong, Z. Jin, and J. Dugelay, “A Cancellable Face Template Scheme Based on Nonlinear Multi-Dimension Spectral Hashing”, 7th International Workshop on Biometrics and Forensics (IWBF), 2019.</a></p><p><a href="https://www.semanticscholar.org/paper/Biometric-Template-Protection-in-the-Image-Domain-Debiasi-Kirchgasser/7e0e9987d96c9bd2244e1b1897cf7581417cea58">[5] L. Debiasi, S. Kirchgasser, B. Prommegger, A. Uhl, A. Grudzien, and M. Kowalski, “Biometric Template Protection in the Image Domain Using Non-invertible Grey-scale Transforms”, IEEE International Workshop on Information Forensics and Security (WIFS), 2019.</a></p><p><a href="https://ieeexplore.ieee.org/document/8410587">[6] H. Kaur and P. Khanna, “Random Distance Method for Generating Unimodal and Multimodal Cancelable Biometric Features”, IEEE Transactions on Information Forensics and Security, vol. 14, no. 3, pp. 709–719, 2019.</a></p><p><a href="https://link.springer.com/article/10.1007/s11042-020-08734-8">[7] H. Kaur and P. Khanna, “PolyCodes: generating cancelable biometric features using polynomial transformation”, Multimedia Tools and Applications, vol. 79, no. 29–30, pp. 20729–20752, 2020.</a></p><p><a href="https://www.sciencedirect.com/science/article/abs/pii/S0167739X18330553">[8] H. Kaur and P. Khanna, “Privacy-preserving remote multi-server biometric authentication using cancelable biometrics and secret sharing”, Future Generation Computer Systems, vol. 102, pp. 30–41, 2020.</a></p><p><a href="https://ieeexplore.ieee.org/document/9107244">[9] A. Sardar, S. Umer, C. Pero, and M. Nappi, “A Novel Cancelable FaceHashing Technique Based on Non-Invertible Transformation With Encryption and Decryption Template”, IEEE Access, vol. 8, pp. 105263–105277, 2020.</a></p><p><a href="https://link.springer.com/chapter/10.1007/978-981-16-4149-7_3">[10] S. Arora, M. Bhatia, H. Kukreja and S. Jain, “Privacy Protection of Biometric Templates Using Deep Learning”, Innovations in Cyber-Physical Systems, pp. 19–27, 2021.</a></p><p><a href="https://link.springer.com/article/10.1007/s00371-019-01715-5">[11] E. Abdellatef, N. A. Ismail, S. E. Abd Elrahman, K. N. Ismail, M. Rihan, and F. E. Abd El-Samie, “Cancelable Multi-Biometric Recognition System Based On Deep Learning”, The Visual Computer 36 (6): 1097–1109, 2019.</a></p><p><a href="https://arxiv.org/abs/1708.02314">[12] Talreja, M. Valenti and N. Nasrabadi, “Multibiometric secure system based on deep learning”, IEEE Global Conference on Signal and Information Processing (GlobalSIP), 2017.</a></p><p><a href="https://ieeexplore.ieee.org/document/8661905">[13] S. Jami, S. Chalamala and A. Jindal, “Biometric Template Protection Through Adversarial Learning”, IEEE International Conference on Consumer Electronics (ICCE), 2019.</a></p><p><a href="https://www.semanticscholar.org/paper/Design-of-a-cancelable-biometric-template-scheme-on-Sadhya-Singh/638e4e0d3d67ee318dfc15525f4f3a389e1dd375">[14] D. Sadhya and S. Singh, “Design of a cancelable biometric template protection scheme for fingerprints based on cryptographic hash functions”, Multimedia Tools and Applications, vol. 77, no. 12, pp. 15113–15137, 2017.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ea28a178fc1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Voice]]></title>
            <link>https://medium.com/@harshitakukreja/the-voice-c5fafc5330d1?source=rss-dd179ae4f463------2</link>
            <guid isPermaLink="false">https://medium.com/p/c5fafc5330d1</guid>
            <category><![CDATA[inner-peace]]></category>
            <category><![CDATA[mindfulness]]></category>
            <category><![CDATA[voices]]></category>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[writing]]></category>
            <dc:creator><![CDATA[Harshita Kukreja]]></dc:creator>
            <pubDate>Fri, 18 Oct 2019 19:38:33 GMT</pubDate>
            <atom:updated>2020-07-02T17:14:39.747Z</atom:updated>
            <content:encoded><![CDATA[<p>It is one of those times. The time when everything is just too much. You feel the walls are closing in on you, even though you are sitting in a spacious room, with no one around you.</p><p>One moment, you feel that you are being coddled by your own thoughts about the ever uncertain future. You try not to think about it, try to calm yourself down. But your mind won’t listen to your voice of reason. It is on its own track, one of revenge upon the only enemy it has in this whole wide world — you.</p><p>The next moment it hits you suddenly that you are not only alone in this room, but in your life too. People are never what they seem. Every day you see a new version of people who you call friends; a version which is deteriorating so much that you would rather not see it.</p><p>This loneliness can be peaceful many times, but today, it is being a thorn in your otherwise smooth going life. Well, not really smooth going, per se, considering all those times you have felt just what you are feeling right now. But you get the drift of these thoughts, or at least try to. After all, they are your very own thoughts.</p><p>Your mind only has one goal now, to calm itself, anyhow. You listen to some songs, take a few deep breaths, or even write a few words. But even after you have stopped or managed the surge of emotions going through you, you are not sure about what to do now. All these thoughts just keep on coming.</p><p>I would’ve helped you with my experience in this, but unfortunately for both of us, I’m still dealing with it. This is me, writing in an attempt to relax my nerves, hoping that everything is going to be alright. And the most amusing part is, I know it. I know that everything is going to work itself out, but I still don’t know what to do with all this energy surging through me right now. Guess I’ll just try to sleep it off, like those human remotes, and hope that I’d be as good as new when I wake up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oMBwXDzr1MgsLHCP0aqpOQ.jpeg" /><figcaption>Picture Courtesy: <a href="https://www.instagram.com/tkaranphotography/">https://www.instagram.com/tkaranphotography/</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c5fafc5330d1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Perfectly Imperfect]]></title>
            <link>https://medium.com/@harshitakukreja/perfectly-imperfect-bd08ed514423?source=rss-dd179ae4f463------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd08ed514423</guid>
            <category><![CDATA[musings]]></category>
            <category><![CDATA[imperfection]]></category>
            <category><![CDATA[perfection]]></category>
            <category><![CDATA[motivation]]></category>
            <category><![CDATA[mental-health]]></category>
            <dc:creator><![CDATA[Harshita Kukreja]]></dc:creator>
            <pubDate>Tue, 16 Jan 2018 16:49:44 GMT</pubDate>
            <atom:updated>2019-10-15T12:21:37.797Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/440/0*xoDPjbvQk1HNLxSI" /></figure><p>Growing up, all we are taught about is to be perfect. This notion of perfection includes excelling at everything under the sun and be the best that is out there. It doesn’t matter if you are competing with one other person or thousands of them. You have to beat them to be perfect.</p><p>Students and children are often taught that they would be perfect only if they excel at academics, art, dance, music and sports (pardon me if I forgot any other category). And God forbid you are average at even one of them, you would not be deemed perfect anymore.</p><p>To me, this whole idea is ridiculous. How can a human being, an imperfection created by God, be perfect in every way? We are after all just another species on this earth and not some kind of robots, though they too are not exactly what we call perfect. We are all ‘programmed’ to make mistakes. And we will make them.</p><p>If these mistakes or shortcoming make us any less perfect, so be it. Because I feel that imperfections are the only way to life. If you consider this for instance, a perfect man could never be happy. After mastering one aspect, he will still feel empty, like his life is still not perfect because there exists this particular thing which he has not excelled.</p><p>Only an imperfect person would have the passion and courage to take risks and actually do things in his life. Why, you ask? Because he has nothing to lose. He is already not perfect for the world, why not go another mile and do something new. The most that could happen is failure and he is not afraid of it. Maybe what he does will give some meaning to his life. At the end of the day, you could be <strong>perfectly imperfect</strong> for yourself and that’s what matters.</p><p><em>Originally published at </em><a href="https://harshitathinks.wordpress.com/2018/01/16/perfectly-imperfect/"><em>http://harshitathinks.wordpress.com</em></a><em> on January 16, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd08ed514423" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>