About Me

I am a Research Scientist working at TECH5 USA. I was previously a graduate student at the Pattern Recognition and Information Processing Lab (PRIP) at Michigan State University, working under the supervision of Prof. Anil K. Jain.

My current research interests include the application of Machine Learning to Computer Vision and Biometrics.

Skills

Machine Learning
Python
Tensorflow
PyTorch
C++

Vishesh Mistry


vsm2209@gmail.com

Education

My academic journey so far!

Michigan State University, United States


MS in Computer Science

Session: 2019-2021

Guide: Prof. Anil K. Jain

CGPA: 3.90 (On the Scale of 4.00)

IIT Jodhpur, India


Bachelor of Technology in Computer Science and Engineering

Session: 2015-2019

CGPA: 8.99 (On the Scale of 10.00)

Atmiya Vidya Mandir, India


AISSCE (CBSE)

Session: 2014-2015

Percentage: 95.6%

Atmiya Vidya Mandir, India


AISSE (CBSE)

Session: 2012-2013

CGPA: 10.00 (On the Scale of 10.00)

Projects

A brief info regarding my projects

T5 Digital ID - Decentralized Verifiable Credential

TECH5, USA

  • Leading the research behind the T5 Digital ID, a secure decentralized offline verifiable credential in the form of a proprietary cryptograph
  • Optimized the T5 Face SDK to facilitate the use of 136 bytes lightweight face templates for real-time verification on smartphones
  • Managing the research team responsible for the AI SDK - detection and processing of the cryptograph before decoding
  • Playing an important role in the integration of this technology into the T5 Idencode platform for digital ID issuance and verification

AI SDK for Remote Medical Injector - PreciHealth

TECH5, USA

  • Developing a comprehensive AI SDK for the use of a remote medical injector through any device with an integrated camera system
  • Led research activities involving data collection, network selection, model training/testing, and deployment
  • The SDK includes real-time device detection and tracking, along with live feedback based on the physical state of the device

Adversarial Fingerprints Synthesis

PRIP Lab, MSU

  • Developed a black-box adversarial fingerprints synthesis method that automatically generates adversarial fingerprints capable of evading model-based and learning-based fingerprint matchers while preserving fingerprint attributes
  • The proposed method drops the TAR of three fingerprint matchers on three datasets from 94.15% to as low as 5.52% (using Innovatrics SDK on FVC 2004 DB1-A at 0.01% FAR), outperforming other white-box state-of-the-art adversarial attack methods

Synthesis Of Realistic Weather Specific Images

IIT Jodhpur, India | IBM Research Labs, New Delhi

  • Constructed a pipeline to convert images taken in summer season to images of winter season taking into consideration fine level features like clothing, etc
  • Employed scene transfer using GANs, followed by person identification and segmentation, followed by cloth changing, and finally warping back the person onto the image
  • Extended the functionalities of the pipeline by incorporating the transfer of clothes to Indian ethnic wear

T5 AirSnap Face - Contactless Capture for Face

TECH5, USA

  • Extended T5 AirSnapFace, contactless face capture technology using smartphones and browsers, by developing and implementing algorithms for ICAO compliance and international standards conformance
  • Improved the then implementation by making the capture faster, fluid, and robust across multiple devices
  • Played an integral role in deploying the technology in the Digital Passenger Declaration (DPD) platform of Australia's Department of Home Affairs

Towards 100 Million Synthetic Fingerprints

PRIP Lab, MSU

  • Implemented an identity loss while training the I-WGAN which guides the generator to synthesize a diverse set offingerprints corresponding to more distinct identities
  • Synthesized fingerprint images are closer to real fingerprints in terms of (i) fingerprint quality and (ii) fingerprint uniqueness

Deep Transfer Learning For The Detection Of Radical Groups' Iconography In Real World Images With Only One Reference Image

BISITE Research Lab, Spain

  • Trained several deep CNNs such as SSD, YOLO, DeepRanking, etc. to detect the presence of radical groups' logos in real-life images. Compared their performance to that of traditional feature descriptors such as SIFT, SURF, rSIFT, AKAZE, ORB, etc
  • Created an artificial dataset from a single radical group logo using various transformations and observed how each degree of transformation affected the performance of the entire network

Contact Lens Detection And Classification Using A Hierarchically Tuned Convolutional Neural Network

IIT Mandi, India

  • Used Keras to develop a hierarchically tuned two-model deep convolutional neural network for detection and classification of contact lens from iris images, outperforming the then existing state-of-the-art techniques
  • The highly generalised network gave remarkable results without using any pre-processing or iris segmentation

Publications/Patents

System and Method for Credential Verification and Associated Identity Binding using Biometrics in a Decentralised Manner

US Patent No. 63/457,306 (Pending), Apr 2023

System and Method for asserting document authenticity based on biometric verification of the photo and 2D code as a security feature

US Patent No. 63/452,057 (Pending), Mar 2023

AdvBiom: Adversarial Attacks on Biometric Matchers

Debayan Deb, Vishesh Mistry, Rahul Parthe

ArXiV Preprint, Jan 2023

ArXiV DOI: 10.48550/arXiv.2301.03966

Contactless Fingerprint Capture using Artificial Intelligence and Image Processing on Integrated Camera Systems

US Patent No. 17/967563 (Pending), Oct 2022

Fingerprint Synthesis: Search with 100 Million Prints

Vishesh Mistry, Joshua J Engelsma, Anil K Jain

IEEE International Joint Conference on Biometrics (IJCB), 2020

IEEE Xplore DOI: 10.1109/IJCB48548.2020.9304885

BRIDGE: Building Plan Repository for Image Description Generation, and Evaluation

Shreya Goyal, Vishesh Mistry, Chiranjoy Chattopadhyay, Gaurav Bhatnagar

IEEE International Conference on Document Analysis and Recognition (ICDAR), 2019

IEEE Xplore DOI: 10.1109/ICDAR.2019.00174

GHCLNet: A Generalized Hierarchically tuned Contact Lens detection Network

Avantika Singh, Vishesh Mistry, Dhananjay Yadav, Aditya Nigam

IEEE Conference on Identity, Security and Behavior Analysis (ISBA), 2018

IEEE Xplore DOI: 10.1109/ISBA.2018.8311471

Other Interests and Hobbies

Table-Tennis

I am an avid table-tennis player and have represented my college and school at various national-level tournaments

Secured 3rd position at the Inter-IIT Sports Meet 2016

iOS App Development

I have tried my hands at developing a few apps on the iOS platform using Swift

The most recent one was developed during Microsoft Code.Fun.Do Hackathon (2018), which checked if a contact in any user’s contact list had updated any of his/her data and automatically merged the update in the user's device