Sharad Joshi, Ph.D.

Current Position: Senior Researcher (Computer Vision and Deep learning), Hitachi Research and Development, Bengaluru

Ph.D. (28 Dec 2015 - 21 Dec 2020 ), Electrical Engineering, Indian Institute of Technology (IIT) Gandhinagar

Advisor: Prof. Nitin Khanna

Collaborators and Mentors:

Prof. Nasir Memon and Dr. Pawel Korus, New York University (Visiting Scholar-6 months project)
Dr. Sainarayanan Gopalakrishnan, Technology Director, HCL Technologies
Prof. Kishor Upla, S. V. National Institute of Technology (SVNIT), Surat (M.Tech. advisor)
Prof. Vikrant Jain, Indian Institute of Technology (IIT) Gandhinagar (Independent project advisor and Doctoral Studies Committee member)

My PhD at IIT was funded by Visvesvaraya PhD Scheme, Ministry of the Electronics & Information Technology (MeitY), Government of India. Prior to joining IITGN, I completed my M.Tech. in Communication Systems from National Institute of Technology (NIT), Surat and before that I worked at Infosys for about 20 months as a Systems Engineer.

Broad Research Areas: Digital Image Processing/ Computer Vision, Multimedia Forensics and Security, Remote Sensing.

Keywords: Image classification, object detection, semantic segmentation, low-level image processing, digital camera processing, data augmentation, transfer learning, designing neural network architecture, deep learning for computer vision,

Contact: sharad (dot) joshi (at) alumni (dot) iitgn (dot) ac (dot) in

Reviewer Work (Partial List)

Media Coverage

Dr Ramesh Pokhriyal Nishank on Twitter“Professor Nitin Khanna from @iitgn & his PhD student, Shri Sharad Joshi have #innovated a new Printer Specific Local Text Descriptor (PSLTD). This descriptor will be helpful in identifying the source printer of a particular text document.”

November 8, 2019: Indian Institute of Technology, Kanpur posted images on LinkedInwww.linkedin.com/posts/indian-institute-of-technology-kanpur_thatsiitk-c3icenter-cybersecurity-activity-6598519324003135489-CBY4/

Printer FingerprintingDid you know that it is possible for electronic printers to have unique fingerprints just like us??

IIT Gandhinagar research traces forged paper to printerUnder Dr Khanna’s guidance, PhD student Sharad Joshi has developed a system to track such forged documentsEvery country has its own unique problems, but one of the most common problems that every country shares is document fraud – a crime that causes severe money losses throughout the world. Realising the negative impact that forged documents can have nationally and internationally, a PhD student and a professor from IIT Gandhinagar have developed a system to trace the source of forged documents.The documents could range from forged mark sheets, property papers, rent agreements, insurance agreements, and even documents like experience letters. In separate incidents the past month alone, five people were caught by immigration officials at Ahmedabad airport when they tried to travel to a foreign country on forged educational certificates. Once developed, such a system could help officials track the forgers, and crack criminal rings.The research was conducted by doctoral student Sharad Joshi and his project investigator Dr Nitin Khanna, an assistant professor with the Electrical Engineering discipline, on 18 printers and 720 documents. They can study any normal paper and trace the laser printer or inkjet printer used. Currently, the research is limited to black-and-white documents. The project is partially supported by the Board of Research in Nuclear Sciences (BRNS) in the Department of Atomic Energy.Student Sharad Joshi said, “We are progressively building a robust system similar to biometric (fingerprint) identification for printers. Just like everyone’s fingerprint and iris is unique, similarly a printed page has unique characteristics/artefacts present. This helps in tracing it to the source printer, and matching the characteristics.“Our present system uses texture patterns present in each printed character. We are trying to create a printer signature to link a known printer to the documents printed from it.”Joshi adds, “We have developed a novel algorithm where a single-printer signature model can be generated using a specific font type. It can further be used to test any text document printed using the same or even some other font type. This is a milestone.“While our proposed system builds a single-printer signature for all types of letters, currently it only works when signatures of all possible printers in our database are available. Now, e will be able to identify documents on the basis of printer signature.”