Computer vision and machine learning researcher with seven years of hands-on experience in design and development of applications and prototypes.
- Interests: Image classification, Action recognition in videos, Video segmentation and summarization, Hand segmentation in egocentric videos/images,Gesture recognition, Quality estimation of camera captured images, Large-scale machine learning
- Over 20 Conference papers published/presented (in CVPR, ICDAR, ICPR, DAS, DRR, CBDAR, PRL)
- Contributed to two public datasets (1) Mobile image quality dataset (2) Egocentric multi-step procedure dataset (hand segmentation, action recognition)
- Reviewer and PC member: T-PAMI, T-IP, ICPR, CVIU, ICDAR, ICIP,
- 13 filed patents in the area of image and video capture/processing
Software development:
- Vision apps on Android using OpenCV
- Classification, clustering, Image quality, mosaicking (Visual Studio C++ and Linux (gcc) + OpenCV)
- Hands-on with Caffe, Keras, Torch for deep learning
Member of Research Staff @ -Developed deep learning methods for counting passengers in HOV/HOT lane vehicles (Xerox Vehicle Passenger Detection System (VPDS)).
-Developed novel methods for egocentric video analysis (hand segmentation, summarization, action classification, hand gesture recognition) and multi-modal sensor fusion.
-Developed advanced video analysis features for Xerox Safe Courier app including flash/no-flash fusion, image quality check, video-based auto capture
-PARC liaison for collaboration on egocentric vision and sensing with Georgia Tech.
-Collaborator on university affairs committee project on computational photography with UMD. Developed low light mosaicking solutions.
-Co-authored 11 patent applications and 3 conference papers From October 2013 to Present (2 years 1 month) WebsterGraduate Research Assistant @ - Developed novel methods for data selection of large data sets for SVM training [ICDAR11]
- Developed an affinity propagation based method for text-line segmentation in handwritten document images [published at DAS10, ICDAR11]
- Designed and developed efficient methods for No-reference Image Quality Assessment [CVPR12, CVPR13, CVPR14]
- Proposed a multi-instance learning method for classification and localization of signatures and text in document images [ICDAR11]
- Proposed and implemented (C++) a method for structure based retrieval/classification of document images [ICPR12] From August 2008 to September 2013 (5 years 2 months) College Park, MDResearch Intern @ -Developed an end-to-end prototype demonstrating the ability to extract high quality images from a mobile video capture of a multi-page document.
-Conducted user study to evaluate user preference and difficulty level for such an application.
-Developed and demonstrated a novel approach for separating overlapping handwritten and machine-printed text in document images. From May 2012 to August 2012 (4 months) Webster, NYResearch Intern @ Developed an efficient method for effectively estimating the sharpness/ blurriness of document and scene images. The proposed method can be used to compute the sharpness in scenarios where images are blurred due to camera-motion (or hand-shake), de-focus, or inherent properties of the imaging system. I implemented the method in C++/Java on Android platform. From May 2011 to August 2011 (4 months) Palo Alto, CAImage Processing Scientist (Intern) @ Designed and implemented a shape codebook based approach for discriminating handwritten and printed text regions in Arabic document images (DRR 2011). From June 2010 to August 2010 (3 months) Boston, MAResearch Assistant @ Project - Online Handwriting Recognition of Indian Languages
(Funded by Ministry of Information Technology, India)
I designed and implemented a tool using OpenCV for online handwritten data collection and automatic segmentation of handwritten data.
Project : Field Extraction from Document Images
(Funded by First Indian Corporation (FIC), Bangalore, India)
I implemented modules in C for automatic field identification and extraction in document images.
While working on these projects I proposed a novel skew detection and correction method for complex camera-based images (published in TENCON 07). I also worked on a novel method for binarization of camera-based images (published in CBDAR 07). From January 2007 to June 2008 (1 year 6 months) Bengaluru Area, IndiaDevelopment Engineer @ Project : Infospace Mobile Search
Implemented change requirements for a product called Infospace Mobile Search ( C#, ASP.NET). From July 2006 to February 2007 (8 months) Bengaluru Area, IndiaProject Intern @ Project - Parallelization of Samba on HP NonStop/VSS
I proposed and implemented a scheduling algorithm for Samba, a file sharing software, so that the work load can be distributed across multiple processors. From 2005 to 2006 (1 year)
MS, PhD, Computer Vision, 3.91/4.0 @ University of Maryland College Park From 2008 to 2013 B.E. (First Class with Distinction), Computer Science and Engineering @ R. V. College of Engineering, Bangalore From 2002 to 2006 Jayant Kumar is skilled in: Machine Learning, Computer Vision, Image Processing, C++, Digital Image Processing, Algorithms, Matlab, C, Pattern Recognition, Computer Science, OpenCV, Linux, LaTeX, Artificial Intelligence, Python, Perl, Parallel Computing, Image Analysis, Image Segmentation, Android Development, Natural Language..., C#, Sentiment Analysis, Java
Websites:
http://www.umiacs.umd.edu/~jayant/,
https://www.quora.com/Jayant-Kumar-2