Ish Rishabh
Position: Ph.D. student
Area: Multimedia and Computer Vision
Advisor: Prof. Ramesh Jain

Contact details:
Office: DBH 3209
Office fax: +1(949)824-4056
Email: irishabh at ics dot uci dot edu

About me
I am a Ph.D. student at UC Irvine. I joined the program in Fall 2007 and am now in the third year of my studies. Before joining UCI, I worked as a Research Scientist for 4 years at Centre for AI & Robotics (CAIR), Bangalore, India. My research here is directed toward smart-environments in which information from various sensors (mostly cameras) is assimilated to help the application make smarter decisions to facilitate the users.

About my research at UCI
My research is focused on making applications smarter by developing models to describe surrounding environment. Once the environment context is captured, applications like communication, assisted living and aware-systems can make use of this context to make smarter decisions.

Following is a brief summary of my work at UC Irvine.

Semantic camera coverage determination using 3D floorplan (MS thesis)
3D floor-plan is used to calibrate a camera with user assisted correspondence matching. To help the user, an algorithm is developed that ranks 3D features (points or lines) based on their ability to provide independent calibration constraints. This way, features that are redundant can be pruned from the list of features that the user can use to specify correspondences. Various feature configurations were studies and analyzed to come up with a set of rules to determine the rank of features. This work comprises my MS thesis.
Environment to environment (E2E) communication system
Aim is to facilitate communication between two or more environments by using a sentient information system that employs adequate number of sensors to detect important events and archive them. A subset of these events can be shared over the Internet to communicate with others. Involves study and implementation of algorithms for real-time analysis of media (images, video, audio, motion), face and person detection, media streaming and archival/retrieval of events.
Environment modeling
Environment Model (EM) is an important module of E2E systems. It describes the ambient sensory and actuation environment. Sensor (camera) calibration, signal to physical environment mapping and 3-D environment floorplan integration are required to describe the parameters of environment model.
Automatic User Availability Detection
Visual, audio and contextual (metadata) information are used to automatically detect and display the status of the User over Skype. Status, as determined from contextual information (calendar, address-book) is combined with the physical status of the user to update the Skype status automatically. This required use of vision-based techniques along with the Skype APIs.

Publications

About my research at Centre for AI and Robotics (CAIR)
Change Detection in bi-temporal images
Pixel based analysis of bi-temporal images for detection of changes in optical satellite imagery. Devised a robust method using RANSAC for radiometric correction of images.
Image Classification
Devised and implemented an object-based method for classification of Synthetic Aperture Radar (SAR) images into one of the four classes: mountain, urban, river and lakes, using Minimum Distance Classifier. Devised an extension of Mahalanobis distance for improved distance measure.
Image Segmentation
Devised a block-based clustering method for segmentation of multispectral images. Implemented Watershed segmentation, both Toboggan and Immersion schemes. Also implemented a few clustering based methods for segmentation.
Progressive Image Transmission
Devised and implemented a method to transmit large images over unreliable and narrow communication media (radio links) based on Symmetric Residue Pyramids that outperformed Progressive JPEG perceptually.