My broad area of Interest is Computer Science. In particular, I am/have/like to work in the following areas:
- Machine Learning
- Information Retrieval
- Vision and Image Processing
Current Project: Masters' Thesis
Automating and Boosting Opinion Mining Framework for Defense Research and Development Organization (DRDO), Government of India
I am currently working on this project as part of my Masters's Thesis with Prof. Parag Singla, Assistant Professor, CSE, IIT Delhi and Prof. Maya Ramanath, Assistant Professor, CSE, IIT Delhi. Aimed at automating the framework for mining opinions and views on various DRDO projects which is currently done manually from different portals and websites.
Past Research Projects
I started working on this project during Summer'12 Summer Internship in Multimedia Search/Video Group at Yahoo! Labs, Bangalore under Research Scientist, Dhruv Kumar Mahajan and now continuing it this semester with Prof. Manik Varma, Researcher, Microsoft Research, India and Dhruv. The Idea is to come up with a Unified Hierarchy which can address drawbacks associated with currently used lines of thought for image classification, i.e., Semantic and Visual with three different labeled links
- Semantic (S)
- Visual (V)
- Semanti-Visual (SV)
which will give a different user experience for Similar Image Search while improving accuracies on image classification and retrieval
I worked on this project during last semester (Jan' 12 - Apr' 12) with Prof. Manik Varma, Researcher, Microsoft Research, India.
This was aimed at developing heuristics by traversing wiki pages of the query words which could be used for extracting subjects for generation of a new Image Search by Subject which is more informative and does not suffer from starvation of subjects and irrelevant subjects as was seen during analysis of current Google Image search by subject.
I also worked on developing algorithms for image re-ranking for optimizing the Image Search results.
The project was aimed at developing an Automated and Clinically-tested method for detection of Brain abnormalities and Tumor-Edema Segmentation using the MRI sequences.
The method uses no prior training, templates or atlases. Instead, it incorporates analysis of Multiple MRI Sequences, Symmetry Integrated Approach and Robust Intensity Thresholding techniques for first abnormality detection and then Tumor-Edema Segmentation.
We have tested this method on more than a hundred real dataset by developing a prototype by the name, AutoCom.
AutoCom secured 3rd Position in the National Innovation Contest, Techtop-2012, Trivandrum, Kerela, India. To know more, have a look on following:
- Automatic Detection of Brain Abnormalities and Tumor Segmentation in MRI Sequences in proceedings of Twenty-sixth International Conference Image and Vision Computing, Auckland, NewZealand (IVCNZ-2011)