Projects at IIT Kanpur
Source Localization using Sparse Reconstruction
I developed a new algorithm for localizing sources of sound in 3D space using spherical microphone arrays. The project involved use of sparse reconstruction technique to get high resolution estimation. My initial work is accepted in CAMSAP 2013 and another in Interspeech 2014.
Speech Source separation using ICA and MUSIC Group Delay
We worked on separating individual speech sources from a mixture of speech signals. This problem is called "Cocktail Party Problem". We came up with a new algorithm to solve the problem where we combined Independent Component Analysis(ICA) and MUSIC Group Delay methods. Our algorithm converged faster than ICA and also improved the Word Error Rate. Our work is published in NCC 2013.
I segmented moving video objects of a video sequence (even in case of occlusion) using optical flow motion vectors. The motion vectors were derived using "Horn-Schunck Optical flow" method.
Sparse Representation Based Blind Image De-blurring
We Implemented an algorithm for sparse recovery of original image from blurry observation. It Resolved multiple optimization problem by iterative alternate minimization scheme. Also Performed comparison with and observed 60% more accurate results than fast de-blur algorithm.