Mining Software Repositories Among the vast set of software maintenance tasks, a task that is unavoidable and time consuming is bug fixing. It has been observed that a programmer will spend approximately 60% of his time fixing bugs and a majority portion of that time is spent in understanding the code. Narrowing down the set of files responsible for the bug is one of the initial steps a programmer must take to solve the bug. This is called the problem of bug localization and is the focus of my PhD dissertation. In this dissertation we address the effectiveness and efficiency of the IR based bug localization tools by bringing together concepts from various prominent fields like Statistical Analysis, Data Mining, Machine Learning and Software Engineering. In order to render credibility to our findings, we carried out rigorous experimented using benchmark datasets created by other researchers as well by creating our own large scale benchmark dataset. The datasets were created by tracking commit history and bug-fixing history of open source software repositories over a span of 10 years
Texture Analysis During my Masters at IIT Madras, I worked on "2D texture analysis" where a 2D image having multiple textures is segmented using tools like wavelets and Gabor filters. We developed an optimal unsupervised classification algorithm where a combination of Gabor and wavelet filters is used to segment a texture-image into different segments based on texture frequency of that segment. My Masters's thesis was focused on "Shape from Texture". This involved studying the problem of obtaining the angle of orientation of a surface from its single available static image and known focal length. The varying texture frequency due to perspective distortion was tapped using the energy distribution across various wavelet coefficients obtained by the wavelet transform. A new theoretical framework using the azimuth elevation projection model was designed to explain the texture variation occurring on the image plane and experiments were carried out to support the validity of the claim.