|
Research |
|
Magnetic Resonance Imaging One of the problems in MRI is that it is a slow imaging modality. Accelerating MRI scans requires partial sampling of the data and sophisticated reconstruction algorithms. Today the majority of the algorithms are based on the precepts of Compressed Sensing. These algorithms exploit the sparsity of the MR images in a transform domain to reconstruct them from sub-sampled data. My research focuses on the developing reconstruction algorithms for a variety of MRI problems: Single Channel Static MRI Multi Channel Static MRI Single Channel Dynamic MRI Multi-echo MRI |
Opportunities Currently I am looking for motivated Master’s and PhD students for short and long term projects in these areas. |
|
Compressive Color Imaging This is a hypothetical problem. It is assumed that the image is captured by a modified Rice Single Sensor Camera. Instead of sampling the channels directly in the pixel domain, Bernoulli projections of each channel is acquired. The problem is to reconstruct the R-G-B channels given these projections. This is an academic exercise for color imaging since sensors in the optical range are cheap. Acquisition devices like the single pixel camera and such reconstruction algorithms become handy when the cost of the sensor is huge, and getting high resolution images is pricey; such as hyper-spectral imaging. Single sensor hyper-spectral cameras would cost a fraction of what they cost today. In such situations, the reconstruction algorithms developed for color imaging can be directly applied. |
|
|
Collaborative Filtering In any online retail, once u view an item, they should you ‘items that are frequently bought with it’ or ‘users who have bought this also bought …’. These are all recommendations. If you like the recommendation, you buy the items, the online retail store makes money. Thus how much money the store will make is largely dependent on how good the recommendations are. Today every online retail store has such a built-in ‘recommender system’. This recommendation is based on your buying pattern as well as from buying pattern from other users. The system tries to predict your choice given the choices of other users. This kind of prediction is known as Collaborative Filtering. Currently my interest is in building computationally efficient recommender systems. |
|