Research

I'm interested in image processing, computer vision, computational photography, and deep learning. Much of my interest lies in constructing mathematical model of a given problem from first principles and in developing respective inverse model; taking into account ill-posedness (if present) via regularization theory, and other computational considerations such as convex optimization, computational cost, and convergence.

Representative publications
Unconstrained Motion Deblurring for Dual-lens Cameras
Mahesh Mohan M. R., Sharath Girish, and A. N. Rajagopalan
ICCV, 2019   (Oral Presentation)
Coming soon ...

Motion deblurring for dual-lens cameras possess an ill-posedness, which calls for a prior for depth-consistent deblurring.

Divide and Conquer for Full-resolution Light Field Deblurring
Mahesh Mohan M. R. and A. N. Rajagopalan
CVPR, 2018
paper / supplement / slides / poster / bib

Full-resolution light field deblurring can be divided into independent subtasks, wherein a single task reinforces other tasks.

Occlusion-Aware Rolling Shutter Rectification of 3D Scenes
Subeesh Vasu , Mahesh Mohan M. R., and A. N. Rajagopalan
CVPR, 2018
paper / supplement / poster / bib

A method for the scenario of a fast moving camera wherein rolling shutter distortions results in intra-frame occlusions.

Going Unconstrained with Rolling Shutter Deblurring
Mahesh Mohan M. R., A. N. Rajagopalan , and Guna Seetharaman
ICCV, 2017
paper / supplement / slides / poster / bib

How can we bridge today's ubiquitous rolling shutter cameras with the well-studied conventional cameras, pertaining to motion deblurring?

Deep Decoupling of Defocus and Motion blur for Dynamic Segmentation
Abhijith Punnapurath , Yogesh Balaji , Mahesh Mohan M. R., and A. N. Rajagopalan
ECCV, 2016
paper / supplement / poster / bib

Deep Learning is used to obtain the attributes of object motion and camera motion, which is then employed for segmenting moving object(s).

Miscellaneous
Noise-aware Detail Enhancement in Scanning Electron Microscope imagery.
Mahesh Mohan M. R., A. N. Rajagopalan and Raj Kuppa,
KLA-Tencor's Neoterix, 2018   (in use at KLA-Tencor)
abstract / (rest is proprietary of KLA-T; sorry.)

A fully-automatic denoising based on a convex prior on SEM images, which eradicate (difficult to infer or possibly erroneous) noise models.

A novel method of Medical Image Denoising using Bilateral and NLM filtering.
Mahesh Mohan M. R. and Sheeba V. S.,
ICACC, 2013   (Oral Presentation)
paper / technical report / slides / bib

Wavelet thresholding (Visushrink) is extended to Contourlet transform, in order to enhance the performance of bilateral and NLM filtering.

Copyright notice: Copies of the papers are provided here for convenient dissemination of scholarly work. They can be downloaded for non-commercial research and education purposes only. Copyrights of the papers usually belong to the publishers of the journals or proceedings and must be adhered to by anyone using these materials. For other materials, feel free to use it at your convenience for good deeds!


Pure mathematics is in its way the poetry of logical ideas. When the solution is simpler, God is answering -- Einstein