Welcome to the Computational Vision lab of IIT Kharagpur

Our research is centered around low-level vision problems such as shape-from-X, deblurring space-variantly blurred images, depth and image inpainting/disocclusion and super-resolution. We aim to propose new optimization-based algorithms to address these problems. In research work conducted earlier, we have demonstrated several improvements to the basic shape-from-focus technique. Significantly, we have proposed an approach to handle parallax as well as space-variant blur caused due to axial relative motion between the scene and the camera. This idea can be extended to perform super-resolution and reconstruct high-resolution image as well as depth map using a practical camera. This is the goal of one of the sponsored projects currently being handled.

We have several publications aiming to remove fences/occlusions from images captured by a smartphone camera with relatively arbitrary motion. This interesting problem has several challenges and we seek to propose machine learning and optimization based techniques to segment out fence pixels and remove them from the input image. Also, we have proposed a multi-modal approach to this problem by using the Microsoft Kinect sensor to obtain RGB-D data.

Recently, we have shifted our attention to understanding multimedia data. Specifically, we seek to understand Indian classical dance from video data. We have used deep learning algorithms for body pose and gesture identification in such videos.