Super-resolution (SR) is a problem of achieving a high resolution (HR) image from low resolution (LR) image, and is an ill-posed problem. Sparsity inducing norm can be used to address the SR problem. However, it may smears the perceptually important edges in the SR result. To mitigate with this concern, we have proposed an edge preserving constraint that preserves the edges of the input image in the SR result. This edge preserving SR strategy is further employed to up-sample LR depth map (or range image). The requirement of higher up-sampling factor for range image is addressed by a pyramidal strategy. Often the initial part of depth estimation pipeline involves structure from motion, which yields depth at selected number of points. Hence, the available depth information is sparse in nature, which can be interpreted as an LR depth map, sub-sampled in a non-uniform grid. To up-sample the non-uniformly sampled LR depth map, we generalize the SR framework using a mask operator. Here, the missing depths at HR grid is filled using the dictionary of exemplars. Dictionary plays an important role in sparsity based SR. Hence, we learn multiple dictionaries by employing structural information (dominant edge orientation) as well as statistical information (mean of intensity values) of patches, extracted from example images. If the example patches are unavailable, we construct image pyramid by up/down-sampling the given LR image, and patches from the pyramids can be used to learn dictionary. Further, we have chosen the image patch details for SR, as it contains perceptually significant information. If the given LR image is contaminated by noise, considering patch detail for SR will emphasize the noise also. To mitigate with this issue, we derive some parameters from the given LR image that reflects the strength of noise present in the image. These parameters are used: i) to derive a threshold that is employed in the sparse coding stage using iterative shrinkage/thresholding algorithm, and ii) to choose between the noise suppressing non-local mean component and the detail component. By enhancing suitable component using iterative thresholding algorithm, we are able to suppress noise while super-resolving a single image.
Srimanta Mandal has received the bachelor’s degree in electronics and communications engineering from the West Bengal University of Technology, West Bengal in 2010. He has joined IIT Mandi, in MS by research program in 2010, and has converted to PhD program in 2013. He has submitted his PhD thesis in November 2016, and is currently working on depth from single image and super resolution of MRI images. His research interests include image and video super-resolution, point cloud completion, and medical image analysis.