Image and Video Resolution Enhancement Using Sparsity Constraints and Bilateral Total Variation Filter

In this thesis we present new methods for image and video super resolution and video deinterlacing. For image super resolution a new approach for finding a High Resolution (HR) image from a single Low Resolution (LR) image has been introduced. We have done this by employing Compressive Sensing (CS) theory. In CS framework images are assumed to be sparse in a transform domain such as wavelets or contourlets. Using this fact we have developed an approach in which the contourlet domain is considered as the transform domain and a CS algorithm is used to find the high resolution image. Following that, we extend our image super resolution scheme to video super resolution. Our video super resolution method has two steps, the first step consists of our image super resolution method which is applied on each frame separately. Then a post processing step is performed on estimated outputs to increase the video quality. The post processing step consists of a deblurring and a Bilateral Total Variation (BTV) filtering for increasing the video consistency. Experimental results show significant improvement over existing image and video super resolution methods both objectively and subjectively. For video deinterlacing problem a method has been proposed which is also a two step approach. At first 6 interpolators are applied to each missing line and the iv interpolator which gives the minimum error is selected. An initial deinterlaced frame is constructed using selected interpolator. In the next step this initial deinterlaced frame is fed into a post processing step. The post processing step is a modified version of 2-D Bilateral Total Variation filter. The proposed deinterlacing technique outperforms many existing deinterlacing algorithms.