Compressed sensing for image processing

Abstract

In the present-day scenario, there are various methods to process and represent a signal according to our desired outcome. This dissertation deals with the image processing operations of ‘Denoising’ and ‘Inpainting’ using Compressed Sensing (CS) measurements (algorithms). This thesis work focuses on the sparsity of real-world signals. Sparse representation of images is a new measure and its applications are promising. Complete and Overcomplete signal dependent representations are the new trends in signal processing, which help in sparsifying the redundant information in the representation domain i.e. the dictionary, which has been discussed in the upcoming chapters in further detail. The objective of signal dependent representation is to train a dictionary from training signals and sample signals. In this dissertation, we have experimented with the CS algorithm, considering two different black & white images called ‘Lena’ and ‘Barbara’. Denoising has been performed for the noisy images and Inpainting has been performed while taking different masks into consideration. The objective is to recover large dimension sparse signals from a small number of random measurements. The thesis work shows that CS algorithm is an effective approach to process images which is evident from the results that are obtained in subsequent chapters. All the other details have been discussed in the subsequent chapters.