A Review on Compressive Sensed Image Reconstruction using Group-based Sparse Representation
Divya Velayudhan1, Salim Paul2
1Divya Velayudhan, Student, Department of Electronics and Communication Engineering, Sree Chitra Thirunal College of Engineering, Pappanamcode, Thiruvananthapuram (Kerala). India.
2Salim Paul, Associate Professor, Department of Electronics and Communication Engineering, Sree Chitra Thirunal College of Engineering, Pappanamcode, Thiruvananthapuram (Kerala). India.
Manuscript received on 13 June 2016 | Revised Manuscript received on 20 June 2016 | Manuscript Published on 30 June 2016 | PP: 64-67 | Volume-5 Issue-5, June 2016 | Retrieval Number: E4603065516/16©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Compressive Sensing (CS) – a novel sensing paradigm asserts that signals can be reconstructed from fewer samples than that recommended by Nyquist sampling theorem, when it can be expressed in a sparse basis. Conventional approaches for compressive sensed image recovery utilized fixed basis (DCT, wavelets) that do not yield higher level of sparsity for the entire signal resulting in poor performance. This paper reviews the performance of Group-based sparse representation (GSR) model for CS recovery which yields high degree of sparsity for natural images in the domain of group. GSR stacks together non-local similar patches in an image to form a group and the sparse representation of each group is achieved using self-adaptive dictionary learning technique. Thus GSR takes advantage of the intrinsic local sparsity and non-local self-similarity of images simultaneously in a unified framework. The GSR driven optimization problem is solved using split-bregman iteration. Experimental results obtained on images for CS recovery reveals the performance achieved by GSR over many current state-ofthe-art schemes
Keywords: Compressive Sensing, Sparse Representation, Self-Similarity, Split-Bregman.
Scope of the Article: Knowledge Representation and Retrievals