Compressed Sensing Based Memory Optimized Representation of Dual Basis Decomposed Image

Abstract

Compressed sensing (CS) has attracted considerable research interest in the past decade because of its ability to surpass Shannon-Nyquist bounds on sampling signals. It is applicable for signals that are inherently sparse or compressible in some suitable basis. Since most of the signals occurring in nature can be broadly classified into one of these categories, CS has immediately found applications in varied fields of engineering. In this paper we applied CS to compress an image. Most of the published literature in this area has considered the standard approach of finding the sparsifier for the input signal and subsequently applied CS on the sparsified data. Innovations were presented on the implementation of the sparsifier viz block based coding, wave-let lifting scheme, etc. In this work we have considered dual basis structure for image decomposition followed by CS. We identified compatible basis pairs which augment each other in terms of data compression and make the data amenable for the application of CS for providing im-proved compression ratios. We have considered multiple configurations of the dual basis kernels and tabulated the compression efficiencies. We have quantified our algorithm using Image Quality Assessment (IQA) metrics viz Mean SSIM, PSNR.