Fast sparsity adaptive matching pursuit algorithm for large-scale image reconstruction

Sheng Wang,Shihong Yao,Qingfeng Guan,Xiao Xie

doi:10.1186/s13638-018-1085-6

Abstract

The accurate reconstruction of a signal within a reasonable period is the key process that enables the application of compressive sensing in large-scale image transmission. The sparsity adaptive matching pursuit (SAMP) algorithm does not need prior knowledge on signal sparsity and has high reconstruction accuracy but has low reconstruction efficiency. To overcome the low reconstruction efficiency, we propose the use of the fast sparsity adaptive matching pursuit (FSAMP) algorithm, where the number of atoms selected in each iteration increases in a nonlinear manner instead of undergoing linear growth. This form of increase reduces the number of iterations. Furthermore, we use an adaptive reselection strategy in the proposed algorithm to prevent the excessive selection of atom. Experimental results demonstrated that the FSAMP algorithm has more stable reconstruction performance and higher reconstruction accuracy than the SAMP algorithm.

Highlights

The explosive growth of information has brought a great burden for signal processing and storage
We propose a fast sparsity adaptive matching pursuit (FSAMP) algorithm where the number of selected atoms is changed from an original linear to a nonlinear growth form
5 Conclusions Obtaining timely and accurate reconstruction results is the key focus of compressed sensing (CS) application for large-scale image transmission

Summary

Introduction

The explosive growth of information has brought a great burden for signal processing and storage. The Nyquist sampling theorem increases the cost and lowers the effectiveness of data acquisition and processing equipment in the transmission and processing of large-scale image data [1, 2]. The methodology of signal processing in CS reduces the number of measurements during the sampling process but still retains sufficient information. It has a great application prospect in large-scale image processing owing to its low CS involves a three-part process, namely, signal sparse representation, signal compression under measurement matrix, and signal reconstruction. The performance of reconstruction algorithm is mainly reflected in the two aspects of reconstruction efficiency and reconstruction accuracy. The key in the application of CS is to design a good reconstruction algorithm that can balance the reconstruction efficiency and the reconstruction accuracy

Objectives

Conclusion