Performance Analysis of Sparse Image Compression and Reconstruction Using Various Sensing Matrix Structures and Orthogonal Matching Pursuit Algorithm

Abstract

Efficient image compression and reconstruction techniques are crucial for reducing storage. This paper presents a comprehensive evaluation of sparse image compression and reconstruction performance using structured sensing matrices in conjunction with the Orthogonal Matching Pursuit (OMP) algorithm. Four types of sensing matrices—Gaussian Random, Bernoulli, Partial Fourier, and Hadamard—are analysed under varying measurement rates and DCT coefficient thresholds. The study measures reconstruction quality through key metrics such as Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), entropy variation, and compression ratio. Experiments on 256×256 grayscale images reveal that Hadamard and Bernoulli matrices generally outperform others in balancing compression efficiency and visual quality, especially at moderate sparsity levels. Additionally, thresholding strategies significantly influence the trade-off between entropy and reconstruction fidelity. The results support the potential of structured sensing in efficient compressive imaging applications and provide insights into optimal matrix and threshold configurations for enhanced performance.