Image Compression Based on Generalized Principal Components Analysis and Simulated Annealing

Abstract

The authors propose a new data dimensionality reduction method that is formulated as an optimization problem solved in two stages. In the first stage, Generalized Principal Component Analysis (GPCA) is used to find a solution with local maximum (local solution) whereas the algorithm Simulated Annealing (SA) is performed, in the second stage, to converge the local solution to the optimal solution. The performance of GPCA and GPCA with Simulated Annealing (GPCA-SA) as images compressors was evaluated in terms of the Compression Peak Signal-to-Noise Rate (CPSNR), memory size necessary to store the resulting compressed image and Contrast-to-Noise ratio. The results show that GPCA and GPCA-SA requires the same amount of memory to store compressed data, but GPCA-SA provides better CPSNR than GPCA. They also compared the performance of our designed method with a wavelet-based compression technique widely used in medical imaging, known as Lifting, to demonstrate the efficiency of GPCA-SA in clinical application.