Comparative Performance Evaluation of SVD-Based Image Compression

Image compression is an absolute need for current multimedia storage/communication situations. Several image compression algorithms are in daily use. In this paper we consider rank reduction algorithm based on singular value decomposition (RR-SVD), when applied to JPEG coded images. The SVD-based rank reduction algorithm compresses the image by eliminating some of its singular values based on human eye sensitivity. We evaluate the performance of RR-SVD when applied to the whole image in comparison with its block-based application on a number of images in different resolutions and with different compression parameters. We focus on compression ratio (CR), compression time and compressed image quality versus the original image. Our results indicate that block-based application of RR-SVD outperforms whole image application in most cases with a noticeable margin depending on the resolution. Also, we provide suggestions for choosing the optimum operating point based on our results.