An Improved SVD based Image Compression

Advent of information technology since last two to three decades led to the enormous generation of multimedia data. Digital images form a major share of multimedia data; hence a large interest over the area of image compression had been seen among researchers. Image compression deals with exploiting visual imperceptibility of human eye to encode data with much smaller number of bits for the same amount of information. Singular Value Decomposition (or SVD as it is commonly abbreviated) based image compression had been extensively studied in the past few decades. We present an improved approach over the technique using the orthonormal property of the matrices produced in SVD decomposition method. We show that we do not need to store full vectors in order to preserve the orthonormal matrices of eigenvectors instead if we store partial vectors then also we can generate the rest of the values using the orthonormal properties of the matrices of eigenvectors. We further derive a mathematical expression for the compression ratio for this scheme and extend it to derive an expression for the gain thus involved. We present the experimental results showing the PSNR, SSIM and compression ratio of the proposed scheme with that of the state of the art SVD compression scheme. The results indicate that the proposed scheme improves the compression ratio while maintain the image quality.