A discrete wavelet transform-singular value decomposition system for image coding

Abstract

A system that combines techniques of wavelet transform (DWT) and singular value decomposition (SVD) to encode images is presented. The image is divided into tiles or blocks of 64x64 pixels. The decision criterion as to which transform to use is based on the standard deviation of the 8x8 pixel subblocks of the tile to encode. A successive approximation quantizer is used to encode the subbands and vector quantization/scalar quantization is used to encode the SVD eigenvectors/eigenvalues, respectively. For coding color images, the RGB components are transformed into YCbCr before encoding in 4:2:0 format. Results show that the proposed system outperforms the JPEG and approaches the JPEG2000.