Color image transform coding based on three dimensional variable vector matrix size segmentation

Abstract

Three-dimensional vector matrix discrete cosine transform coding (3DVMDCT) deals with each component of color image in a unified model, fully eliminates the correlation between them and gets an obvious advantage. But its block-size of image segmentation is fixed and does not fully take into account that there are different statistical properties in different regions. In this paper, we take gradient as the image measure activity (IAM), propose a new method—variable matrix-size three-dimensional vector matrix image segmentation (VMS-3DVMDCT), and implement the corresponding multi-dimensional vector matrix discrete cosine transform coding, experimental results show that, compared with the fixed block partition, the proposed algorithm improves PSNR at most 1dB, and the quality of the reconstructed image is improved in subject evaluation, much better than JPEG.