Abstract: paper, a lossy image compression algorithm based on a prediction and classification scheme is discussed. The algorithm decomposes an image into four subimages by subsampling pixels at even and odd row and column locations. Since the four subimages have strong correlations to one another, one of them is used in predicting all the others and the resulting differences between the predicted subimages and the original subimages are encoded. Estimated differences tend to be large in a region where pixel values change rapidly, while the differences are small in a monotonous region. This redundancy is explored by dividing the estimated differences into subsets based on the slope of pixel changes, the basis for which is found in some human perception models used to measure the visibility of distortion. The resulting classified estimated differences having different visibilities are encoded with classified vector quantizers.!14
展开▼