Adaptive Block Matching Based Quantization for Lossy Image Compression
Nikolay N. Ponomarenko, Karen Egiazarian · 2019
We propose an efficient method of quantization of discrete cosine transform (DCT) coefficients for lossy image compression. The main novelty of our approach is in utilization of masking ability of image regions having large values of error of block matching (searching patches similar to a given one). We propose to calculate a quantization level for a given image region in a proportion to a block matching error for this region, and describe the method of additional quantization of DCT coefficients for JPEG compression. We collect the mean opinion scores for a designed image test set containing compressed images for three different quantization schemes, and show by numerical analysis of 300 test images ofTAMPERE17 database, that the proposed method for the same compression ratio is able to provide significantly better MOS value, than one based on the conventional quantization. We also show that for different fixed values of the CSSIM4 metric, the proposed quantization provides a compression ratio increase by 30%-50% comparing to the conventional quantization.