COMPRESSION OF SIGNS OF QUANTIZED DISCRETE COSINE TRANSFORM COEFFICIENTS OF JPEG IMAGES

Abstract

Images compressed in accordance with the JPEG standard may be additionally lossless compressed to reduce the amount of data transmitted over the communication channels or to reduce the space occupied by such images on data storage devices. In this case, a JPEG image is partially decoded to obtain quantized coefficients of the discrete cosine transform (DCT) of image blocks. Then these coefficients are compressed by some more efficient method. Most of such methods are able to efficiently compress amplitudes of the DCT coefficients, while signs of the DCT coefficients are compressed weakly or even are not compressed. In this paper, we propose an effective method of compressing of signs of quantized DCT coefficients of JPEG images. The proposed method takes into account correlation between DCT coefficients inside of an image block as well as between DCT coefficients of neighboring blocks by forming the corresponding frequency models and by combining these models using logistic mixing. For a test set of real JPEG images formed by various digital cameras, it is shown that the proposed method allows to compress signs of DCT coefficients in 1,1 ... 1,3 times, significantly outperforming nearest analogues