Performance improving in skew-coordinates DCT method for images by entropy coding based on gaussian mixture distribution model

In the skew-coordinates DCT coding method, an image is partitioned along the edge into variably shaped blocks and coded using the skew-coordinates DCT adopted to the edge direction. Compared with the conventional square block DCT, it is known that the skew-coordinates DCT can improve power packing efficiency and can reduce mosquito noise in the reconstructed image. In this paper, the entropy coding method of the skew-coordinates DCT is studied in order to improve the compression ratio. As an entropy coding method, we adopt an adaptive code allocation method based on the Gaussian mixture distribution model and study the construction of the mixture model. Statistical characteristics of the DCT coefficients in real images are investigated and it is shown that the warping effect of the skew-coordinates DCT can reduce the local variation of the variance distribution of the DCT coefficients and that, consequently, a simple mean-power model is suitable as the mixture model. Finally, the result of a computer simulation experiment shows that the proposed method is useful in improving coding performance.