Research on statistical distributions of transform coefficients for H.264/SVC

As a substitute to discrete cosine transform (DCT), a more complicated transform scheme has been adopted in H.264/AVC and inherited by H.264/SVC. For supporting the scalability, H.264/SVC incorporates the inter-layer prediction mechanisms, which lead to the variety of the composition of residual frame as the input of the transform. However, so far scarcely any study has been carried out on the transform coefficients distribution of H.264/SVC. With the method of Maximum Likelihood estimation, this paper presents the study of performing two goodness-of-fit tests, the Kolmogorov-Smirnov (KS) test and the χ² test, to decide the best distribution among three famous distributions, the Generalized Gaussian distribution (GGD), the Laplacian distribution (LAP) and the Cauchy distribution (CCH), for the transform coefficients of the luminance components of video sequence coded by H.264/SVC encoder. The results indicate that the Cauchy distribution can still be considered as the best description for the transform coefficients in most cases of H.264/SVC.