Improved MPEG-4 visual texture coding using double transform coding

MPEG-4 Visual Texture Coding (VTC) was recently adopted as an image coding technique using a wavelet-based approach. For images with significant texture and sharp contrast, MPEG-4 VTC spreads the signal energy over several wavelet trees. To improve the coding efficiency of zerotree entropy coding, we propose a double transformation coding (DTC) approach. The wavelet trees are reorganized as a sequence of wavelet blocks, which are categorized into four groups where the similarity among the blocks within a group is increased. For each class, we use an adaptive Karhunen Loeve Transform (KLT) bases to predict the quantized AC coefficients of the same class. The classification process is based on the variances of the quantized DC coefficients considering the spatial orientation. Thus, the classification information can be computed at the decoder. At the same time, the property of continuous scalability is maintained using the new approach. For the same bitrate we found that the DTC achieved a coding gain over MPEG-4 VTC by about 0.5-1.0 dB in PSNR.