Image coding using Variable-Size Transforms and a full binary tree recursive optimization

In this work, we present the first results of a new algorithm for image compression, called Variable Size Transform Coder (VSTC). This algorithm uses spatial transforms and recursive optimization functions combined with a highly adaptive segmentation scheme, similar to the one proposed by the MMP algorithm (Multi-dimensional Multiscale Parser), that uses pattern matching. We replaced the pattern matching by spatial multiscale transforms, which include symmetric and asymmetric blocks, and we investigated and implemented some of the required tools for encoding these new blocks. The proposed method uses a much more dynamic adaptive-size transforms than the H.264/AVC standard and compares favorably with this encoder. It also consistently outperforms the pattern matching scheme of the MMP algorithm. Moreover, we have considerably reduced the computational complexity when compared to the MMP algorithm.