New approaches to fractal image and very low bit-rate video compression

Because of the rapidly increasing demand of audiovisual services over today's existing communication channels, very low bit rate video coding has become a hot research topic. Although there are some video coding standards such as ITU-T H.261 and MPEG-1 which have been established recently, their compression ratios are not high enough to allow audiovisual bit stream being transmitted over most communication channels including telephone lines and wireless com-munication channels. A new standard called ITU-T H.263 is just published. However, it is believed that ITU-T H.263 is only a short term solution for very low bit rate video applications. Thus new techniques other than hybrid motion estimation and discrete cosine transform are needed for the long term solution of very low bit-rate video compression. It has been shown that fractal coding has the potential to achieve huge com-pression ratio, which makes fractal a popular candidate for the new video coding standard called MPEG-4. Nevertheless, its great computational complexity pre-vents it from being applied to real applications. In this thesis, we propose new techniques to reduce the computational com-plexity. These techniques include the Statistical Normalization and Frequency Domain Comparison, which achieve high reduction in computational complexity with only little degradation in visual quality. We also try both to reduce the bit allocation for the fractal parameter and to maintain the visual quality using de-blocking techniques. Finally, we propose a hybrid fractal and feature-based video coding scheme. The simulation results show that we can obtain acceptable image quality by using our proposed scheme in very low bit-rate applications