Content area
Vector Quantization (VQ) is an attractive technique for image coding because of its low bit-rate requirement. Many studies focus on this topic attempting to improve the VQ performance. The existing improved VQ techniques have achieved 0.2-0.7 bit/pixel for monochrome and 0.4-1 bit/colour-pixel for colour image coding with satisfactory reconstructed images in which gray scale transitions look natural and edges are clear.
In this thesis, several techniques for low bit-rate image coding are introduced, which are the Hierarchical Multirate Vector Quantizations, the Hierarchical Finite State Vector Quantization, the Edge Modified Vector Quantization and the Trimming Vector Quantization. These techniques are developed based on the hierarchical vector quantization combined with finite state predictions and multistage processing. They are also successfully applied in colour image coding. Another type of image coding scheme called the Hierarchical Finite State Visual Pattern Image Coding is presented, which is rather similar to vector quantization but has extremely low computational complexity. All these techniques make image coding strategies more adaptive to image structures and features, so that we can further reduce coding redundancies and more flexibly adjust reconstruction qualities in different regions based on human eye sensitivity to the coding distortions. With the same reconstruction quality achievement as the ones of the existing improved VQ techniques, the bit-rates for the new image coding techniques can be as low as 0.1-0.3 bit/pixel for monochrome image coding and 0.2-0.4 bit/colour-pixel for colour image coding. The computational complexities in the new techniques are 15- 20% of the related VQ techniques without hierarchical structures. The computations of the Hierarchical Finite State Visual Pattern coding are fewer than 50% of the JPEG's. On the other hand, limitations of the presented coding techniques, which also reflect the common limitations of VQ techniques, are discussed in the thesis.