Researches On High Performance Image Coding

In the field of image coding, one of the goals is to pursue compression algorithm with higher performance. To choose suitable image coding methods under different situations is the key to apply in actual system. In this dissertation, main work on image coding in recent years are systematically introduced, emphasis on the algorithm design of newly emerged wavelet theory and the application of vector quantization in the domain of remote sensing image. As the development of harmonic analysis, wavelet theory has met wide application dependent on its engineering background, which is very suitable for the research of image coding. At first, basic concepts in wavelet theory is expounded, focused on wavelet transform, multi-resolution analysis, Mallat algorithm and filter-banks etc. By examining the time-frequency characteristics of the wavelet basis, the rule of best basis selection is proposed. Two new image coding schemes based on the biothogonal wavelet transform with Classified Vector Quantization (CVQ) and embedded coding are presented. The experimental results show that the proposed schemes can improve the performance and reduce the complexity. Since I 990s, the research of vector quantization (VQ) on image coding has gradually become popular in application. At first, basic principles and common technology of VQ are brief introduced, and a new rapid and high-quality VQ initial codebook generating method, Adaptive Decision- Directed Clustering (ADDC) is presented. Secondly, A high-speed and real-time remote sensing image coder is implemented, which has an input bit-rate of 240 Mbps, and good quality of reconstructed image with a compression ratio of 8. Base on it, a new fast VQ search algorithm using Energy Band Segmentation (EBS) is presented, which is recommended in the actual satellite codec system. Over along period of time, a puzzle in the area of image coding is that MSE is not well coincided with the subjective perception. By inspecting physiology and psychology on human vision, we attempt to study the perceptual characteristics of the image. Based on the human face location technique, a novel Quality Assessment Measure (QAM) is given which can correctly reflect the subjective visual reaction of the head-and-shoulders scene, and it can be incorporated to design the quantizer used in low bit rate sequence image coding. The last part of the dissertation is the conclusion. After a brief overview, we simply promise some future research directions and challenging problems.