The Research Of Image Compression Coding Technology Based On Bandlet Transform

Multiscale geometric analysis (MGA), prosposed in recent years when wavelet theory is in-depth study, is a mathematical analysis tool which integrates the latest research results of different disciplines, such as mathematical analysis, computer vision, pattern recognition, statistical analysis, etc. The performance of sparse representation of MGA surpasses Fourier analysis and wavelet analysis, and therefore it is more suitable for compressing natural images rich in textures or SAR images. Bandlet is a successful MGA theory in theoretical research and practical applications. Centering on the theme of Bandlet transform and image compression coding, the main work of this thesis is as follows:(1) Some basic theories, algorithms and techniques are summarized, such as image coding framework, sparse representation, geometric regular image, approximation theory and so on; A new type of image quality evaluation system - SSIM is introduced; The basic concepts and ideas of the first generation Bandlet (1G Bandlet) and second generation Bandlet (2G Bandlet) are expounded, mainly including geometric flow, construction of basis, framework, approximation, transform in wavelet domain, bandletization, quadtree decomposition, transform steps ,etc.(2) A two-layered compression method is presented. The method combines 2G Bandlet and Wavelet together. 2G Bandlet does well in compressing high-frequency textural information, and Wavelet is a powerful tool for low-frequency structure. So the two tools can be combined into an efficient two-layered image compression method. Numerical experiments prove that this method outperforms JPEG 2000 and singular Bandlets both in visual effect and objective SSIM values.(3) An image compression technology integrated 2G Bandlet and EBCOT is studied. Based on the research, a lossy compression system for still gray images is designed and implemented. In the system, 2G Bandlet and EBCOT are the core transform and encoding style respectively. The framework of the compression system is given integrally in the thesis, and the input and output of each module are also parsed. The system designed in this thesis can not only achieve image compression, but also some advanced features similar to JPEG 2000 system, such as resolution and quality progressive transmission.