Algorithm Research And Improvement On JPEG2000 Standard

JPEG2000 is a new standard for the compression of still image established by the International Organization for Standardization (ISO) and the International Telecommunications Union (ITU). Compared with former standards, JPEG2000 provides a set of features, such as superior low bit-rate performance, excellent compound image performance, progressive transmission by pixel accuracy and resolution, Region of Interests coding, robustness of bit errors. Because of these features,JPEG2000 standard applies to many fields such as image compression and network transmission and gradually become major standard of still image compression. In this paper, we analyzed its major algorithms and improve two parts of them.Firstly, we studied the structure of the standard and codec procedure of its part one. The encoding procedure includes several parts: preprocessing, wavelet transform, quantization and entropy coding. Preprocessing includes tiling, DC level shifting and component transformation. After that, wavelet transform is used for the analysis of the tile components into different decomposition levels. Then quantizedwavelet coefficients are compressed using entropy coding and embedded bit stream is acquired.Wavelet transform is one feature of JPEG2000 standard. We have reviewed the history of wavelet transform and analyzed multi-resolution theory firstly, then we studied basic method of lifting scheme and discussed how to realize traditional wavelet transform using lifting scheme.EBCOT is the kernel part of entropy coding in JPEG2000. We proposed a new algorithm which can find the reconstruction point self-adaptively which based on EBCOT. Some features such as SNR and resolution scalability and "random access" are achieved through bit-plane coding, and excellent compression performance through PCRD-opt In order to overcome the disadvantage that the reconstruction point is fixed to the middle of quantization section in the step of anti-quantization, a new method is proposed. In this method, the reconstruction point can be selected self-adaptively at the cost of trivial bits increase so that reconstruction coefficients are closer to original coefficients. The result indicates that this method can increase PSNR and improve the quality of reconstructed image.At last, we focused on the ROI method in JPEG2000 standard: maxshift method and scaling based method and a new ROI method is proposed. The two traditional methods implements ROI through shifting background coefficients, which waste lots of time and resources. To overcome this problem, "S-twice coding" are proposed. In this method, the highest bit-planes are encoded twice instead of shifting the background coefficients, implementation complexity are reduced, and the decoder is simpler. The result shows that the encoding time shortens 5%-10%.