Research Of The Image Algorithm Of Compession And Coding Based On Wavelet Transform And Its Parallelization

Digital image has the features of intuitive, clear and efficient , but its mass of data to the storage and transmission are also presented problems, so image compression is inevitable and the study of image compression algorithm has been now hot spots in information processing.In recent years, image compression based on wavelet transform has achieved rapid development. Different quantization and coding of transform coefficients of images can get different compression algorithms. During a variety of image coding algorithms based on wavelet transform, Shapiro's embedded zerotree wavelet EZW algorithm and the improved EZW algorithm for set partitioning in hierarchical trees SPIHT coding algorithm are widely recognized as two of the more successful methods. This thesis separately studies the two traditional algorithms,and proposes improvements based on the shortcomings of the two algorithms, and demonstrates that the improved algorithm is effective.The traditional embedded zerotree wavelet EZW algorithm adopts zero-tree algorithm ideas, successive approximation quantization and Z fonts scan order to guarantee the coding efficiency. However, there are still insufficient, such as a large energy loss of the low-frequency, few zero roots and large encoding, and so on. The improved algorithm in this thesis uses Z97 biorthogonal wavelet for image decomposition and reconstruction, because that Z97 has a good biorthogonal wavelet vanishing moments and smoothness, higher regular order, to guarantee good performance, which will help compression. In addition, the improved algorithm separately codes the image of the lowest frequency by DPCM coding to avoid the great loss, and ensure the quality of image restoration when in low bit rate. The strategy not to judge the important future generations of wavelet coefficients when subgraphs on the edge of the high-frequency coding, eliminate the coding redundancy greatly. It changes the Z shaped scan order, using different scaning order for different levels of images to help generate more zero roots. Simulation results show that, the PSNR peak signal to noise ratio of reconstructed image has improved to some extent of improved algorithm compared with the traditional EZW algorithm, mean square error is decreased, and encoding and decoding efficiency is improved.The traditional SPIHT algorithm uses a unique approach in a subset of the partition coefficient and the transmission of important information, to transfer coefficient implicitly ordering information when achieving large amplitude transmission coefficient at the same time priority. But the traditional SPIHT algorithm does not consider the characteristics of human vision fully and takes too much space and other issues. To address these issues, this thesis proposes improved algorithm. Improved algorithm still select the Z97 biorthogonal wavelet as wavelet basis, and adopts special handling in the threshold processing considering the human visual system HVS. Simulation results show that the noise ratio of the reconstructed image signal of the improved algorithm is higher than the traditional SPIHT algorithm, and the visual effect is better especially significant in the details of the image.Finally, this thesis implements parallel processing in the coding part of improved algorithm of EZW. The computation of the coding part of EZW is large to get less efficient serial processing code. This algorithm uses the inherent parallelism of the scanning process in improved algorithm of EZW, to implement parallel processing based on MPI, and experimental results show that, the parallel coding computation has achieved a higher efficiency, which improve the effectiveness of the parallel algorithm.