Research And Improvement Of Several Image Encoding Methods

Fractal image compression (FIC) is developed in the last 10 years and takes advantage of self-similarities which exist in the images commonly. There is a certain kind of self-similarity in natural images. Fractal image compression can be achieved so long as a group of affine transformation whose fixed point is the approximation of the original image is found. FIC realize high compression ratio because the code consists of parameters of the iterated function systems instead of the pixels of the image. FIC became a hot research in the field of image compression encoding.Based on fractal and wavelet theory, the following works are done in order to improve the compression ratio and reduce the huge encoding time of fractal image compression and improved the image quality of wavelet image compression:Firstly, this paper proposes a fractal image encoding algorithm based on matching error threshold. First the authors set up two kick-out conditions to reduce the capacity of the codebook, and then set up a matching threshold when searching the best matching blocks, which can shorten its runtime greatly. Meanwhile, the authors discard the isometric transformations that mentioned in most literatures, because the usage of the isometric transformations only increase the computational complexity, the same or even better reconstructed image can achieve through reducing the sliding step of producing domain blocks. Experimental results indicate that the proposed algorithm can both shorten the encoding time greatly and achieve the same or better reconstructed image quality as compared with the basic fractal encoding algorithm with full search.Secondly, we have presented a fractal image coding algorithm based quadtree partition and variance sorting. Initially the original image is partitioned into a set of range blocks with size of 32×32 using quadtree scheme. For each range block with size larger than 4×4, we use no search scheme to compute its collage error. If the collage error is unacceptable, i.e. E(R,D) > T, then we partition this range block into four identical square subblocks. If the size of the subblocks is larger than 4×4, then we also use no search scheme to compute its collage error, else we search the best matched domain block from the domain pool. Experimental results show the proposed algorithm is a better choice than some other algorithms.Finally, we introduce a popular image coding method based wavelet transform, and depict the SPIHT (set partitioning in hierachical tree) algorithm in detail. At the same time, we propose a improved SPIHT algorithm. Be different with SPIHT algorithm, the proposed scheme always set LIP and LIS with the fixed pixels, which make all the coefficients be scanned. Expermental results show that the improved SPIHT can exceed the traditional SPIHT when the bit rate is less than 1.