A Study Of Deblocking Algorithms For Highly Compressed Images Based On Discrete Cosine Transform Domain

The block-based discrete cosine transform (BDCT) is widely used by the standards of image and video compression for its high compression rate and low computational cost. But it often causes heavy blocking artifacts near the block boundaries at high compression rate for the quantization of DCT coefficients, which could impact visual effect of the decoded image. So it is extremely meaningful to research the algorithms of deblocking artifacts for highly compressed images based on BDCT.Firstly this thesis analyzes the causes of blocking artifacts in JPEG images, and introduces several common deblocking algorithms. Then, proceed from the correlation between adjacent macro blocks in JPEG images, this thesis proposes a new algorithm based on c-means clustering method to reduce blocking artifacts. The proposed algorithm is based on a image database which contains a lot of images, after self-defined macro blocks are created from the image database, several different characteristic macro block set are obtained by c-means clustering method. Then, this algorithm classifies each macro block of the processing JPEG image into one matching macro block set, then estimates the image noise and achieves the effect of reduce blocking artifacts, finally the blocks which contains strong texture near its edge is filtered. Then, the proposed algorithm is programmed, the experimental result shows that this proposed algorithm is effective and has certain superiority by comparison with several other algorithms. Next, experiments based on different compression factor of compressed images or different size of the image database while using the same compression factor are studied respectively, and the PSNR performance of the proposed algorithm is summarized while using different experimental parameters. At last, the BWP parameter is introduced to explain the performance of the proposed algorithm.In addition, this thesis also proposes a new example-based deblocking algorithm. This algorithm extracts pairs of pixels near the block boundaries from the image database which contains one original image and the corresponding JPEG image, then clusters the extracted pixels by c-means clustering method, and through computing many noise coefficient matrixes are generated. Next, this proposed algorithm estimates the image noise and achieve the effect of reduce blocking artifacts by the generated matrixes, finally the blocks which contains strong texture near its edge is filtered. Then, this proposed algorithm is programmed, and the experimental result shows that this proposed algorithm is effective and has certain superiority by comparison with several other algorithms.