With the increasing development of computer technology, digital image is applied in a lot of areas as one of important media of modern information communication system because it is visualized, vivid, understandable and rich of information. However, image degradation resulting from imperfection of real digital image acquirement system leads to loss of image information, which hampers the understanding or analysis of human visual organ or system. The main goal of digital image restoration is to eliminate the degradation happened in the digital image acquirement system as much as possible and to restore and approach theoriginal real image. This thesis focus on research of several problems related to digital image restoration. Related areas are deinterlacing, supper-resolution and compression image post-processing. In this thesis, my main work and contributions are in the following. 1 .We propose fuzzy edge-based line averaging algorithm and local edge-basedline averaging algorithm. The former is based on fuzzy logic and determinethe edge direction in a fuzzy way. The latter judge the direction according to the local statistics. Compared with previous algorithm, proposed methods can obtain smooth estimation taking advantage of local area information. 2. We propose an image supper resolution algorithm in compression domain. The compression process is modeled as an additive and spacial related Gaussian noise and the image prior is currently popular Field of Expert model. To further enhance the edge area, total variation model is also use there. 3. We proposed two rules, mutuality rule and potential propagation rule of similar patches to accelerate collaborative filtering. To be specific, the search area of the reference patch is confined in the latter half and then in the second order neighborhood of reference block, which reduces most computation of this step. What'more, discrete cosine transform(DCT) and inverse DCT(IDCT) are proposed to be taken in blocks instead of in groups, which reduces computational load by 15/16. 4. We apply collaborative filtering to post-processing of compressed image. We deal with the compression noise in collaborative filtering frame by an empirical relation which links the noise level to quantization step. The whole algorithm only takes a dozen of seconds and outperforms H.264 intra-frame compression deblocking filtering about 0.3dB in PSNR. In a word, in this thesis, we developed some new technologies to digital i...
修改评论