Research On Image Restoration Method Based On Prior Constraint

Image information is an important source for human to know the world. However, due to the bad conditions in the process of imaging or transferring, the quality of image could be degraded, thus deteriorating its use and post-processing. How to restore a clear, content-rich image from degraded one is a problem of focus, which has an enormous practical value and theory research meaning in economy and national defence. This is exactly what image restoration does.Image restoration plays a significant part in image processing. It is an ill-posed problem. Some constraint conditions must be introduced to algorithms and transform image restoration to be in well-posed. This thesis used prior constraint model as a thread, mainly targeting at image denoising and image super-resolution. The innovations of this thesis are as follow:1) Done research on image super-resolution restoration methods based on bilateral total variation prior constraint condition. Aiming at the image frame selection problem in practical implementation, selection rules for reference frame and image frame sets are proposed; an optimize image sequence framework step-by-step for image super-resolution is designed, which preliminary solved the robust problem of super-resolution restoration in the existence of large range movement.2) Done research on the problem of robustness of image super-resolution restoration; taken localized motion outliers of low-resolution image sequence into account; using layer information in initialization, an image super-resolution restoration method based on multi-layer model is proposed, which preliminary solved the problem of super-resolution restoration in the presence of localized motion outliers.3) Studied image restoration methods based on Fields of Experts (FoE). Firstly, according to iterative inference of traditional FoE, it proposed an automatic stopping criterion in FoE image denoising, which improved the denoising effect and overcomed the condition that traditional FoE must known the noise variance a priori. It extended FoE denoising method to real image denosing tasks well; Secondly, it analyzed the insufficiencies of traditional FoE, and then proposed an edge-preserving FoE, obtained better edge-preserving results than traditional FoE; At last, it introduced high-order Markov Random Fields (MRF) prior into image super-resolution, proposed a super-resolution restoration method based on FoE prior, extended the model expression ability of super-resolution restortation methods based on traditional low-order MRF, improved the effect of image super-resolution restoration.4) Done research on image restoration methods based on non-local prior constraint. It proposed a fast Nonlocal-means (NL-means) image denoising method with projection. This method used the fast and recursive computation of Walsh Hadamard kernel, projected image blocks onto the space of a set of complete Walsh Hadamard kernels, then used efficient energy packing feature of Walsh Hadamrd projection and rejection strategy to fast discard those impossible image blocks in the image blocks matching procedure of Nonlocal-means image denoising method. Experimental results show that the proposed method can accelerate the NL-means algorithm with no image quality degradation.