Image Blind Deconvolution Method Based On Gradient-Projection Algorithm

Image is a kind of information acquired through sensors, which can reflect the true outside scenery. Sometimes, we need to convert it into visual information, and then record, store, and reproduce it. In the process of image acquisition, transmission, and display, the presence of different levels blurring and noise is unavoidable because of the restriction of external environmental condition. This can result in image degradation. In the case of little or no prior knowledge of images and imaging systems acquired, we need a technique to extract the degradation information from the observed images and estimate the true images, which is called blind image restoration. Now, blind image restoration is a useful technique to enhance quality of degraded images. It can be applied to some fields such as remote sensing images, medical images, video surveillance, and computer vision and so on.At present, researchers are seeking and exploring the image blind restoration mathematical model, the theoretical framework and efficient numerical method. In this thesis, we have researched blind image restoration technology deeply, and improved blind image restoration based on gradient-projection algorithm. The results of our experiment show that it has attained good effect in denoising and deblurring. The main contribution of the thesis as follows:(1) We have improved the denoising model of Yunmei Chen by adopting an adaptive method to choose parameters, and then carried on numerical compute in detail and denoising simulation experiment using the proposed denoising models. The improved method is effective learned from the theory of Yunmei Chen's total variation denoising models. It can distinguish smooth areas and edges by setting a threshold. On edges, denosing process is similar to total variation denoising which can be effective protected; on smooth areas, denoising process approximate to anisotropic diffuse of L2 norm denoising. What's more, we obtain adaptive fidelity term by defining local variance. It means imposing a spatially varying variance constraint on noising images. The experiment results show that the improved algorithm not merely can increase SNR of images, protect image edges but also can de-noise in smooth areas and better retention of image texture details.(2) We have improved the blind image restoration based on gradient-projection algorithm. Firstly, considering the shift and distortion frames amount images, we carry on motion estimation before gray image restoration. Then, we combine the adaptive fidelity total variation denoising method with blind image restoration based on gradient-projection algorithm, de-noise the estimated image and PSF respectively. After that we have realized multi-frame gray image restoration. At last, we have extended out algorithm from to gray image restoration to color image restoration, and realized color image restoration.(3) Several groups of blind restoration experiments for the degraded images had been conducted using our image blind recovery algorithm. Experimental results show that this algorithm can do well in denoising and deblurring, and recover image detail and edge. The simulated experiment results on color image also show that our algorithm has good convergence and stability.