Non-parametric Qualified Support Domain Of Blind Image Restoration Algorithm Research

In the process of image acquisition, the relative motion, atmospheric interference, lack of defocus or noise leads to the decline of obtained image quality. In order to reduce or eliminate the impact of this degradation and restore the true image as much as possible, the image restoration processing is very necessary. Image restoration techniques are always based on certain priori knowledge of the image degradation. In the conventional image recovery, the PSF (point spread function) is supposed to be known. But in actual image applications, the PSF is difficult to identify, only to reconstruct the true image from the observation scene; this kind of algorithm is called blind image restoration.In actual, noise generation is difficult to avoid. In this paper, the author introduces the application background and the basic method of blind image restoration, analyses the image degradation model and the ill-posed problem in the process of image restoration and pay more attention to non-negativity and support constraints recursive inverse filtering. We proposed an improved NAS-RIF algorithm based on the lifting scheme for blind image restoration and an iterative blind de-convolution image restoration algorithm based on adaptive selection of regularization parameter. In the former, we use the lifting wavelet de-noising technique to smooth the blurred image, and then in the each iteration of NAS-RIF algorithm process adopt the image segmentation to find out the precise support constraints. In the latter, the proposed ASPR method can obtain the regularization parameter adaptively according to the edge information of the image which guarantees the restored image is the best result in the total field. The simulation shows that the new method can get a good performance, even under the existing noise.