Multi-priori Models Collaboration For Image Restoration

In today’s information age, information acquisition, processing and utilization hasbecome increasingly important. The information obtained in the human visualaccounted for70%, so the image acquisition, processing and utilization is veryimportant. However, since the image forming apparatus, transfer, human and otherfactors may cause the inevitable degradation of the pictures. Therefore, the imagerestoration problem has important practical significance. But also in the field ofdigital image processing of image restoration is one of the most fundamental issues.Image restoration problem is usually given in a degraded image, hoping to get atrue original image. Throughout the degradation process can be a degradation modelto simulate the model input is a real image, the output is a degraded image, imagerestoration problem is often a counter. When the entire degradation procedure is notvery clear, people tend to take the blind restoration techniques, which may estimatethe degradation process. But when the degradation case is clear, the accurate imagerestoration will need to rely on the accuracy of the degenerative process. However,relying solely on the information degradation processes, restoration after effect isoften relatively poor. To improve the recovery effect, we need to use a prioriknowledge of image, the a priori knowledge to image restoration, constrainedrecovery solution space.Many low-level vision algorithms assume a prior probability over images andmany priori models have been applied for the image restoration. But no studies haveshown that a model has been proposed can grab all of the statistical information. Sobuilding an accurate image priori model is still an unsolved problem. A more precisea priori model can be used to solve these basic image processing problems, but alsocan guide people to discover the limits of these image restoration problems. Such asfor image denoising problem, the use of a priori knowledge can be more precisestudy of image denoising limit results. On the one hand people continue to studynew image prior model, on the other hand people are trying to combine multipleprior model used in image restoration and other issues. However, the current way ofcombining multiple prior models focused on the regularization method, and the apriori knowledge as a regular entry into the objective function to obtain the desiredsolution, which is the restoration of the image quality is better. This regularizationmethod, not only the numerical complex and need to construct the objective functionin advance, this is not a common framework. Motivated by the idea of modelcombination in machine learning, we put a priori model of each individual orindependent algorithm as a "classifier", according to each individual’s recovery results, combined with the recovery of their final result. In order to test theeffectiveness of proposed framework, the paper five classic algorithm experimentswere carried out and did some comparative experiments, the results show the effectof combined algorithm than a single model or better results.