Video And Image Restoration Based On Total Variation And Kernel Regression

Video restoration refers to the process of recovering the original sharp videofrom degraded video. It can extract the clear detail information of interested region,and reduce the negative influence of the outer noise and improper operation. So it hasbroad prospect in military, transportation management and social management. Videorestoration is a challenging task because of its large amount of data, and that willcause the time complexity of the video restoration process is high. In this paper, wewill study the video image restoration; mainly explore it with the follow two aspects:First, according to the character of the space-time discrete gradient preserve thespatial-temporal relationship of nearby frames in the process of total variationrestoration, this paper proposed a total variation video restoration based on3Dweighted generalized difference. Different from kinds of the current video restorationmethod, it is not just take video restoration as a series of image restoration problem,but regard the video sequence as a space-time volume, apply total space-time totalvariation regularization to enhance the smoothness on space and time, ensure thecontinuity on space and time. And use the3D generalized difference in the process ofcomputing the total variation regular term to obtain richer constraints about theoriginal video. The proposed method transform the non constraint problem toconstraint one. And employ the augmented Lagrangian method to solve the constraintproblem, the alternating direction method to get the solution of sub-problems whichincluding image restoration, auxiliary variable resolution and updating of Lagrangianmultiplier. The proposed algorithm can restore the blurred video quickly, and obtain adesirable signal to noise ratio.Second, We proposed a geometric local adaptive sharpening algorithm for thefeature of the image took is weakly blurred and with strong noise because of shotenvironment and the photographer’s mistake in the real life, the proposed methodapply steering kernel regression to get the structure information, and use it to obtainthe local metric of analysis window, finally determine the degree of sharpening of theanalysis window according to local metric. It is capable of capturing local imagestructure and sharpness and adjusting sharpening accordingly so that it effectivelycombines de-noising and sharpening together without either noise magnification orover-sharpening artifacts. It also uses structure information from the luminance channel to remove artifacts in the chrominance channels.