Research On Image Restoration Algorithms For Mixed Degradation

The image restoration is a process that recovering the original sharp image fromdegradation image. Images are likely to be noised under many conditions, such as theinterference of ambient during transmission, degradation in the image acquisitionprocess or other kind of image degradation caused by device defection. Actually, thegeneral image degradations are image blur, noise, geometric distortion, mosaicphenomenon and resolution decline. Degradation of the image makes the imagedifficult to understand in the process of image processing, brings great difficulties tomany image processing areas, such as pattern recognition.Motion blur mainly is caused by relative motion between the imaging system andthe target. In the study of motion-blurred image restoration, the point spread function(PSF) estimation of motion blurred image and restoration algorithm is a hot issue inrecent years. The main method of motion blur restoration bases on convolution modelor transforms to frequency domain to solve problem. In this paper, in the case ofknown point spread function, the projection on convex set method applies to imagerestoration with product model. Based on the principle of two-dimensionalconvolution calculation, this paper transforms the convolution model to a productmodel, supplies blur kernel matrix transformation formulas. Application of the threetraditional image restoration methods: inverse filtering, Wiener filtering, RL filteringmethod, and compared with POCS method. Inverse filtering method is sensitive tonoise, recovery results were affected by noise largely. RL image restoration method isbetter, but as the number of iterations increase, the noise is amplified. Wiener filtermethod is a kind of the image frequency domain processing method, there is a certainnoise suppression effect. POCS method can recover the edges and details of the image,after iteration the image can be restored well in no noise situation, but can’t removethe image noise. The advantage of POCS method is it can be a data consistent item tobe added a de-noising model, so motion blur method could suppress noise. In thispaper, minimization total variation POCS method (TV-POCS) is applied to removenoise with TV item. TV-POCS method is based on POCS, and minimization the totalvariation as the optimization objective function with data consistent item and non-negative constraints item. Experiments show the effect of noise suppression.For the mixed noise in degradation process, this paper is on the basis of the ROFmodel, then introduces a data constrains item in " salt and pepper " noise point, andprovides a constrained ROF model for mixing noise removal. The constraints of the newmodel ensure the data consistency of mixed noisy image on the uncorrupted pixels, reducinglosing of the details of image information. The results show that the proposed algorithm has muchbetter performance than adaptive median filter and unconstrained TV-L1de-noising model. It iscapable of smoothing noise as well as retaining the details of the images.