Support vector regression based image restoration

The point spread functions (PSF) responsible for degrading the observed images are very often not known. Hence, the image must be restored only from the available noisy blurred observation. This paper proposes two new image restoration algorithms, which are based on support vector regression (SVR). The first algorithm uses local variance and the second algorithm utilizes the concepts of fuzzy systems to counter blur in a given image. These algorithms significantly reduce the training time through their effective sample selection mechanisms. Experimental findings show that the proposed techniques deliver superior results for a variety of blurs and PSFs.