Blur Identification Based on Maxima Locations for Color Image Restoration

Blind deconvolution essentially involves the determination of the original image and the blurring function given only a degraded image. This paper proposes an approach to simultaneously estimate these two quantities by utilizing a reference point spread function (RPSF) as a learning basis in the regularization cycle. This is derived from the behavior of image maxima in the presence of blurs. The extraction of the RPSF is solely dependent on the image and it does not change with respect to the estimated values so a more stable learning process can be achieved. A modified cost function is also used such that the contours of piecewise non- Gaussian blurs are accounted for. This cost function is formulated for color images so that spurious color artifacts can be avoided. Experimental results show that better image quality is attainable with fewer iteration counts.