Abstract: Due to the various limitations in real restoration process, it is difficult to get the image blur
mode or point spread function (PSF). A new method for image deblurring is proposed in this paper. At
first, the proposed deblurring method uses the different of Gaussian operator (DoG) to detect the
counters of blur image. Then the information of transition region of original image can be predicted
according to the contours of blur image. Then the objective function is established according to the
original image, transition region, and point spread function. In order to overcome the influence of
noise, the nonnegative penalty term and space correlation penalty term with anisotropic features are
added in objective function, and the PSF is solved using the minimization method of hysteresis
iteration. Finally, the clear image can be gained by the existing the non-blind image restoration
methods. Experimental results show that the proposed method can effectively restore the blur images
caused by various factors. It does not need to know the image blur model.

Key words: image deblurring, clearness, transition region, point spread function