Fine Estimation of Blur Parmeters for Image Restoration

In this paper we consider the problem of image blurring as a result of relative motion between image sensory and the scene. We propose an approach which is able to properly estimate orientation and extent of blur. A fine estimation of the blur parameters is a key point for restoration of blurred documents in photographic images. In this method we estimate the orientation of the blurred image using the main smear lines in the Fourier spectrum of the image. A fine segmentation of the main smear lines is performed using EM algorithm. In order to estimate the blur extent we first de-correlate the blurred image along the direction perpendicular to the blur orientation. A differential filter is proposed for this stage. The blur extent is then estimated from the process of minimizing the autocorrelation of image profiles along the blur orientation. The estimation of the extent is refined by statistical analysing the extracted extents along a number of image profiles. The experimental results shows the superiority of the proposed method to the existing methods for the estimation of blur parameters.