Abstract: To accurately estimate direction of motion blurred images, in the part of theoretical derivation, the relationships among motion blur direction, image size and direction of parallel dark stripes in the spectrum images are derived based on knowledge of definite integral, Fourier transform and the Sinc function. In the part of algorithm optimization, the principle and deficiency of Radon transform, Gabor transform and spectrum sub–blocks method are analyzed. And then the improved algorithm based on edge detection of spectrum and Radon transform is presented. In the part of numerical experiments, by use of Matlab programs, comparison results of the three methods proved the accuracy of our improved method is the highest, which shows our method is more suitable for estimating direction of motion blurred images.

Key words: direction estimation of motion blurred images, edge detection of spectrum images, Radon transform, Gabor transform, spectrum sub-blocks method