The basic idea of identifying the direction of motion blurring by directional derivatives is that the original image is an isotropic first-order Markov random process. However

the real result are not good. The main reason is that the images do not meet this assumption strictly. Judging from the entire image

the shape of the objects

flat areas

and some other factors tend to weaken the physical premise

and cause inaccurate calculation result of the differential image. In this paper

we propose an optimization method. First

multiple feature blocks are extracted via the image local variance

and then the statistical results of the motion blur direction identification from every block are retrieved. The experimental results show that our method can still obtain good identification accuracy even when the error of the identification of traditional directional derivative method is big.