Phase correlation method to the optical flow using neuronal networks

The motion estimation problem in image sequences is one of the most important tasks in computer vision. Thus, many methods were proposed to resolve this problem, but no universal method has been developed to determine the motion in all the situations and for all the types of objects in motion. In this paper, we propose to combine the advantages of neural methods in particular EMAN method and the frequency phase correlation method. The new method is consider as a connectionist neural method of motion estimation with discontinuities. The proposed method consists of two phases, the first one; the most likely motion in each pixel is estimated by exploiting the principle of Self-Organizing Maps of Kohonan with the algorithm of winner takes all. The second phase is a regularization phase of the displacement field with consideration of discontinuities. The new method is tested on the two real sequences.