Motion Estimation by Segmentation and Prediction of Direction

In this paper we show that the motion direction of a block is correlated with those of the neighboring blocks of the current frame and those of the previous frame. We design a search algorithm which is called SAPDS (segmentation assisted predicted direction search). This algorithm uses the mentioned correlation and hence predicts a direction for the movement of the block. Therefore, based on this prediction an initial search point is found. Around this initial search point a local search is performed to obtain the best matched block. For the hardware implementation of the algorithm we developed a circuit to perform the direction prediction part of the routine. This circuit uses only five 2-input gates and two 5-bit comparators. Comparing our algorithm with the existing routines we were able to reduce the required hardware as well as computational complexity. This is shown by simulating a large number of standard video image sequences.