Fast motion estimation by early search termination
Shaifali Madan Arora
In this paper, a refined dynamic threshold estimation technique for zero motion prejudgment (ZMP) is proposed. In most video sequences, especially containing slow motion, large number of stationary blocks are present. Early and accurate determination of these stationary blocks reduces the computational burden without compromising the video quality in any motion estimation (ME) algorithm. Stationary blocks can be determined by comparing the block distortion with a predetermined threshold which varies depending upon the motion activity of a block. Very large or small value of this threshold results in false determination of stationary or moving blocks. So there is a need of a threshold estimation technique that follows the distortion pattern of stationary blocks in a frame. A two level dynamic threshold estimation technique is proposed in this manuscript which is based upon the distortion measure of previous stationary blocks. The performance of the proposed technique is checked by implementing ZMP before ME process in ARPS algorithm and is able to save computations by 70-95 % for slow motion sequences with very small degradation in PSNR and SSIM. Further in comparison with other fixed and dynamic threshold estimation techniques, the proposed ZMP estimation improves the PSNR and reduces the decision errors.