International Journal of Advanced Research and Development

Authors
Mani Laxman Aiyar, Chang Y Choo

Abstract
Despite increased complexity sub-pixel motion estimation and compensation significantly outperforms integer motion estimation and compensation in MPEG4/H.264 Video Codec’s, since moving objects do not necessarily move by integer pixel locations between successive video frames. Typically, fractional pixel accuracy is obtained by means of bilinear interpolation producing a spatially blurred predicted signal. The motion estimation and compensation is improved in this paper by means of the filtering effect using a very effective spatial digital low pass FIR filter, which allows the motion vectors to be determined with higher levels of precision and accuracy than existing algorithmic implementations. The fractional pixel accuracy was achieved using a total of 112 8-tap digital FIR filter for one-eighth pixel precision, which includes half and quarter pixel accuracy. The design has been implemented on a 28nm TSMC process, with a speed of 1.101 GHz and it has achieved 2262 GOPS at this speed, outputting data at the rate of 1.8 Tera bits per second, for one-eighth pixel accuracy. Computational complexity, Memory & I/O Bandwidth has been reduced by inputting the Mean Square Error Map of the pixels to the Fractional Pixel Estimator and then searching in the sub-pixel grid.