Fast Sub-pixel Search Control by using Neighbor Motion Vector in H.264

Abstract

We propose a novel fast sub-pixel search control algorithm for H.264 encoder by using neighbor motion vectors. This fast sub-pixel search control algorithm cuts down the motion estimation time by referring neighbor motion vectors and current integer motion vector. And this algorithm can improve the performance of motion estimation time more by merging other fast search algorithms with its independent discrimination.

Novel fractional pixel motion estimation algorithm using motion prediction and fast search pattern

“…1. Since most of MVs concentrate on (0, 0) and have the horizontal or vertical directivity [6], we firstly search up, down, left and right positions and then search diagonal positions. After that, SDSP is iterated around the minimum SAD position.…”

“…However many integer MVs are selected as final MV and many integer MVs are concentrated in the center position (0,0) of the search area [6]. In other word, there are many blocks having no motion movements.…”

An Efficient VLSI Architecture of Fractional Motion Estimation in H.264 for HDTV

“…However, only a few FME implementations have been discussed in spite of FME having a strong impact on the peak-signal-to-noise ratio (PSNR) and the amount of computation required for FME is even more than needed for IME. Several algorithms have been proposed to speed up the FME process, although they decrease the video quality to some extent, such as those based on early termination techniques [12] (average ΔPSNR= −0.02 dB, ΔBitrate=2.91%), search control by using neighbouring motion vectors [13] (ΔPSNR=−0.03 dB), size reduction of tap filters [14] (ΔPSNR=−0.003 dB) or reduction of search area [15] (ΔPSNR=−0.17 dB, ΔBitrate= 4.08%). Different FME implementations which make a trade-off among input bandwidth of reference pixels, hardware overhead and number of clock cycles for processing all MBs have been described.…”