Lossless interpolation optimization for H.264 decoder

To improve compression capacity, motion compensation (MC) is utilized in H.264, which includes 6-tap FIR for 1/2-pel interpolation and 2-tap filter for 1/4-pel interpolation. MC uses up most decoding time, and is the expensive part of H.264 decoder. Some consumer electronics, such as smartphones, may not have enough processing capability to decode high definition video program in real time. MC must be simplified at first to reduce its computational cost for real time decoding. As for interpolation cost, 1/2-pel interpolation is the dominant one for it adopts larger tap filter and is more often used than the 1/4-pel interpolation. In this paper, we proposed an interpolation mechanism to adaptively adjust tap length of 1/2-pel interpolation filter considering image content's variation. For low frequency signal, there is strong correlation among adjacent pixels, so it is possible to use a short filter to replace original one, without introducing additional interpolation error. Therefore, we examine the frequency spectrum of image blocks and design an optimal filter at the criterion of the least interpolation distortion and tap length. We further adopted different simplification strategies for different kind picture interpolation, because the interpolation distortion in the reference pictures will be propagated into whole group of pictures and significantly reduce video quality. The experimental results show that the proposed algorithm can reduce more than 20% interpolation costs while with neglectable quality degradation.