Adaptive intra prediction filtering (AIPF)

Abstract

Intra prediction is an important coding tool to exploit correlation within one picture in image and video compression. Before the ultimate intra prediction values are generated for current block along oblique angles, a fixed low-pass filtering with 3-tap filter (1, 2, 1) will be applied to the three prediction pixel values to avoid the effect of pulse noise. In this paper, we use adaptive intra prediction filter (AIPF) to replace the fixed filter to minimize the prediction errors. To get the adaptive filter coefficients in an on-line way with an acceptable accuracy and no coding overhead, we combine it with template matching (TM). After the best estimation of current block through template matching, the optimal adaptive filter coefficients are calculated with least-square optimization through considering the best estimation as `current' block. The adaptive filter is used to obtain intra prediction values instead of the 3-tap fixed low-pass filter. Experimental results show that the AIPF can get a stable coding gain on all test sequences, and reduce the bit-rate by up to 1.74% comparing with that using only TM.