文摘
In this paper, a content-adaptive method to generate side information at the block level is presented. First, motion compensated temporal interpolation (MCTI) algorithm is used between the reconstructed key frames at the decoder to acquire initial motion vectors. Second, the image is segmented and the edge of moving region is detected by obtained the residual frame between two consecutive key frames. Furthermore, hierarchical motion estimation (HME) and motion vector filter (MVF) are adopted for edge region and an adaptive motion vector filter (AMVF) is introduced in non-edge region to correct the false estimated motion vectors. The proposal is tested and compared with the results of the state-of-the-art DISCOVER codec and RD improvements on the set of test sequences are observed.