| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 94 === Most video coding standards use block-based discrete cosine transform (DCT)and motion-compensated prediction (MCP) to reduce both spatial redundancy andtemporal. For very low bit rate coding, coarse-quantized DCT coefficients suffer discontinuities at block boundaries, namely blocking artifact, after reverse transform. Furthermore, motion-compensated prediction propagates blocking artifacts to inner-block regions, which makes artifact detection difficult.
Blocking artifacts caused by motion-compensated prediction can be solved by a loop-filter, which is integrated to the codec and filtered frames are used as reference frames. The main drawback of loop-filter is its in-flexibility, i.e., users cannot enable/disable filtering freely but restricted by the received bitstream. In this study, a hybrid deblocking filter for H.264 video is proposed.
To estimate inner-block discontinuities, a motion-compensated based approach is proposed. A map of detected blocking artifacts within each video frame is stored. Combining motion vectors and blocking artifact maps of previous frames, possible locations where blocking artifacts may occur can be estimated. Cooperating with oriented blocking artifact detecting and filtering, we successfully realize loop-filter performance with a post-filter. The proposed approach can also used in other DCT/MCP based codecs.
|
|---|
