Abstract
In current video coding standards, the encoder exploits temporal redundancies within the video sequence by performing block-based motion compensated prediction. However, the motion estimation is only performed at the encoder, and the motion vectors have to be coded explicitly into the bit stream. Recent research has shown that the compression efficiency can be improved by also estimating the motion at the decoder. This paper gives a detailed description of a decoder-side motion estimation architecture which assumes temporal constant motion and compares the proposed motion compensation algorithm with an alternative interpolation method. The overall rate reduction for this approach is almost 8% compared to H.264/MPEG-4 Part 10 (AVC). Furthermore, an extensive comparison with the assumption of spatial constant motion, as used in decoder-side motion vector derivation, is given. A new combined approach of both algorithms is proposed that leads to 13% bit rate reduction on average.
Highlights
All existing video coding standards, such as MPEG-2 Part 2 or H.264/MPEG-4 Part 10 (AVC), are essentially based on similar structures: the encoder estimates motion between the current frame to be coded and already encoded reference frames to exploit temporal dependencies within the video sequence
The resulting motion vectors are used to calculate a prediction of the current frame by displacing the content of the reference frames
Since the interpolation is crucial for the performance of this approach, a detailed description with additional information on the motion estimation algorithm proposed in [6] is given in this paper
Summary
All existing video coding standards, such as MPEG-2 Part 2 or H.264/MPEG-4 Part 10 (AVC), are essentially based on similar structures: the encoder estimates motion between the current frame to be coded and already encoded reference frames to exploit temporal dependencies within the video sequence. In case DMVD is selected, no motion vector has to be coded, and only the mode has to be signalled Another approach, discussed in more detail in this paper, is decoder-side motion estimation (DSME, [3]). Since the interpolation is crucial for the performance of this approach, a detailed description with additional information on the motion estimation algorithm proposed in [6] is given in this paper. A performance comparison with the spatio-temporal autoregressive (STAR) frame rate upconversion proposed in [7] is made Both approaches, DMVD and DSME, try to reduce the rate used for transmitting the motion vectors, different assumptions on the motion are made as mentioned before. An extensive comparison of the new combined approach with the AVC reference as well as DSME and DMVD is given in that section.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
