Motion estimation for video resolution down-conversion and frame type conversion
by Wong Wai Chuen
M.Phil. Electrical and Electronic Engineering
xi, 92 leaves : ill. (some col.) ; 30 cm
In this thesis, we investigate three main problems: (I) fast motion estimation algorithms associated with resolution conversion of video, (2) fast motion estimation associated with frame type conversion and (3) motion vector recovery for error concealment of compressed video....[ Read more ]
In this thesis, we investigate three main problems: (I) fast motion estimation algorithms associated with resolution conversion of video, (2) fast motion estimation associated with frame type conversion and (3) motion vector recovery for error concealment of compressed video.
In many video applications, a compressed video sequence needs to be converted to a lower-resolution compressed video. In such applications, one typically needs to decompress the original sequence, down-sample each frame, and re-compress the down-sampled sequence. The challenge is that full re-compression has very high complexity due to the computational intensive motion estimation. In this thesis, we propose three fast motion estimation algorithms (PME, MPME and ME-SVF) for eficient re-compression using the motion vectors in the original compressed video. Our simulation results suggest the proposed fast algorithms can achieve a significantly higher quality than existing fast algorithms with low additional complexity. The proposed algorithms can also provide diRerent trade-08 between complexity and accuracy.
In some application such as transcoding video from IPPP format to IBPBP or IBBP format, there are needs to change the frame types of many frames. Again full re-compression can be very complex due to the computational intensive motion estimation. In this thesis, we propose two fast motion estimation algorithms (P2P, P2PS) for the P-frame to P-frame conversion, and two algorithms (P2B, P2BS) for the P-frame to B-frame conversion. Our simulation results suggest that the proposed algorithms can give very high quality results with much lower complexity than full exhaustive search.
Moreover, delivery of digital video through different kinds of packetized network is very common. During the transmission, motion vectors might be lost or erroneous after going through some noisy channel, and the effect can propagate as the video frames are usually coded differentially. In this thesis, we propose two algorithms (SSPS and SSPS-IO) for the recovery of lost macroblocks and the perjormance of the proposed algorithms is found to be significantly better and more robust than the existing algorithms.