210 likes | 368 Views
Compressed Video Downscaling. Zhijun Lei MCRLab, SITE University of Ottawa. Outline. Introduction Problem Statement Algorithms System Architecture Ongoing Implementation Problems and Future Works. H.263 Video Coding Standard. Based on ITU-T recommendation H.261;
E N D
Compressed Video Downscaling Zhijun Lei MCRLab, SITE University of Ottawa
Outline • Introduction • Problem Statement • Algorithms • System Architecture • Ongoing Implementation • Problems and Future Works
H.263 Video Coding Standard • Based on ITU-T recommendation H.261; • Very low bitrate compression. (below 64kbits/s); • Input video format: Sub-QCIF, QCIF, CIF, 4CIF, and 16CIF. CIF (352288), QCIF (176 144) • Encoder & Decoder source code available in C. (TMN V3.2)
Problem Statement • Dynamically scale down the compressed video sequence for small display device. • VOD applications. • Quality and bit rate. • Computation cost.
Algorithm for downscaling P-frame • From incoming compressed video, we get: • Motion Vectors ( 1 MV for every Macroblock) • Predication difference DCT coefficients. • Basic idea: • Resample motion vectors to bypass Motion Estimation, because ME contributes 60% ~ 80% computation cost. • Motion compensate the new P-frame according to new motion vectors. • Calculate the new predication difference and encode it.
Motion Vector Resampling • Method 1: Average • Method 2: Median
Motion Vector Resampling • Method 3: Adaptive Motion Vector Resampling where denotes the motion vector of block i in the original M N video and Ai denotes the activity measurement of residual block i (in the original video).
Motion Vector Resampling • Method 4: Adaptive Motion Estimation
Implementation • Pixel domain transcoder as test bed. • H.263 (CIF) -> H.263 (QCIF). • I-frame processing • Motion Resampling algorithms • P frame processing.
I-frame Result CIF (352 X 288) QCIF (176 X 144) Foreman.263 (100 frames)
P-Frame Result CIF (352 X 288) QCIF (176 X 144)
Problems & Future Work • Problems • P frame Motion Vectors Resampling Algorithms. • Bandwidth Reduction. • Total computation complexity. • Future Works • Finish the implementation • Evaluation and analysis. • Motion Vector Refinement.