University of Ottawa - Carleton University
Ottawa-Carleton Institute for Computer Science (OCICS) Presentation
|
November 23, 2012 @ 10:00a.m.
Halving H.264/AVC Frames in the Compressed Domain
|
Speaker: James McAvoy
Location: 3101 CB (Canal Building) |
ABSTRACT
Internet based delivery of video content over wired or wireless networks is
expanding. These content providers need to adapt the precoded or real-time
video to meet a broad range of end-user with different bandwidths and device
capabilities. One strategy to transmit video over bandwidth constraint networks
is to reduce the video's spatial resolution and transmit the low-resolution
version of the video as trade off to bitrate. The straight forward approach of
inverse transform, spatial domain resizing and forward transform to the
required resolution is undesirable due to inherent high computational cost.
Developing fast algorithms to resize video frames in the compressed domain will
be worthwhile. A popular video compression standard to encode video is H.264
Advance Video Coding (AVC). Early research in the area of image halving in the
Discrete Transform Domain (DCT) assumed the compressed picture is coded using a
popular 8x8 DCT block framework. However, H.264/AVC employs 4x4 integer
transform, an approximate form of the DCT, to create blocks of 4x4 DCT
coefficients. Also, researchers developed their algorithms to resize
images or frames in the DCT domain. Hence, inverse quantization was
applied to the 8x8 DCT coefficients before resizing operation. In H.264/AVC,
transform process includes quantization and scaling, making inverse quantization of DCT coefficients difficult. The presentation will show that these DCT resize algorithms can be modified to work on H.264/AVC frames. In experiments, these algorithms generated images with similar or greater quality at lower computational cost than comparable operations in the spatial domain. |
| Return to Schedule |
|
