Resizing H.264/AVC frames in the compressed domain.

摘要

One of the most popular standards to encode video is H.264/Advanced Video Coding(AVC), which provides a compression gain of 1.5× to 2.0× over previous standards such as MPEG2 or H.263+. Resizing H.264/AVC video for storage or transmission is often required because of imposed constraints dictated by network bandwidth or device capabilities. Usually, systems resize H.264/AVC video in the spatial domain. However, it would be more attractive to resize the video directly in the compressed format, which would avoid high computational overhead associated with decompression and compression operations. This thesis proposed methods to resize H.264/AVC frames that leveraged contributions made in image and video resizing in the DCT domain. We also proposed a new efficient DCT resizing algorithm that exploits fixed-point integer transforms. This approach has shown to reduce the computational cost by more than 60% with small dB loss when comparing similar algorithms using floating-point transforms.