A unified MC-DCT framework for reducing the computational complexity of frequency-domain transcoders for MPEG-2 video

In order to transmit pre-encoded digital video over heterogeneous networks, it becomes necessary to employ transcoding techniques that convert pre-encoded video streams into streams having different bit rates and quality. Video transcoders save computational overheads incurred in the IDCT and DCT operations, by performing motion compensation (MC) in the frequency (block-DCT) domain (MC-DCT). Most of the computational complexity of the frequency-domain video transcoder comes from the MC-DCT operation. MC-DCT involves pre- and post-multiplication of 8 × 8 DCT blocks with 8 × 8 matrices derived from the motion vectors of that block. The specific contribution of this work is towards developing a unified framework for handling all MC-DCT modes for MPEG-2 video input to a frequency-domain video transcoder. Use of this framework can lead to a potential saving of up to forty-eight 8 × 8 block multiplications and sixteen 8 × 8 block additions per reconstructed 16 × 16 macroblock.