Skip to Main Content
For downscale (spatial resolution reduction) video transcoding, previous work has shown that motion vectors in the input video can be reused so that the costly motion estimation process can be avoided. Since the derived motion vectors may not be perfectly aligned with the input ones, a drift compensation loop consisting of inverse transform and motion compensation (or transform domain motion compensation) modules is required. However, the motion compensation is always performed based on macroblocks in the full resolution, which makes the full-resolution drift compensation inevitable. This fact renders the drift compensation loop as the next computational bottleneck. Through analytical modeling of the transcoding architectures, we propose a new architecture that employees a submacroblock motion compensation method. This method enables the drift compensation loop to be performed entirely in the reduced resolution. In addition, the memory space required for the architecture is significantly reduced since the full-resolution frame buffers are no longer required. Extensive analysis and experiments show that the proposed system achieves further reduction in computing load comparing to prior systems with minimum quality degradation.
Circuits and Systems for Video Technology, IEEE Transactions on (Volume:15 , Issue: 10 )
Date of Publication: Oct. 2005