Video compression with output traffic conforming to leaky-bucket network access control

This paper investigates how video source data should be compressed to conform to the leaky-bucket (LB) specification for transmission over a communications network. The buffer occupancy of the compressed data is used to effect compression of future video frames. Two general strategies for using the tokens in the LB are investigated. The greedy transmission strategy will send out a cell whenever a token is available-that is, a token will not be saved while there is a data backlog in the buffer. We show a general way to modify any existing constant bit rate compression (or rate-control) scheme for a corresponding LB compression scheme using the concept of a virtual buffer. The non-greedy transmission strategy uses “extra” tokens only when buffer overflow is impending and saves up tokens for future usage when underflow is about to occur. The non-greedy strategy is generally better than the greedy strategy. The overall conclusion is that tokens should be used only under the “emergency” situation when overflow occurs and not during the normal situation just to smooth image quality from time slot to time slot