A WAVELET-BASED CONDITIONAL REPLENISHMENT COMPRESSION SYSTEM FOR SCALABLE LIVE VIDEO STREAMING

摘要

We consider the problem of encoding individual live video streams in a single streaming server for a large number of clients. Our streams utilize the wavelet transform, require reliable transport to ensure a defined state and exploit this state by transmitting only changed rectangular image blocks for particular clients. To reduce cost we compute the wavelet transform once for all clients and separate its creation from the entropy encoding done for groups of clients with identical state. We analyze the dependencies between image pixels and wavelet transform coefficients for the biorthogonal 5/3 and 9/7 filter to update the transform in a blockwise fashion. This leads to significant improvement of the execution time if the fraction of changed blocks falls below a certain threshold. We construct a coefficient change mask to identify those coefficients that must be encoded in order to transmit the changed image blocks for groups of clients with identical state. To entropy encode them we modify the Wavelet Difference Reduction (WDR) algorithm. This improves execution time up to a factor of four. We implemented the proposed compression system and present results for the execution time of the partial update of the wavelet transform and the rate-distortion performance.