Adaptive picture slicing for distortion-based classification of video packets

We present a new algorithm for the dynamic identification of perceptually important regions in pictures of a video sequence. For each macroblock, the distortion that its loss would cause at the decoder is computed. High-distortion macroblocks are then grouped together into slices to be protected with forward error correction or to be sent as "premium" packets. We developed the approach for the case of video sequences encoded with the ISO MPEG-2 video coding standard. We then applied the algorithm to classify video packets within a 1-bit differentiated services architecture: slices were grouped either into premium packets, to be sent on a "virtual wire", or into regular packets, to be sent as best-effort traffic. In packet losses, the proposed distortion-based classification scheme outperforms source-transparent packet-marking techniques and provides substantially higher PSNR values than the regular best-effort case sending as little as 10% of the packets as premium traffic. Video samples are available at http://multimedia.polito.it/mmsp2001/