Multiple Descriptions , Error Concealment , and Re ned Descriptions for Image

{ This paper demonstrates the conservative nature of a multiple description approach to image and video coding, by showing that error con-cealment of a sub-optimally partitioned coded image stream yields comparable results under realistic conditions. Rather than use the non-orthogonal transforms of multiple description coding, which result in excess rate when both streams are received, a standard DCT-coded output stream is partitioned to give two descriptions of the source. It is shown that, by using a combination of DC averaging and maximal smoothing to conceal errors, a better performance than multiple descriptions is achieved for block error probabilities up to approximately 0.05. If a feedback loop between the decoder and encoder is added so that the encoder conditionally sends reenement data, this range is greatly extended.