ABSTRACT
Error concealment belongs to the general problem of image recovery or restoration. However, because errors are incurred at the compressed bit level, the resulting error patterns in the pixel domain are very peculiar, and special mea sures are usually needed to handle such errors. The use of prediction and variablelength coding (VLC), although leading to high coding efficiency, makes the compressed video stream extremely sensitive to transmission errors. On the one hand, with predictive coding, the reconstruction error in a single sample will af fect all the samples that are directly or indirectly predicted from it. On the other hand, the use of VLC makes it impossible to decode received bits following a single-bit error until a special synchronization codeword is encountered. To contain the error effect, various measures can be taken in the encoder to make the com pressed stream more error resilient, at the expense of a certain degree of coding gain and/or complexity. A simple measure is to add synchronization codewords periodically, so that a bit error will have only limited effect. Upon detection of a synchronization codeword, the decoding process can be restored. Another mea sure is to limit the extent of prediction, both spatially and temporally, so that the effect of a transmission error will be confined within a small spatial/temporal seg ment. More substantial coverage on error-resilient coding is provided in Chapter 8. Here we assume that synchronization codewords are inserted periodically within a picture and that the prediction loop is periodically reset-for example, by means of an I frame-so that a bit error or packet loss will cause damage to only a confined region in a picture, usually several rows of macroblocks.
