In video communication bit-errors

block-errors and macroblock-errors often happen owing to channel noises. They seriously affect video quality and QoS. At the same time retransmission can't match the needs of real-time transmission(RTP) when finding errors. Up to now many methodshave been proposed for error concealment in spatial

temporal

frequent domain and their joint domain. They all tr out to conceal the error with simple algorithm to meet the demand of RTP. In this paper we proposed an error concealment method using the frequent information of error blocks' neighborhood to decide how to conceal the error. It employs Markov random field to describe the relationships between pixels and between blocks. Based on the description we get the neighborhood of pixels and blocks by which to restore the error pixels and error blocks. According to the characters of human vison sstem different methods are used to restore different frequent parts. The high frequent parts

i.e. the deges of image are restored first

then the lower parts. In a least-squear sense linear fitting and multi-directional extrapolation are used to obtain optimal concealment. The experiment results have proved its robust error processing ability and good error concealment performance.