DCT coefficients recovery-based error concealment technique and its application to the MPEG-2 bit stream error

This paper presents a novel error concealment technique based on the discrete cosine transform (DCT) coefficients recovery and its application to the MPEG-2 bit stream error. Assuming a smoothness constraint on image intensity, an object function which describes the intersample variations at the boundaries of the lost block and the adjacent blocks is defined, and the corrupted DCT coefficients are recovered by solving a linear equation. Our approach can be regarded as a special case of Wang et al.'s (1991). However, we show that the linear equation in the proposed algorithm can be decomposed into four independent subequations, requiring much lower computational load and simpler hardware structure than existing algorithms, while providing adequate performances. To develop a generic error concealment (EC) system, the blocks corrupted by the random bit errors are identified by a multistage error detection algorithm. Thus, the proposed EC system can be applied to more realistic environments, such as concealment of random bit error in MPEG-2 bit stream. Computer simulation results show that the quality of a recovered image is significantly improved even at a bit error rate as high as 10/sup -5/.