A Hybrid Error Concealment Technique for H.264/AVC Based on Boundary Distortion Estimation

Abstract

Video compression technologies have been extensively studied in recent years. The basic concept of video compression is to reduce the amount of bits for video representation by exploiting spatial and temporal correlations in image sequences. In recent years, H.264/AVC (Advanced Video Coding) is the state-of-the-art video coding standard established by ITU-T Video Coding Experts Group and ISO/IEC Moving Pictures Experts Group. H.264/AVC provides a better compression efficiency and visual quality than prior standards, owing to it adopts some unique techniques to reduce the redundant information, such as multiple reference frames, variable block size, quarter-sample-accurate motion compensation, etc. In H.264/AVC encoder, integer DCT procedure transforms residual data into the frequency domain. Further through quantization will generate many continuous zero coefficients. Two excellent entropy coding schemes can reduce coding redundancy: context-adaptive variable length coding (CAVLC) (Bjontegarrd and Lillevold, 2002) and context-adaptive binary arithmetic coding (CABAC) (Marpe et al., 2003). Therefore, H.264/AVC has higher compression ratio than prior standards and is more appropriate to limited transmission channel. However, this highly compressed video bit stream is very fragile over transmission environments. In the error-prone transmission channel, packet loss of the highly compressed video bit stream will cause the serious distortion. The distortion will propagate to its successive frames. This is because video coding standards utilize complex predictions to enhance the coding efficiency, especially as H.264/AVC. Thus, how to recover the lost video data in the decoder is critically essential. Since erroneous data would not only make seriously degrade in the current frame but also propagate to the following frames. For solving abovementioned problems, the error resilience and the error concealment techniques have been proposed in many literatures. The error resilience is a mechanism in the encoder for resisting packet loss. These preventative mechanisms are designed to improve the robustness of bit streams in noisy networks. On the other hand, the error concealment is an effective mechanism in the decoder. It applies to concealing corrupted regions by referencing previous decoded data. As a video sequence usually has strong spatial and temporal correlation, the corrupted