An object-based error concealment technique for H.264 coded video

Abstract

The H.264 video coding standard has become popular owing to its excellent compression efficiency. However, the H.264-coded video is very vulnerable to data loss. Conventional error concealment techniques interpolate the lost data in units of rectangular blocks, which limit the performance because a visual object is not equivalent to an image block. In this paper, we propose a new error concealment technique that uses visual objects as concealment units. The H.264 error-resilience tool FMO (Flexible Macroblock Ordering) is also incorporated at the encoder side for utilizing the spatial correlation. A lost region is concealed at the decoder side in three steps, namely object segmentation, object matching, and region-based patching. Objects are formed in the reference pictures based on color similarity, and adjacent objects of small area or the same motion are grouped as a unity. Motion estimation is performed on detected objects to find the associated motion vector. A lost region is concealed by the object in the reference picture with the best boundary-matching score. The proposed method provides considerably higher PSNR (peak signal-to-noise ratio) than conventional block-based approaches, especially for traditionally difficult cases and high-quality videos.