Switchable video error concealment using encoder driven scene transition detection and edge preserving SEC

Abstract

Error concealment, a decoder based technique, attempts to reconstruct the corrupted regions of a video frame, using spatial and/or temporal correlation. For sequences with high temporal correlations, the performance of Temporal Error Concealment (TEC) techniques are better than Spatial Error Concealment (SEC) techniques in terms of PSNR of the reconstructed frames. However, the performance of TECs deteriorate drastically at the scene boundary due to low temporal correlation with the reference frame. In this paper, we propose a novel encoder controlled transition detection scheme which would facilitate selection of concealment strategy, that makes a choice between our newly proposed edge-direction based SEC technique and a TEC scheme. Exploiting Motion Vector (MV), Sum of Absolute Difference (SAD), and Motion Compensated Histogram Difference (MCHD), scene transition is detected at the encoder. A single bit transition status flag per frame is then sent to the decoder as a part of the header information to select the concealment policy between TEC and SEC. The present work also proposes a new edge-directed spatial error concealment, termed as "Directional Edge Based Spatial Error Concealment" (DEBSEC), which outperforms existing techniques in terms of PSNR and SSIM of the concealed frame.