Error Concealment for Scalable Video Coding Based on Deformable Convolution Network

Scalable Video Coding (SVC) is a layered coding mechanism with advantages in network adaptation and error resiliency brought by the multiple layers. SVC encoder can encode a video into bitstreams, including a base layer that is usually given more protection and several enhancement layers. Although the enhancement layer is not as important as the base layer, frame loss in the enhancement layer can also result in quality degradation. Therefore, we propose a novel error concealment method for the enhancement layer. The proposed method consists of two modules, the RSR module and the QE module. The RSR module fully uses the relationship between the base and enhancement layers of the previous frame through deformable convolution to predict the intermediate image. The QE module post-processes the intermediate image using the difference between the enhancement and the upsampled base layer from the previous frame through a deep residual network to obtain the final predicted image for error concealment. We train our model on a public dataset and test it on a set of sequences. The experiment result demonstrates that the proposed method outperforms traditional error concealment methods and other deep-learning-based methods.