Performance Evaluation of Spatial Domain Error Concealment for Image Recovery

Many spatial domain error concealment techniques, which are used in decoder to improve visual quality of damaged images, have been proposed in the past years. Each algorithm is based on different assumptions and has different performance under different conditions. However, there is no extensive comparative evaluation available in literature. Performance evaluation helps us to understand the state-of-art of techniques and indicates further research directions. In this paper, evaluation results of spatial domain error concealment for image recovery are presented. Many algorithms are realized and evaluated, such as pixel domain interpolation (PDI), multi-directional interpolation (MDI), projection onto convex sets (POCS) and maximally smooth recovery (MSR), etc. Different algorithms have different performances on different type of regions, e.g. smooth region, region with edge or texture region, etc. In addition to natural images, several artificial images, in which we can control the size, location and relative ratio of different types of regions, are used to test some specified aspects of each algorithm. Our experiments show all algorithms perform well on smooth regions. However, fine texture image is still a challenge. MDI performs well on regions with edges. Maximally smoothness recovery approach has relatively good performance, and low computation complexity. Based on the above observation, we proposed an adaptive version of MSR, using different smoothness measures for different regions. Experiments show, our improvement can get better results than MSR.