Implementation of MPEG-2 video compression artifact reduction system on GPU

Abstract

In this paper, we present our study on an MPEG-2 video compression noise reduction system that is based on the total variation regularization method and that uses the GPU. This system is very effective in removing image compression noise without losing picture sharpness. The problem with applying the total variation regularization method to motion pictures is the huge computational cost. Therefore, we implemented this system on the GPU, which is capable of massive computation in parallel, and we measured how fast it is processed. The experimental results show that the processing time of our system for Full HD images on the GPU is 7.81 ms, which is less than the 16.7 ms, for one frame time of HDTV. This result shows that our system can be applied to consumer devices such as HDTVs and PCs. Our system enables the adoption of noise removal for the next-generation HDTVs with 4K × 2K panels.