A CNN-Based Post-Processing Algorithm for Dark Video

Low light scenes often come with acquisition noise, which not only disturbs the viewers, but also brings special characteristics to video compression. Common video compression method adopts filters to deal with annoying distortion and artifacts such as blocking, blurring, and ringing. Versatile Video Coding (VVC), the most recent video standard, adopts four in-loop filters for the improvement of the coding efficiency, including In-loop reshaping, Deblocking filter with strong longer filter, Sample Adaptive Offset and Adaptive Loop Filter. In a certain extent, elaborate filters help reduce the nonlinear distortion and improve the video quality. But the search of better universal filters become more exhausting. Recently, the progress of deep learning shows the possibility to settle the complex problems in the computer vision field. According to the compressive sensing theory, the post-processing method at the decoder end can further enhance the coding efficiency. In this paper, we propose a multi-scale residue learning dark video restoration CNN (DRCNN). We feed the decoded pictures to the model at the decoder and apply luma pictures and chroma pictures to the model, respectively. In order to comprehensively evaluate the coding performance of both luma and chroma components, the PSNR and VMAF is exploited to measure the effectiveness of the proposed method. Compared to VVC anchor streams, our approach achieves the BD-rate reduction of 33.7% on Y-PSNR and 36.08% on average-PSNR, and our approach also has a promotion on the objective quality Compared to VTM-7.0 baseline.