Real-time video postprocessing algorithms and metrics

Abstract

The main purpose of video postprocessing is to reduce compression artifacts, such as blocking, ringing, and temporal noise. Many existing postprocessing algorithms either require very high computational cost or cannot visually remove the artifacts, especially at very low bit rates. In this research, we propose an adaptive de-blocking algorithm, a clustering-based de-ringing algorithm, and a motion-compensated temporal filtering with a down-sampled pattern. These algorithms can smooth out the compression artifacts while preserving the strong edges and texture areas. In addition to effective artifact reduction, our algorithms have low computational cost. The implementation of our de-blocking and de-ringing algorithms on a mediaprocessor demonstrates the feasibility of real-time video postprocessing. In addition, we have also developed new metrics to measure the blocking artifacts in the reconstructed or postprocessed images. To validate our algorithms and metrics, we have designed statistics-based subjective quality assessment experiments. The experimental results confirm that our proposed postprocessing algorithms have achieved a significant improvement over other methods while the proposed blockiness metric is more consistent with subjective evaluation than peak signal-to-noise ratio (PSNR).