Block-based detection systems for visual artifact location

The core of many video coding standards is formed by the Discrete Cosine Transform (DCT) for de-correlating spatial video data. When quantizing DCT coefficients, artifacts may appear such as mosquito noise and ringing. Spatial artifact reduction requires artifact-location information, to control the filter process, thereby avoiding unnecessary blurring of non-visual artifact-contaminated regions. This location information can be derived either in the spatial or frequency domain. In this paper, we will present two systems: one for each domain. As coding artifacts are most annoying in flat or low-frequency regions, the objective of the detector is to localize these artifact-sensitive locations. The detection accuracy, coverage and sensitivity differ between the two different detection systems. Experiments have revealed that the spatial-domain system has a better performance on edge tracking and detection of small-sized flat region. However, the system degrades in separating originally present low-amplitude texture from artifact contamination, compared to the same capability of the frequency-domain system.