Clustering-based Optimization for Side Window Filtering

Side window filtering (SWF) is a new technique that significantly improves edge preserving capability. Based on our observation, it still suffers from edge blurring caused by filtering across edges. To address the issue, a clustering based optimization is proposed for SWF, which is motivated to only use the pixels that are not across edges in the filtering process. With the clustering strategy, the image is first grouped into perceptually homogeneous regions, so that the pixels on two sides of an edge arc divided into different clusters. Each cluster is assigned with a unique label, and the pixels in the same cluster share the same label. In each side window, only the pixels that have the same label with the one being processed are used for filtering. Extensive analysis and experimental results show that the proposed optimization can further improve edge preserving capability as compared to SWF. Meanwhile, the increased complexity of the proposed optimization is only O(N), which is linear in the number of image pixels.