A Hue-Based Quad-Tree Method for Color Image Segmentation

Image segmentation is an important technique in image processing. It be used for separating the objects or foreground from the background. This paper focuses on the split step of a split-and-merging approach for color image segmentation. For splitting, we propose a quad-tree method based on hue angles (QTHA). The QTHA introduces a novel metric criterion for measuring hue homogeneity within a block in the quad-tree. The homogeneity is guaranteed by monitoring the within-block parameters, such as angle difference, saturation radius, and light difference. Experimental results show that compared with a conventional method working on the RGB space, the QTHA is more consistent with the human visual perception and provides the segmentation that is more accurate.