Image segmentation using a hue-guided quadtree

Abstract

Finding the boundary of the object from the image is an important step in image processing. Splitting and region merging is one of the typical approaches for image segmentation. This paper proposes a slipping-and-region-merging method. The image is segmented by the quadtree to establish a relationship list between the blocks. The block is treated as a seed and merged for similar hue angles between adjacent regions. Our system adopts Lab2000HL color space, which is closer to human vision of color perception. The experimental results prove that our method can more accurately find the actual boundary of the object and effectively reduce unnecessary blocks.