A robust image segmentation method using hierarchical color clustering

A robust and fast image segmentation method is provided based on color distribution. In this paper, unlike the widely used training and optimizing process, an image is recursively divided only according to the color differences, a whole region is divided into two regions that have the most difference Gaussian Mixture Model distribution in color space, and the image is segmented into separate and continuous regions after a few iterations. Morphological post process is introduced to get correct segment results with closed and semantical boundary. At last of this paper, experimental comparison is provided on different hardware platform. A robust image segment result is got with high calculation performance according to contract experiments on different platforms including normal PC and high performance server with GPU support.