Color images segmentation using new definition of connected components

This paper proposes a definition of the connected components of color images, (/spl epsiv//sub H/, /spl epsiv//sub C/)-connected components ((/spl epsiv//sub H/, /spl epsiv//sub C/)-CCs). A systematic segmentation method using (/spl epsiv//sub H/, /spl epsiv//sub C/)-CCs is also presented. The similarity of pixels is measured in the IHC (intensity, hue and chroma) color space, which was proposed by the authors previously, and the similar pixels in a given image are grouped into (/spl epsiv//sub H/, /spl epsiv//sub C/)-CCs. Experiment results demonstrate that color images can be effectively segmented in accordance with the perception of human using the definition of (/spl epsiv//sub H/, /spl epsiv//sub C/)-CCs and the systematic method. A hybrid system composed of MEBML (mind-evolution-based machine learning) and MLCNN (maximum likelihood clustering neural network) is used to cluster features of small windows of an image. The hybrid system has good performances on clustering and makes the color images segmentation algorithm efficient.