Color image clustering segmentation based on SMCL for mobile robot

For conventional clustering segmentation of a color image, it is necessary to predetermine cluster number and centers of the color image. If they are not appropriately predetermined, results of segmentation may become considerably worse. To fulfill unsupervised clustering segmentation of visual color images for a mobile robot, this paper proposes a multiprototypes-take-one-cluster (MPTOC) strategy and splitting-merging competitive learning (SMCL). Based on MPTOC, SMCL can adaptively detect the appropriate cluster number of color images. An experiment on the mobile robot CASIA-1 validates MPTOC and SMCL.