SUBSAMPLING-BASED CLUSTERING METHOD FOR FAST SUPERPIXEL SEGMENTATION

摘要

An image clustering method according to the present invention includes: an initialization step for designating values of an interval S and a normalization variable m, setting a plurality of cluster centers in an image, and classifying all pixels in the image into G_1 and G_2 sets; a cluster assignment step for obtaining distances to the cluster centers with respect to the pixels included in the G_2 set and then obtaining the distances to the cluster centers with respect to the pixels not included in the G_2 set but included in the G_1 set; a changing step for re-calculating and changing the plurality of cluster centers in view of the cluster-assigned pixels; a step of comparing a predetermined threshold value (T) to an amount (E) changed in the changing step; and a step of re-executing the cluster assignment step in a case where E exceeds T and terminating the cluster assignment step in a case where E does not exceed T. The present invention provides a better image clustering processing effect than SLIC.