| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a region segmentation method and device enabling region segmentation with high accuracy at low load by reducing the influence of noise existing in a target image and to provide an image recognition device that uses the region segmentation device. <P>SOLUTION: The target image acquired is subjected to region segmentation using different characteristic quantities and groups of generated regions consisting of a plurality of generated regions characterized by the characteristic quantities used in the region segmentation are generated by the number of kinds of characteristic quantities (S130). One of the groups of generated regions is used as basic regions and the characteristic quantities of the generated regions overlapping with the basic regions are incorporated into the characteristic quantities of the basic regions so as to generate a single group of basic regions consisting of a plurality of basic regions characterized by the plurality of kinds of characteristic quantities (S140). Based on the characteristic quantities of the basic regions that constitute each group of basic regions, a weighted Euclidean distance between the basic region Rtar of interest and the basic region R<SB>k</SB>adjacent to it within a multidimensional characteristic space represented by the characteristic quantities is calculated as an index that represents the similarity between the basic region Rtar of interest and the adjacent basic region R<SB>k</SB>. When the distance meets integration requirements, both regions are integrated (S150). <P>COPYRIGHT: (C)2006,JPO&NCIPI |
