公开(公告)号：JP2003216956A

公开(公告)日：2003-07-31

申请号：JP2002017902

申请日：2002-01-28

Applicant: ATR ADVANCED TELECOMM RES INST

Inventor： NAKAI TAKAHIRO ,  TADENUMA MAKOTO

IPC: G06T7/40 ,  G06T7/60

Abstract: PROBLEM TO BE SOLVED: To make correctly determinable whether a texture has a flat pattern to not. SOLUTION: An image is divided into small regions of a prescribed size in a step S3, and the texture feature quantity of image data corresponding to each small regions is calculated in a step S7. For example, the image data corresponding to the small region are converted into complex numbers by a two-dimensional DFT, and the absolute values of the complex numbers are distributed on a two-dimensional frequency flat plate. The center of gravity of the right half of the two-dimensional frequency flat plane is found based on the distribution, then a first feature quantity expressed by the distance between the origin and the center of gravity of the two-dimensional frequency flat plane is obtained. The smaller the first feature quantity is, it is judged to be a texture of clear pattern, and the larger the first feature quantity is, it is judged to be a texture of obscure (flat) pattern. COPYRIGHT: (C)2003,JPO

公开(公告)号：JP2003242508A

公开(公告)日：2003-08-29

申请号：JP2002039485

申请日：2002-02-18

Applicant: ATR ADVANCED TELECOMM RES INST

Inventor： NAKAI TAKAHIRO ,  TADENUMA MAKOTO

IPC: G06F17/18 ,  G06T7/00

Abstract: PROBLEM TO BE SOLVED: To enhance the reliability to enhance the convenience for successive processing by enabling the clustering to an optimum number of clusters. SOLUTION: The distance between a sample point within a cluster and a representative point (center-of-gravity point) of the cluster is calculated, and a first sum is calculated by summing up the distances (errors) over all sample points belonging of the cluster. A second sum value is calculated by summing up the first sum values over each cluster (S59-S63). When the number of cluster is reduced by 1 to perform the clustering to a, two clusters having the shortest cluster-to-cluster distance in the previous clustering result are collected, whereby a second sum value when reducing the cluster number by 1 is calculated. The number of cluster corresponding to a second sum value of allowable magnitude among the second sum values in several positions around the position where the second sum value is suddenly changed is determined as the optimum number of cluster (S71-S75). COPYRIGHT: (C)2003,JPO