摘要 |
PROBLEM TO BE SOLVED: To obtain a mechanism for eliminating unnecessity to collect data preliminarily determined to be normal when preparing normal data distribution in the detection technology of abnormal data, and for, when similar data are added, adding the similar data to the normal data for relearning. SOLUTION: This abnormal image detection device is provided with an image input part 11 for inputting an image, a characteristic vector extracting part 12 for extracting featured values from the input image, a cluster configuring part 15 for preparing the cluster of the images inputted based on the extracted featured values, and for reconfiguring the cluster based on the featured values transmitted from a decision result output part 14, similarity deciding part 13 for determining whether or not the input order of the input images exceeds a threshold, and for determining whether or not the input image whose input order is larger than the threshold is belonging to the cluster and a decision result output part 14 for outputting the result when it is decided that the input image is an abnormal image, and for transmitting the feature values of the input image to the cluster configuring part 15 when it is decided that the input image is not any abnormal image. COPYRIGHT: (C)2007,JPO&INPIT
|