| 摘要 |
PROBLEM TO BE SOLVED: To provide a classification execution method of a digital document and image for facilitating countermeasures to the variability of features without depending on template matching, inductive rule or high level feature analysis of an OCR or the like. SOLUTION: This system for performing classification through the generative models of features occurring in an image includes a storage unit for storing the category-conditional probability distributions of features occurring in a plurality of training images respectively assigned to one category and a classification unit. The classification unit includes: a feature identification part for specifying features occurring in an unclassified image; a likelihood evaluation part for determining category-conditional likelihoods for the specified features by using the category-conditional probability distributions for each category; and a category classification part for assigning the unclassified image to one such category based upon the category-conditional likelihoods. COPYRIGHT: (C)2008,JPO&INPIT
|
