| 摘要 |
PROBLEM TO BE SOLVED: To properly detect a text part and a natural part of an image and to discriminate both parts from each other by deciding a fuzzy detection rule set to discriminate a text-like part from a non-text-like part of the image and applying this rule set to the image that is divided into plural segments. SOLUTION: In a rule decision process, a fuzzy detection rule set is decided to discriminate a test-like part from a non-text-like part of an image. Then the image is divided into plural segments, and the fuzzy detection rule set is applied to each segment of the image. Thus, these segments are sorted into the text-like and non-text-like parts. In the rule decision process, plural image features are also discriminated to separate the different parts of the image and the combinations of different features are applied to both text-like and non-text-like learning images. Thus, plural fuzzy detection rule sets are obtained and then minimized, and such rules that are not supported by a learning image of a prescribed amount are excluded. |
