| 摘要 |
PROBLEM TO BE SOLVED: To overcome the disadvantages such as an overlap of areas (noise components) outside a character caused in the result of obtaining a character row area information based on a multi-valued color image, unintended parting of a character row due to a punctuation mark, and integration of connecting components where a ruled line exists internally. SOLUTION: With a run (horizontal continuous pixel string) obtained from an input color image as an object, a run of the same color contacting in the vertical direction is integrated as a connecting component to generate a character candidate. A ruled line extract means 106 extracts depending on the feature of the ruled line from the connecting component obtained on the basis of the input original image. A circumscribing rectangle integration means 107 prevents parting of a character row by recognizing a punctuation mark, and inhibits area integration of character rows extending over the extracted ruled line. An overlap area eliminating means 109 eliminates an area outside the character row area according to the character feature amount, and an area re-integration means 110 performs integration of parted character rows caused by a punctuation mark or the like. COPYRIGHT: (C)2003,JPO
|
