2007 年 36 巻 4 号 p. 509-519
Text information in natural images is formally different from that in the images captured by traditional scanners. A text locating algorithm for natural color images is presented. The apparent non-text regions are removed by two-step processing of the gray-scale version of the original image, leaving the possible text regions. The first deletion is based on connected-component (CC) analysis of strokes. The second step uses a rules-oriented verification process for improving the performance.