An algorithm for locating characters in color image using stroke analysis neural network
Anto Satriyo Nugroho, S. Kuroyanagi, Akira Iwata · 2003
Character segmentation is a significant part in a character recognition system. Particularly, when the system is assumed to work in a color image with multi-segment characters such as Japanese Kanji characters, the complexity of the characters and the background properties bring the difficulties to the segmentation problem. Discussion in this paper is focused on designing an automatic system for locating text regions, assuming that the texts are composed by Kanji characters. The principle of the proposed model is the inclusion of recognition phase to give a feedback in controlling the segmentation task, yielding a robust algorithm to solve the complexity of the characters. The algorithm is assumed to work with color images, which makes it suitable for practical applications. The evaluation of the model shows that the algorithm promises an appropriate approach to deal with the complexity of Kanji character segmentation.