Text extraction in natural image by using OCR score

With the growth of the portable device with camera, we can take a variety of photos. In addition, with the higher resolution of the camera, it becomes possible to recognize a character in the scene taken by the camera. There is much text information around us. It also provides us a useful information. Therefore, if the system recognizes a text information in the scene, it is possible to cooperate with various systems. To recognize it in the scene, it is necessary to identify the position of the character in first, however, it is difficult. In this paper, we use a connected component method and CSER to detect a character candidate by separating a background. In addition, to reduce a character candidate, we propose the system that uses OCR score and compares a histogram of character that is taken out and a histogram of the other character. Moreover, by using OCR score, it is considered to be possible to extract a single character that has not been extracted in previous studies. Experimental result shows that the proposed method has the extraction accuracy of 74.6 %.