A Binarization Method for the Character Extraction from Scene Image by the Blanket Method

抄録

In this paper, a binarization method based on fractal dimension for character string extraction is proposed. In order to deal with a scene image that has more than one character strings which have different colors (gray scale values), we have to take multiple threshold values. The proposed method can obtain multiple threshold values which correspond to each character string by detecting stable intervals of fractal dimension FD. The binarized character regions can be white or black, so we add reversed version of each character string region, that is, we generate “negative-positive problem” and then, obtain the black character regions by solving nega-posi problem. We also introduce two steps of noise reduction based on the shape of bounding box and edge information. The experimental result on datasets of ICDAR2003 shows that the proposed method has higher performance than existing binarization methods.