A Blanket Binarization Method for Character String Extraction

In this paper, a binarization method based on fractal dimension for character string extraction is proposed. In character extraction from a scene image, a major problem is how to deal with much different type of characters in a complex background. The proposed method can obtain multiple threshold values which are correspond to each character regions by detecting the stable intervals of fractal dimension FD. The stable interval is a relatively low and flat valley of the FD which indicates the binarized image has the stable connected regions, and therefore fine character regions have been appeared. The character regions may contain some noise and has conflictions between the regions derived with another threshold values. We call these character region as a ”Candidate Character Region Images”(CCRI), and will be processed by noise-reduction consists of two steps. After that, CCRI are integrated into one binarized image as output image through the contention resolution process. We show the performance of the proposed method by comparing Niblack’s method as a local method and Otsu’s method as a global method on the dataset provided at ICDAR 2003.