Fuzzy threshold method for off-line Chinese handwritten character segmentation

Off-line Chinese handwritten character segmentation usually includes the coarse and fine steps. In order to identify which regions that gotten through the coarse step should be merged or segmented again, a threshold is usually adopted. In this paper, a fuzzy threshold method is presented to adapt for the variety of unconstrained writing styles. First, the coarse segmentation region´s distribution is gotten by taking statistics with its width and height. Second, the region membership values for broken, isolated and touching character are gotten by the fuzzy functions, and then the region class is determined by the max membership value. Last, the fine segmentation step is instructed by the fuzzy threshold and feedbacks from recognizer. Experiments are performed and the results show that the fuzzy threshold method is quite simple and effective.