Ranking Segmentation Paths Using Fuzzified Decision Rules
Zhongkang Lu, Zheru Chi, Pengfei Shi, Eam Khwang Teoh · Studies in fuzziness and soft computing · 2000
Connected characters are usually recognized using a segmentation-based approach in which individual characters are segmented first by segmentation paths before they are fed to a character classifier. Good ranking of segmentation paths can significantly shorten the processing time in evaluating segmentation paths and improve the recognition rate by avoiding misinterpretation. In this chapter, we present a method of using fuzzified decision rules to rank segmentation paths. The fuzzified decision rules are obtained by fuzzifying the decision rules extracted from a decision tree learned from examples. Nine measures of the properties of a segmentation path are extracted as the input and the centroid defuzzification method is adopted to defuzzify the output. Our method of ranking segmentation paths has been tested on the NIST Special Database 3 with good results.