RECOGNITION OF PERSIAN HANDWRITTEN NUMBERS BASED ON ASSEMBLY OF REINFORCED CLASSIFIERS

Abstract

According to the economic justification study in field of recognition of handwritten letters and numbers and specially the more sensitiveness of handwritten numbers, in this report has tried to do a deep survey on this field. One of the general weaknesses that in most of research in Persian language and often English language are raised is using simple classifiers that finally lead to recognition with Low accuracy. This especially is raised about recognition of numbers. Hitherto, different methods have introduced for increasing classifier's efficiency that in most of them efficiency of a particular classifier optimize for a particular sets of data. According to this that the number of recognition categories in these issues are a lot, using binary classifiers can efficiently leads to more accurate detection. In this report it will be tried to present new and efficient method compared to the existing methods for recognition of single handwritten digits. Detector classifiers of each category are trained especially. Then by using classifiers combination methods, the final decision has predicted for category of an input digit.