Bangla handwritten digit recognition using dynamic search process

Abstract

Pattern matching method is one of the major principles used for character recognition. This method is applied to the hand written numerical characters. In this thesis we investigated dynamic search process to recognize a digit. This method conducts an adaptive matching between reference pattern and a handwritten input digit. This input digit is actually converted to bit map format which is subsequently compared to the reference digit. This thesis uses the reward-punishment approach for all the algorithms developed for recognition. For this purpose it uses a special type of weighting factor for reward and punishment. This is used as conditional weighting factor. According to this weighting factor, the segments of the reference digits are classified into three types of segments to give a digit either reward or punishment. According to this weighting factor every reference digit gains or losses scores in proportional to the length of the segment and weighting factor. The reference digit giving maximum score can be declared as the candidate for a valid digit if it can overcome its threshold score. The first approach was to search only in the specific direction and area in input digit. It permits no flexibility, no rotation and no shearing of the input numeric character. The recognition of hand written characters is a special kind of complex mathematical problem, because the different distortions present in hand written character set make it difficult to produce a distinct set identification. For this reason at the later stages, the reference pattern vectors were permitted to accept aberrations to some extent in the direction of the input digit to cope with abrupt aberrations in the input pattern line segments. This improved the flexibility of matching. For this purpose, rotation and shearing to some extent is performed in the input pattern. This approach helps to recognize the hand written numeric characters. which are italic.