Bangla printed word recognition using histogram based approach and fuzzy logic

This paper aims to present a technique for Bangla word recognition, using a histogram based approach and fuzzy logic. The recognition of a Bangla word proves to be challenging because most of its characters are joined together by a line named `matra´ which makes traditional procedures of character segmentation difficult. The system we propose is divided into three parts: image segmentation, character recognition and word recognition. In image segmentation, the image is preprocessed at first to remove noise and correct alignment, if rotated. Then the `matra´ is removed, thereby separating the characters from each other. Each character is then segmented by finding out the connected components. The recognition of these characters is done by dividing each 32×32 image into 8×8 slices. The histogram for all the slices is compared with the histograms in the database using fuzzy logic. At last, the word is recognized based on the order in which the characters were arranged. This system gives 94% accuracy for image segmentation and around 76% accuracy for character recognition. We conclude by inferring that this system works best for words with small number of characters.