Kannada handwritten word conversion to electronic textual format using HMM model

摘要

This project specifies a Hidden Markov model-based approach which is considered for identifying off-line words of kannada language which is written by hand. After preprocessing method, an image of word will be segmented to letters or each word in a line is segmented into frames. The procedure of this segmentation technique is used for transforming express image into frames which are in sequences. Preprocessing techniques includes binarization. From a line which consists of about 3 to 4 words in a line a technique is used which identifies each word from that line which uses green formula. The words are classified based on aspect ratio. Each word image is represented by its contour information. The system is used to first extract a set of robust features on binary handwritten images by using SIFT as well as feature detection and description(ORB). Later the system learns word HMM models using training samples from the Feature extraction. Finally, best word which maximizes the a posterior will be located through HMM. Because of the nature of the writing kannada handwritten word recognition is a challenging task. The recognition task of kannada words is prone to problems because of problems of many difficulties, such as the overlap, variability of character shape and the presence of ligatures in the word.