Online Urdu Handwritten Text Recognition For Mobile Devices Using Intelligent Techniques

Abstract

Recognition of online cursive handwriting is difficult and challenging task. This become more cumbersome when it comes to mobile devices where we have small writing area, different writing styles and stroke sequence variation. This research introduces different approaches for recognition of Urdu handwritten characters and word for mobile devices. Comparing all points is resource hungry approach especially for mobile phones. Looking at challenges of mobile phones, Urdu characters are recognized using vectors. It results in fast processing and efficient recognition of Urdu characters consuming lesser resources of the devices. Nevertheless, for complex inputs i.e. online handwritten words proposed technique did not provide good results. Urdu words are compounds of different characters connected with each other. Correct segmentation of online cursive handwriting is difficult to achieve on small writing area. Segmentation free approach is used to recognize Urdu words written on mobile screen. In segmentation free recognition of cursive handwriting, writing rules plays very important role in good and accurate recognition. Looking at shape and location of each character within word, writing rules are designed for Urdu language. A new algorithm is proposed to preprocess the complex input and preserve shape of the actual input. Fuzzy association rules are used to link secondary stroke with their respective primary strokes. Different classifiers such as hidden Markov model (HMM), fuzzy logic, K nearest neighbor (KNN), hybrid HMM fuzzy, hybrid KNN fuzzy and convolutional neural network (CNN) are used for the classification. Statistical tests are applied to find the significance of classifiers results. Experimental results show that by applying proposed preprocessing algorithms and writing rules, recognition rates are impressively enhanced. Vector quantization method provides better results for recognition of online Urdu handwritten characters and hybrid KNN and fuzzy classifier for recognition of ligatures and words in mobile phones.