Stroke Sequence Identification in Handwritten Urdu Alphabets Using Convolutional Neural Networks

There is a lack of attention to early childhood handwriting in developing countries like Pakistan, where a single teacher teaches 50 to 100 students at a time. Children need a lot of attention from the teacher during their early stages (4-8 years), especially when they start writing the alphabet to assure that they adopted a correct stroke sequence. It is almost impossible for a teacher to pay attention to all his/her students to observe their stroke sequences. Due to a lack of teacher attention, the students may write the alphabet with correct or incorrect stroke sequences. If a student follows an incorrect stroke sequence, it badly affects the writing speed and the beauty of his/her cursive handwriting. As the research on this topic is concerned, so far, it has been conducted mainly in Chinese and Japanese languages to find out the correct stroke sequence. Arabic has received far fewer studies and Urdu is the most neglected till now. In this paper, we have proposed Urdu Handwriting E-Tutor (UHET). UHET is a Computer Vision based method that continuously monitors the handwriting activity of a child and successfully points out whether the stroke sequence (while writing an Urdu alphabet) is correct or not. To conduct this study, we created a new dataset that consists of images and videos of five Urdu alphabets. UHET exploits Convolutional Neural Network to train the model for predicting the alphabet (written by the child) and its stroke sequence. Results show that our UHET performs well achieving 80% accuracy on the average on the given data set.