Offline Handwritten Dogra Script Recognition Using Convolutional Neural Network

The handwritten optical character recognition is a challenging and active branch of pattern recognition. The recognition of Indian scripts is potentially a complex problem due to several challenging issues like complex shapes of characters, similar shaped characters, positioning of diacritics, dots, etc. This paper focuses on the recognition of handwritten Dogra script. This paper also contributes to the construction of a handwritten Dogra script dataset. The handwritten document images are manually collected from several sources like archival departments, libraries, and museums and then pre-processed and segmented in order to obtain the final handwritten character dataset. The resultant handwritten Dogra dataset is evaluated using a convolutional neural network classifier, reporting a promising recognition accuracy of 88.95%.