Recognition of Offline Hand Written Telugu Script using Deep Learning

Abstract

Automatic character recognition is one of the significant parameters that allow the data processor to distinguish letters and numbers, using contextual data. Various efforts have been made to resolve this problem using different choices of classifiers and attributes, however problem remains complex. In this work, mainly focuses on the problem of handwriting recognition (HWCR) for a script in Telugu. A crosscutting structure is used for segments an image from text, classifies characters, and selects strings using a language model is provided. Segmentation is based on mathematical morphology. Writing in Telugu is a complex alpha syllable. For this suitable language is required, which complicates the problem. Conventional methods used artisanal characteristics, which required a priori knowledge of the field, which is not always possible. In this case, automatic feature retrieval could potentially attract more interest. In this work, a classical Convolutional Neural Networks (CNN) for the identification of Telugu symbols offline is presented. Qualified analysis demonstrated the effectiveness of the proposed CNN compared to previous methods with an interesting data set. In the test data set, the classification method provided with n accuracy of 98.7%.