HANDWRITTEN CHARACTER RECOGNITION OF ASSAMESE HANDWRITTEN RECOGNITION USING CONVOLUTION NEURAL NETWORK

Abstract

The involvement of technology is reshaping our perception of the world. The inherent desire to understand human consciousness and intelligence has led to the widespread development of the fields like Artificial Intelligence and Machine learning. [1] Artificial intelligence has paved its way into most disciplines and blended in as an essential tool to boost efficiency and non-conventional enhancements. Linguistics is one such field; with the involvement of AI, communication and text extraction have become vividly easier. The presented work involves the development of one such application: Handwritten Text Recognition for the Assamese language. The presented work analyses text extraction from images and understanding by classifying it into proper categories for machines to understand it using the Assamese language, which is spoken in the Indian state of Assam. The framework of the work can easily be utilized for other languages just by scanning or capturing the text of the mentioned language. In this Project, the use of convolution neural networks(CNNs) is analyzed and proposed as the feature extractor for the handwritten Assamese characters. The classification for successful recognition of the scripts is achieved using the final layer of the CNN as a softmax activation layer. The dataset is obtained from the UCI repository for the training and testing of the proposed model. The results achieved by the testing of the model are quite satisfactory, with an accuracy of 99.87%.