High-Performance Character Recognition System Utilizing Deep Convolutional Neural Networks

Abstract

Accessible reading is still a major challenge for those with visual impairments in our digitally-driven world, particularly for those who are born blind. The creation of a Letter Recognition System (LRS) for the blind is the novel solution to this problem that this research suggests. With the help of this device, blind people can access printed letters according to its reading solution. In order to obtain images of printed letters, this study presents practical hardware that uses a webcam and a manual printing machine. The paper can be moved to acquire the written letters on it. Hence, a large collection of data containing letters known as Printed English Letters-version 2 (PEL2) dataset was gathered for the printed English letters (A–Z). Following acquisition, the input images undergo preparation, segmentation, and resizing. After that, a Deep Convolutional Neural Network (DCNN) is used to recognize letters from them. Ultimately, it is recommended that the identified letter be transformed into letter-to-speech audio to enhance the system's efficacy in assisting blind or visually impaired individuals with reading. In this work, a very high accuracy of 99.70% for letter identification has been calculated and attained.