iReader: An Intelligent Reader System for the Visually Impaired

Abstract

For visually impaired persons, it is quite difficult to read printed text. Non-visual forms of reading materials, such as Braille, are available as Blind Aiding Technology amoung many others. In recent times, many devices and assistive equipment have been developed and technologies made available to assist visually impaired persons with reading. Most of these research works and products support reading from printed text-based manuscripts only. Due to this limitation, it may not be possible for a visually impaired person to describe and comprehend a printed image. In this paper, we develop iReader, an Intelligent Reader system that not only helps a visually impaired reader to read but also vocally describes an image available in the printed text. The Convolution Neural Network (CNN) is employed to collect features from the printed image and its caption. The Long Short- Term Memory (LSTM) network is used to train the model for describing the image data. The resulting data is sent as a voice message using Text- To-Speech to be read out loud to the user. The efficiency of the LSTM model is examined using the ResNet50 and VGG16. The experimental results show that the LSTM-based training model delivers the best prediction of a picture's description with an accuracy of 83