Word prediction for a real-time reader device for blind people

Abstract

The aim of this project, taking the software developed in a previous work as the starting point, is to increase the recognition reliability and robustness. The main goal of the future global system is the ability to be the closest possible to the way that blind people read, increasing the accessibility to this group of people. If this system can considerably help blind people to read, these people would probably get more reliability to access new technologies, due to the fact that unfortunately, nowadays, a great amount of blind people do not use computers because they can not access them. Therefore, a way to increase the system reliability is to make it more robust. The current system based on artificial neural networks processes a character and tries to recognize it only taking into consideration its acquired image from the camera. In consequence, the system does not take into consideration other information which would increase the system accuracy. Other information could be the use of previous characters or some orthographic notions of the language in use, which are useful to avoid errors when a bad recognition has occurred. For this reason, a character and word-level prediction systems have been implemented. On the one hand, useful to add a simultaneous way of recognition and, on the other hand, the starting point of a system able to correct characters or words in text.