Optical Character Recognition with Word Prediction Feature for the Blind

Chew, Shaun Han Wen (2021) Optical Character Recognition with Word Prediction Feature for the Blind. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Shaun Chew Han Wen.pdf Restricted to Registered users only Download (3MB)

Abstract

There are 253 million people that are blind and visually impaired as according to the World Health Organisation (WHO) in 2017. They will encounter some difficulties when accessing any materials to read or study. Besides that, some books do not have audio or Braille versions. Some Optical Character Recognition (OCR) software will also show some errors in the output. Therefore, this project is done to allow the blind to read more conveniently and to reduce the error rates of the Optical Character Recognition device. The OCR device is done using a Raspberry Pi microcontroller and a camera module to capture the image. A text-to-speech engine is also included to read the text shown at the output. To reduce the error rates, a Word Prediction feature is implemented. The method used to perform this feature is done using the First Order Hidden Markov Model. There are 3 novels and 3 short stories that are used to test the accuracy. The experiment is done and the total average accuracy rates for the novels and short stories is 94.06%. The accuracy obtained is lower than expected due to some limitations found. Moreover, the OCR device is created and is able to read the text using the text-to-speech engine. The results obtained shows that the error rates can be reduced and the OCR device can be created using a Raspberry Pi microcontroller with a camera module attached. Further improvements can also be made to achieve the expected output.