Handwritten Bangla character recognition to braille pattern conversion using image processing and machine learning

Abstract

There has been significant development in the recent Education System with the rapid development of technology however there are very few facilities that can help those people with disabilities such people without sight. Braille has been one such method however it has not yet been digitized or has not been incorporated in the education system. On the other hand, many papers and works have been conducted on English Character Recognition with satisfactory outcomes with considerably good accuracy however such cannot be said about Bengali Characters. Some works have been done with considerable accuracy however its frequency is still low. This paper introduces the development of a prototype system which takes an image of a handwritten Bangla Character and applies the concepts of Image Processing using OpenCV and Machine Learning to capture and process an image then recognize the character and finally with the help of a device present the recognized character in a Braille pattern. Furthermore, this paper will also look into the different machine learning modules and assess the reliability and accuracy. For the machine learning part Deep Neural Network was applied on the image and then VGG-16 , Resnet-50 and DenseNet-121 which are modules of Convoluted Neural Networks where used. Then the outcome is passed into a device which will engrave the corresponding character into its respective Braille Pattern. The device will consist of Atmega2560 chipset (Arduino Mega) , Servo Motor and LCD which will process the data and finally present it in Braille pattern using servo motor and will show the corresponding character on the LCD monitor.