Letters Recognition System by Deep Convolutional Neural Network

This paper proposes a new system to address challenge for reading accessibility through the development of a Character Recognition System (CRS) for blind people. This system has reading solution for aiding blind individuals by accessing printed letters. This paper introduces useful hardware for capturing images of printed letters using a manual printed machine and webcam, where a paper can be provided and moved in order to acquire its printed letters. So, a big dataset is collected for the printed English letters (a-z), it is named the Printed English Letters (PEL) dataset. After the acquisition, the input images are prepared, segmented and resized. Then, they are sent to a Deep Convolutional Neural Network (DCNN) for letters recognition. At the end, the recognized letter is suggested to be converted into audible speech, character-to-speech, in order to make the system more useful in helping blind persons. The main results were impressive and superior. That is, values of statistical metrics were recorded as 99.44%, 92.77%, 92.77%, 99.71% and 9 2.7 7% for the accuracy, precision, sensitivity, specificity and f1_score, respectively.