Handwritten Tamil Character Recognition using Convolutional Neural Network
P. Gnanasivam, G. Bharath, V. Karthikeyan, V. Dhivya
Abstract
A handwritten Tamil character recognition system using deep learning algorithm is proposed. Convolutional Neural Network (CNN) is used for image classification and text classification. In this project the character image is converted into text and also represented as an audio output with the help of Google cloud Translation API. Using Deep learning algorithm, the undigitized Tamil letters are converted into readable format. We have trained for 12 vowels (soul letters), 18 consonants (body letters), 90 soul-body letters and one special character, in total we trained 121 characters out of 247 for which the dataset is collected from students as well as from Kaggle website. Thus, we obtained around 40,000 image samples. Out of 121 characters we get an accurate result (99%-100%) for 114 characters. The other seven characters will be accurate if the characters are captured in a proper lighting environment. The model accuracy is 95%. The system can also recognise the words with the combination of the above-mentioned letters.