An Enhanced Deep Learning Model for Handwritten Tamil Character Identification

• Main Authors: Babitha Lincy, R, Jency Rubia, J, Sherin Shibi, C, Kavitha, M
• Format: Conference Proceeding
• Language: English
• Subjects: Adam Optimizer Algorithm; Deep learning; Enhanced Convolutional Neural Network; Image resolution; Image segmentation; Industries; Machine learning algorithms; Optical character recognition; Support Vector Machine; Support vector machine classification; Tamil Character Recognition; Writing
• ISSN: 2832-3017
• DOI: 10.1109/ICSSIT55814.2023.10060920

In the data processing industry, data entry is regarded as a great bottleneck. An eminent Character Recognition (OCR) system acts as a key solution for this concern. The OCR is being familiaized in universal level for different languages. Till now, only a limited explores were made for the Tamil Handwritten character recognition (HCR). Existing research works provides best performances for many languages, but for the work towards Tamil language is not satisfactory in the recognition area. To resolve this demerit this paper proposed an enhanced CNN for Tamil HCR utilizing Adam optimizer. Initially, encode the target variable using "one-hot4ncode" method. Then, the encoded image is preprocessed using RGB to grey conversion, image resizing, slant removal and array conversion. Subsequently, word and characters are segmented as of the preprocessed image. Next the segmented images are provided as input to the CNN classifier. The CNN is enhanoed by utilizing Adam optimizer algorithm (AOA) which optimizes the weight value for minimizing its loss function. finally, the experiential outcomes of the proposed CNN are contrasted to the prevailing SVM and some other machine learning algorithms. The proposed recognition system shows preeminent performance when comparing with the prevailing methods.