LAUSR

Overview: Handwriting recognition (HR) involves converting handwritten text into machine-readable text. Tamil handwritten document recognition remains a challenging process in various text real-world applications owing to the differences in the sizes, styles and orientation angles of Tamil alphabets. Prior studies concentrated only on character-level segmentation, and each character was subsequently classified. The recently developed machine learning (ML) and deep learning (DL) approaches can be utilized for Tamil handwritten character recognition (HCR). Objective: This paper attempts to present an end-to-end DL-based Tamil handwritten document recognition (ETEDL-THDR) model. Methods: Segmentation is used, first at the word level and then at the line level. ETEDL-THDR text recognition can be accomplished using two modules: line segmentation and line recognition. Initially, the text ETEDL-THDR model targets improving input image quality using the median filtering (MF) technique. To create meaningful regions, more line and character segmentation activities are performed. A deep convolutional neural network (DCNN) based MobileNet approach is also applied to derive feature vectors. Finally, the water strider optimization (WSO) algorithm with a bidirectional gated recurrent unit (BiGRU) model is used to identify the Tamil characters. Results: Extensive experimental analyses of the text ETEDL-THDR model have been carried out, and the results show that the text ETEDL-THDR model performs better than more recent methodologies, with a maximum accuracy of 98.48%, a precision of 98.38%, a sensitivity of 97.98%, specificity of 98.27% and text F-measure of 98.35%. Conclusion: The comparison results show that the proposed model can recognize Tamil handwritten documents in real time.