Handwritten Text Recognition using VGG19 and HOG Feature Descriptors
Saikrishna Devendiran, Jyothiratnam, Prema Nedungadi, Raghu Raman R · 2024
Recognition of Tamil characters has been a challenging task, and especially handwritten text recognition has been a bigger hassle. The intricacy of the script with its inherent cursive nature and the existence of numerous diacritical marks adds to the complexity of automating its recognition. In this research, we delve into the realm of Optical Character Recognition (OCR) with a focus on handwritten Tamil characters, employing two distinct deep learning models: Model A using Histogram of Oriented Gradients (HOG) for feature extraction and Model B leveraging the robust architecture of VGG-19. We present a comprehensive dataset, meticulously compiled from over a thousand individuals across various age groups to capture the rich diversity of handwriting styles. Through a multi-stage preprocessing pipeline involving cropping, resizing, noise reduction, and grayscale conversion, we condition the images for optimal feature extraction. The study contrasts the performance of HOG-based feature extraction against the convolutional neural network approach of VGG-19, demonstrating the latter's superior efficacy in handling the complexity of Tamil character recognition. The results indicate a significant advancement in the accuracy and reliability of OCR for Tamil, with potential applications in digitizing historical manuscripts, facilitating digital learning, and supporting the preservation of this classical language.