CALLIGRAPHY IDENTIFICATION OF MULTISCRIPT TELUGU DOCUMENTS

Abstract

The rapid growth of digital document processing and multilingual information exchange has intensified the need for efficient script and calligraphy identification systems. In complex document environments, particularly those containing South Indian scripts such as Telugu, Tamil, and Malayalam, accurate identification of handwritten and printed text remains a challenging task due to variations in writing styles, structural similarities, and noise in scanned images. This work addresses the problem of calligraphy identification in multi-script Telugu documents through an integrated image analysis and deep learning framework. The methodology incorporates systematic pre-processing techniques including noise removal, normalization, and segmentation at multiple hierarchical levels such as text lines, words, and characters. Discriminative statistical and structural features are extracted to capture script-specific characteristics. These features are then utilized by deep learning-based classifiers, particularly convolutional neural networks, to achieve robust and scalable classification performance. The approach effectively handles hybrid documents containing both handwritten and printed content, improving recognition accuracy in diverse real-world scenarios. The proposed framework contributes to the advancement of intelligent Optical Character Recognition systems by enabling reliable script identification, which is a critical prerequisite for downstream text recognition tasks. The results demonstrate improved adaptability, accuracy, and efficiency in multi-script handwritten document analysis.