Recognition of Historical Handwritten Kannada Characters Using Local Binary Pattern Features

Abstract

Archaeological departments throughout the world have undertaken massive digitization projects to digitize their historical document corpus. In order to provide worldwide visibility to these historical documents residing in the digital libraries, a character recognition system is an inevitable tool. Automatic character recognition is a challenging problem as it needs a cautious blend of enhancement, segmentation, feature extraction, and classification techniques. This work presents a novel holistic character recognition system for the digitized Estampages of Historical Handwritten Kannada Stone Inscriptions (EHHKSI) belonging to 11th century. First, the EHHKSI images are enhanced using Retinex and Morphological operations to remove the degradations. Second, the images are segmented into characters by connected component labeling. Third, LBP features are extracted from these characters. Finally, decision tree is used to learn these features and classify the characters into appropriate classes. The LBP features improved the performance of the system significantly.

Hybrid manifold smoothing and label propagation technique for Kannada handwritten character recognition

“…The literature and artistic diversity of the language makes it a priceless repository of information and culture. Many of these regional languages need the power of technology to retain the language directed at them (Thippeswamy and Chandrakala, 2020;Parikshith et al, 2021). The preservation of the language's scripture is greatly aided by advances in digitization, which also give the language a significant edge in terms of reaching a wider audience given the pervasiveness of internet access around the world.…”

Deep Convolutional Neural Networks for Degraded Printed Kannada Character Recognition

“…In a later work by Thippeswamy G, and Chandrakala H. T [11], local binary pattern features are employed for historical handwritten Kannada character recognition. However, the features employed are not efficient towards handwritten character recognition.…”