Lanna Dhamma Handwritten Character Recognition on Palm Leaf Manuscripts Based on Hierarchical Classification

      Lanna Dhamma is an ancient alphabet. It was used through the Lanna Kingdom for inscribing knowledge and legends into documents such as palm leaf manuscripts. During the last hundreds of years, those manuscripts had been damaged. Large number of valuable manuscripts have been surveyed and digitized by many organizations. To extract useful information, transforming these manuscripts in form of digital images into machine-encoded texts is often considered to be the first step towards text mining. So, this thesis presents a handwritten character recognition of Lanna Dhamma alphabet on palm leaf manuscripts. Our works focus on character segmentation and recognition tasks that the complex characteristics of Lanna Dhamma are challenges in. In character segmentation, we propose a segmentation method that utilizes feedback from character recognition model in order to determine segmentation points. An optimal solution are retrieved using a graph partitioning algorithm. For character recognition, we propose a recognition model designed based on the hierarchical classification. Local discriminative features are used to distinguish character classes. The two proposed method were evaluated through cross-validation and writer-independent tests. The proposed methods have produced promising performance, and outperformed existing methods.