PNN and Deep Learning Based Character Recognition System for Tulu Manuscripts
C K Savitha1, P J Antony2
1C K Savitha, Department of Computer Science and Engineering, KVGCE, Sullia (Karnataka), India.
2P J Antony, Department of Computer Science and Engineering, A.J Institute of Engineering & Technology, Mangalore (Karnataka), India.
Manuscript received on 18 June 2019 | Revised Manuscript received on 25 June 2019 | Manuscript published on 30 June 2019 | PP: 1854-1861 | Volume-8 Issue-5, June 2019 | Retrieval Number: E7890068519/19©BEIESP
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: The main intention of the work is to implement a machine learning based offline Hand written character recognition (HCR) system for South Indian ancient language called Tulu. Degraded images are preprocessed with adaptive thresholding based binarization, median filter based noise removal and skeletanization processes. The classification of characters is done by the help of a Probabilistic neural network (PNN) and Deep convolution neural network (Deep ANN) models. Wavelet transform and Zone wise gradient direction values of skeletons of characters are extracted to form feature vectors, which are used for training the PNN model. Best recognition efficiency of 97.05% achieved for Tulu characters from degraded paper documents, 98.12% for Tulu numerals and 88.07% is achieved for Tulu palm leaf characters using Deep CNN model compared to PNN. The results verified that the proposed methodology outperforms from the present state of art models.
Keywords: Handwritten Character Recognition; Palm Leaf; PNN; Deep ANN; Tulu; Wavelet.
Scope of the Article: Deep Learning