CNN -LSTM Based Approach for Recognition of Devanagari Manuscripts
Aditi Moudgil, Saravjeet Singh, Bhavna Sareen · 2022 International Conference on Data Analytics for Business and Industry (ICDABI) · 2022
Devanagari scripts are a good source of information as it consists medieval and prestigious content which is very useful for the upcoming generations. Digitization of these scripts becomes one of the most important task to make the content available. Pattern recognition emerges its one of the applications as Optical character recognition. For the character recognition from manuscripts, the authors have used CNN which gives higher character recognition accuracy. This paper implements Convolution Neural Network + Long Short Term Memory for Devanagari character recognition. The total of 250 manuscript pages were considered and the dataset of multiple characters were divided into 33 classes of basic characters. The proposed model was run using 15:85 and 25:75 as test:train ratio of characters. Also, the number of epochs were varied for better recognition accuracy. The authors observed the best recognition accuracy of 93.63% when the proposed model was run with 25:75 test:train data on 27 epochs.