Character Recognition of Malayalam Palm Leaf Manuscripts Using Convolutional Neural Networks

A wealth of knowledge, tradition, and ancient culture can be found in the palm leaf manuscripts. These scriptures must be preserved to the greatest extent feasible because they are an essential component of our rich culture. To determine the optimal way for digitalizing the Malayalam language, an extensive literature survey was done on the various Character recognition techniques used in Malayalam as well as other languages. The proposed method transforms characters from ancient Malayalam palm leaf manuscripts into a computer recognizable text file format. The input of these systems should be images of palm leaf documents and the obtained output will be recognized characters in current version of Malayalam text. In this Malayalam character recognition system we integrated Ostu’s method based binarization and Horizontal, Vertical projection profile for segmentation. And for training the model, we used Convolutional Neural Networks (CNN) which simplifies the entire process by avoiding feature extraction. And we obtained 95.30% accuracy for the developed model.