Ethiopic Base Characters Image Recognition using LSTM
Ruchika Malhotra, Maru Tesfaye Addis · 2021
Ethiopic scripts are widely used in Ethiopia by acting as a writing script in many languages. In the scripts, there is a shape similarity of characters and an abundant number of letters in the alphabet. The visual or shape similarity and a large number of alphabets are a problem for any optical character recognition system. To address this issue, in this study, we developed the Recurrent Neural Network-based character recognition model for 26 base Ethiopic characters. In the proposed model, we used our own collected 28 by 28 size image datasets, which has 8912 samples to train and test the recognition model and the Long Short-Term Memory networks used as the recognition algorithm. In the recognition model, a promising testing accuracy reported as a performance measure.