Recognition of Tifinagh Characters Using Optimized Convolutional Neural Network

The choice of a convolutional neural network architecture and its parameters to extract optical character recognition is very difficult and tedious. The main objective of this work is to build a very robust and very fast architecture for the recognition of the handwritten characters of the Amazigh language written in Tifinagh character. This work has two optimized architectures, one to recognize RGB images and the other to recognize binary images. Very promising results are obtained compared to other existing systems in terms of precision, recall, F-measure and execution time.