A New Optimization Approach to Improve an Ensemble Learning Model: Application to Persian/Arabic Handwritten Character Recognition.

Due to the advancement of technology, handwriting recognition has become more important than ever. As a result, several methods for document image recognition have been developed in the literature. This paper presents a new ensemble model based on the Feedforward Neural Networks (FFNN) to accurately recognize Persian and Arabic handwritten characters. As training and optimizing FFNN models have a significant role in obtaining optimal results, two optimization algorithms are integrated into the proposed handwritten recognition method. The Particle Swarm Optimization algorithm is integrated into the proposed model to improve the Neural Networks learning process. The FFNN architectures are further optimized using the League Championship Algorithm. The ensemble model is fed by a set of handcrafted features, including directional and intersection features, extracted from handwritten text. The proposed model is evaluated using three different datasets. Results obtained from the proposed models demonstrate higher accuracies compared to the state-of-the-art models.