Threshold center-symmetric local binary convolutional neural networks for bilingual handwritten digit recognition

The writing style of the same writer varies from instance to in Arabic and English handwritten digit recognition, making recognition challenging. Currently, deep learning approaches are applied many applications, including convolutional neural networks (CNNs) modified produce other models, such as local binary (LBCNNs). An LBCNN is created by fusing a pattern (LBP) with CNN reformulating LBP convolution layer called (LBC). However, LBCNNs suffer random assignment 1, 0, or -1 LBC weights, less robust. Nevertheless, using another LBP-based technique, center-symmetric patterns (CS-LBPs), can address issues. In this paper, new model based on CS-LBPs proposed (CS-LBCNN), which addresses issues LBCNNs. Furthermore, an enhanced version CS-LBCNNs threshold (TCS-LBCNNs) proposed, issue related zero-thresholding function. Finally, models compared state-of-the-art proving their ability producing more accurate significant classification rate than existing models. For bilingual dataset, TCS-LBCNN enhances accuracy CS-LBCNN 0.15% 0.03%, respectively. addition, comparison shows that acquired second-highest MNIST MADBase datasets.