A robust CNN model for handwritten digits recognition and classification

At present, machine learning and deep learning models are playing a key role in various domains of image classification including handwritten numeral images. Handwritten digits recognition (HDR) via machine learning has received great attention of researchers due to ambiguity in learning methods. Hitherto, several researchers made significant efforts to improve the recognition process by selecting appropriate parameters and feature design. But there is always room for improvement in the conventional methods. Convolutional neural network (CNN) is a deep neural network most commonly applied to analyze image classification, object detection, face recognition, etc. To execute the task of HDR, robust CNN architecture is used for feature extraction and classification. In this paper, a Java-based framework known as Deeplearning4j (DL4J), is used for recognition and classification of the MNIST database. Results demonstrate that, compared to existing techniques, the proposed model is superior in terms of accuracy for handwritten digits recognition.