Modeling of Hybrid CNN Framework for Handwritten Character and Digit Recognition System

Purpose: In this paper highlights the recognition of hand-written Content/character problems that have been proposed. This study shows the emerging and widely used in many vision-based applications in recent years. In this work, the drawback of conventional handwriting recognition utilizes the statistical characteristics of individual Content/characters, and different subsets to recognize the different Content/characters are overcome. Design/Methodology/Approach: To mitigate this problem several works have been motivated to incorporate a deep learning model for handwriting recognition and achieved significant performance improvement in recent times. In deep learning models, Convolutional neural networks (CNNs) are providing promising solutions to automate the distinct feature extraction from all given training samples. In this work pre-processing measures have been used to explore the boundary details of input samples for CNN-based handwritten character and digit recognition systems. Here, we proposed a hybrid novel CNN model which combines the detailed feature extraction for final classification. The proposed CNN framework is trained using the Modified National Institute of Standards and Technology (MNIST) database. Findings/Result: The performance validation includes both character and digits classification from given benchmark image sets. The validation also includes some self-build handwritten characters and digits for analyzing its resistance to the real-time capture. From the experimental results, it is observed that the variation inaccuracies are significant with the number of layers used in the CNN network. Finally, to validate the performance metrics of the proposed handwritten character recognition system it is compared with the existing state-of-the-art HCR models in terms of overall recognition accuracy. Originality/Value: In this work, the drawback of conventional handwriting recognition utilizes the statistical characteristics of individual Content/characters, and different subsets to recognize the different Content/characters are overcome. The objectives for accommodating handwritten content/characters written in different formats are difficult to accomplish irrespective of the methodologies used for classification