Various Approaches of Convolutional Neural Network-Based Recognition of Handwritten Devanagari Characters
Prashant Sopanrao Kolhe · 2023
In the past, handwriting recognition systems based on handmade qualities and a great deal of historical data. An OCR system that depends on these specifications is challenging to train the HR research that has produced ground-breaking results recently is centered on deep learning techniques. However, the rapid expansion in the volume of written data and the accessibility of massive computing resources call for an improvement in detection performance and call for further research. The most effective approach to solve the handwriting identification problem challenges is to use CNNs, which are particularly adept to address the issue of handwriting recognition automated extraction of distinctive traits. The goal of the proposed study is to investigate different design choices for handwritten digit recognition using CNN, such as the quantity of layers, phase size, facility available, kernel size, padding and dispersion. Our objective is to achieve comparable accuracy using a pure CNN architecture. As a result, a CNN design is suggested to attain accuracy that is even greater than ensemble systems while also reducing operational complexities and expense.