Analyzing Different Optimizers for the Recognition of Gurmukhi Handwritten Text by Employing CNN

Abstract

Automatic recognition of handwritten data is an important application area in various fields. Recognition of handwritten text, which is cursive in nature, is a cumbersome task. Moreover, handwritten text by different writers makes the recognition even more difficult due to the different writing styles of the individuals. In this paper, a CNN model employing different optimizers is proposed for the recognition of Gurmukhi handwritten dataset. For the purpose of classification and recognition, five different classes of Gurmukhi handwritten text have been created where each class has 1,000 handwritten samples. Results are obtained using four different optimizers: Adagrad, Adam, Adamax, and RMS prop. Maximum validation accuracy of 99.20% is achieved using Adam optimizer.