Comparative Analysis of Handwritten Text Recognition using CNN and SVM

Abstract

Handwritten Character Recognition (HCR) is an active area of research in the recognition domain. Many handwritten character recognition systems have been put forth in recent years for real-world applications that require high identification accuracy and dependability. Many of these organizations, like the banking and healthcare sectors, demand extremely accurate HCR. HCR systems are also used in newly developing fields, such as the creation of electronic libraries and multimedia databases, where handwriting data entry is necessary. The goal of the study is to evaluate the models of a potent Convolutional Neural Network (CNN) and a Support Vector Machine (SVM) for the most accurate recognition of handwritten characters. Due to its receptive field, CNN can instantly learn local features. Each layer's output serves as the following layer's input. The Support Vector Machine (SVM)'s primary function is to represent a multi-dimensional dataset in a space where data items from various classes are divided by a hyperplane. The SVM also seeks to reduce generalization errors resulting from unobserved data. The topic of this paper is understanding Kannada numerals. The results show that CNN is the most optimal machine learning technique to classify handwritten text with an accuracy of 98.85 percent whereas SVM shows only 98.62 percent. CNN model outperforms the SVM model by 0.23 percent.