A Novel BPSO-Based Optimal Features for Handwritten Character Recognition Using SVM Classifier

Abstract

The handwritten character recognition is a line of research within the image processing for which there have been developed many techniques and methodologies. Its main objective is to identify a character from a digitized image that is represented as a set of pixels. This paper presents a handwritten character recognizing framework with the help of rows and column-wise segmentation, feature extraction using discrete wavelet transform (DWT) and histogram of oriented gradients (HOG) features and support vector machine (SVM)-based classifier. Further research is extended for optimal feature selection using BPSO for the training of SVM to reduce the machine overhead maintain near to same accuracy. Achieved accuracy of the proposed method is 99.23%.