Indian Number Handwriting Features Extraction and Classification using Multi-Class SVM

Abstract

In this paper, an Indian Number Handwriting Recognition Model (INHRM) is proposed. Mainly, the proposed model consists of four phases which are the image acquisition, image preprocessing, features extraction, and classification model. Initially, the captured images of the handwritten Indian numbers were enhanced and preprocessed to obtain the skeleton for the interested object. The extracted features of the handwritten Indian numbers were obtained by calculating four parameters for each captured number sample, these parameters are the number of starting points, the number of intersection points, the average zoning which consists of four values, and finally, the normalized chain vector of length of 10 elements. So, the resulted 16 values of the four parameters were arranged in a vectors of length of 16 elements. These features vectors were used in the training and testing processes of the proposed INHRM model. Multi-class SVM (MSVM) approach is suggested for the classification phase. An accumulation of 600 samples of various handwritten Indian numbers styles has been gathered from a group of 60 students. These samples were preprocessed, features extracted, then delivered to the classification phase by utilizing 500 samples of them for training while the remaining 100 samples were used for testing of the MSVM-classifier model. The results showed that the proposed INHRM achieved relatively high percentage of exactness of around 97%.