Handwritten Digit Recognition Based on Support Vector Machine

Abstract

In this paper, we propose a handwritten digit recognition method based on Support Vector Machined (SVM). Firstly, some main features are extracted from the handwritten digital images (Euler Number, roundness, moment feature, crossing density, pixel density). Secondly, adopt the method of SVM for classification. This paper adds the feature values normalized, using radial basis function and Cross-validation important parameters. Our approach has been implemented with MNIST database and we have achieved an average recognition rate of 96.3%, the lowest single digit recognition rate of 93.5% when the training data are 100×10. Introduction Handwritten digit recognition is a branch of optical character recognition technology. The main research of the handwritten digit recognition is how to use the computer automatically recognize handwritten digits. Due to the extensive use of digital symbols (such as postal code, bank check, financial forms), the Handwritten digit recognition technology is widely attention. It has been one of the hotspots with high practical value in the field of pattern recognition. At present there are four main methods in the field of handwritten digit recognition. They are Bayesian Classification [1], Clustering Analysis [2], Neural Network [3] and SVM [4, 5]. A major problem in handwriting recognition is the huge variability and distortions of patterns. Elastic models based on local observations and dynamic programming such Hidden Markov Model (HMM) is not efficient to absorb this variability [6].The SVM is a machine learning algorithm developed based on the statistical theory, and it has advantages in solving small sample size and nonlinear pattern recognition problems. Image Preprocessing Feature Extraction Classification Fig. 1 Three main operations As shown in Fig. 1, there are three main parts of handwritten digit recognition (image preprocessing, feature extraction and classification). Different features have different impact on the recognition rate, the feature extraction plays a key role in the recognition. Some scholars focus on the study of the feature extraction [7, 8]. In this paper, we select the SVM as a classification method. In the next part, we will describe the selected features in detail. Meanwhile, in order achieve higher recognition rate we adopt some optimization steps before the data training operation. Feature Extraction In this section, several main features used by this paper will be introduced. Euler Number. Euler number is a topological parameters used to describe the object structure. Euler number is defined as: the difference between the connecting number and the number of holes. It can be expressed in the following formula: E L K = − (1) E represents the value of the Euler number, L represents the number of the connecting, K represents the number of holes. International Conference on Information Sciences, Machinery, Materials and Energy (ICISMME 2015) © 2015. The authors Published by Atlantis Press 941 Roundness. Roundness feature can be used to describe the complexity of the object boundary. Roundness obtained by the ratio of the circumference and area of the square. It can be expressed in the following formula: