Numeric Digit Classification Using HOG Feature Space and Multiclass Support Vector Machine Classifier

Abstract

Pattern recognition is one of the major challenges in statistics framework. Its primary goal is to extract efficient feature and accurately classify the patterns into categories. A well-known and vital application in this field is the handwritten digit classification and recognition where digits have to be assigned into one of the 10 classes using some classification method. There are several approaches for handwritten digits classification and recognition. This paper proposed an efficient image appearance feature based approach which process the acquired digit image using Histogram of Oriented Gradients (HOG). HOG is a very efficient feature descriptor for data discrimination and very stable on illumination variation because it is a gradient based descriptor. For the efficient classification of the HOG features of numeric digits, a linear multiclass Support Vector Machine (SVM) classifier has been proposed, because it has better responses for nonlinear classification cases also. Mixed National Institute of Standards and Technology (MNIST) hand written numeric digit dataset has been used to test the classification accuracy of the proposed numeric digit classification system. For the implementation and testing of proposed system MATLAB 2015 (a) software platform has been used. The proposed system has been evaluated against the Neural Network based classification system. The classification and recognition efficiency of the proposed system and NN classifier based system has been evaluated using True Recognition Efficiency (FRE) and False Recognition Rate (FRR) parameters.