Experimental Analysis on Character Recognition using Singular Value Decomposition and Random Projection

Character recognition, a specific problem in the area of pattern recognition is a sub-process in most of the Optical Character Recognition (OCR) systems. Singular Value Decomposition (SVD) is one of the promising and efficient dimensionality reduction methods, which is already applied and proved in the area of character recognition. Random Projection (RP) is a recently evolved dimension reduction algorithm which can scale with large dataset. In this paper, we have applied SVD and RP as feature extraction technique for character recognition and experimented with k-Nearest Neighbor (k-NN) and Support Vector Machine (SVM) classifiers. Our experiments are conducted on MNIST handwritten digit database and Malayalam character image database. On both databases SVD features could achieve lower misclassification rate than RP based features. SVD features with SVM classifier using RBF kernel could achieve a misclassification rate of 2.53% on MNIST database and 2.87% on Malayalam character image database. Keyword-Character Recognition, Singular Value Decomposition, Random Projection, Pattern Recognition, Support Vector Machine, Dimensionality Reduction