An approach to numeral recognition based on improved LDA and Bhattacharyya distance

This paper presents a novel pattern classification approach-improved LDA and Bhattacharyya distance based classifier for numeral recognition. Improved LDA is used as a dimensionality reduction technique that can estimate the within-class covariance and between-class matrices more accurately for classification purposes. Bhattacharyya distance utilizes the distribution characteristics of the samples in each class to improve the recognition performance. Experimental results conducted on UCI database show that the improved LDA combined with Bhattacharyya distance process achieves a good recognition performance using low feature dimensions.