A modified concept of PCA to reduce the classification error using kernel SVM classifier

This paper focuses on the mathematical technique PCA with the drawback of its mixing of data pixel. We have extracted principal directions of the covariance ellipse as done in PCA, but we will not blindly take the Eigen vectors corresponding to k largest values. Instead, we transform the data vectors into the new n– dimensional (n is dimension of old input space) vector space spanned by the Eigen vectors of the covariance matrix of the input data and then take one attribute at a time to perform classification. Then, we choose attributes corresponding to k largest accuracy measures this approach of Eigen vector selection shall prove to be more effective than the traditional one with the improved approach to the conventional method in which we have considered the feature vector corresponding to the kminimum error in order to reduce the dimensionality. The work has been implemented on more than 545 images of men and women to give efficient consequences in context of advance approach of PCA. Index term: Principal Component Analysis, Support Vector Machine classifier classification error Euclidian distance kenel svm , matlab