A new method for face recognition with fewer features under illumination and expression variations

In this study, a new adaptive feature extraction method has been presented based on multi-dimensional discriminant analysis (MLDA) over multi-dimensional principal components. Proposed work has been aimed to design a method that can predict required number of features for a particular dataset. This method use only effective features which have better discriminant power in different dimensions of an image. In order to ease the pre-processing we controlled the variance in each mode to make the feature selection adaptive in different datasets with facial variance present in the image. The Experiments with different datasets has been performed in order to check suitability for larger dataset, with lesser computational cost and higher efficiency. Moreover, when support vector machine operated as classifier, proposed algorithm shows its superiority of recognition over previous known methods like PCA, PCA-LDA, MPCA.