Nonlinear Face Classification with Modified 2DDTW

Facial physical appearance normally have several variations which occurred due to changes in expression, illumination, occlusion, head pose, and aging. In actual, human eyes can able to justify the authenticity of a person by the use of single image per class. In this paper, a new framework is proposed for nonlinear classification of face images with only one training image per class. Histogram Equalization is used for contrast stretching, Gabor wavelets and Kernel 2D PCA is used to extract local and nonlinear features and those are invariant towards orientation & spatial locality. Those features are fused with PCA fusion and then Two Dimensional Dynamic Time Warping is used for the classification of those feature vectors. The constraints Continuity, Monotonicity and bounded properties of DTW will identify the non linear optimal path between two dimensions of feature vectors simultaneously. The proposed method has evaluated on standard bench mark face databases like ORL, Grimace & Yale and yielded better performance such as 91.35%, 98.5% and 97.16% respectively.