Face recognition poses great challenges due to different variations of pose, illumination, expression, resolution, and motion in employing supervised machine learning model on basis of low level features on each training and testing samples contains a set of face images for identification and verification based secure authentication applications. Those methods using supervised and unsupervised learning methods remain undesirable due to some invariant feature of the human beings. In order to mitigate those issues, a novel neural network architecture on basis of probabilistic neural network using optimization of grey wolf optimization in proposed in this article. Probabilistic Neural Network uses pattern layers and summation layer to classify the features of the test images with support of the radial basis kernel function. In this paper, image preprocessing is carried out using image enhancement and normalization through Contrast limited Adaptive Histogram Equalization (CLAHE). Pre-processed image is employed to viola Jones segmentation which segments the image into four segments and it is represented as integral image. Integral Image contains many Harr like features discriminators is extracted efficiently using principle component analysis. Features namely edges, line and four sided features is suppressed to generate the less no of optimal features for recognition using grey wolf optimization as multiple correlated tasks. Grey wolf optimization process the features with respect to fitness function on basis of the alpha (α), beta (β) and delta (δ). Finally Probabilistic Neural Network has been carried out with optimal invariant feature representing the landmark points, visibility factors, pose factor and gender factors as multitask approach. Class membership is established for recognition with losses mitigation through correction component on the optimal features. Extensive experiments shows that the proposed architecture are able to capture both global and local information in face and performs significantly better than many existing machine learning algorithms on following measures recognition rate and computation cost. Performance evaluation of the proposed model is carried out on the Yale dataset with 10 fold validation to measure the accuracy and efficiency model against the conventional approaches.

Face Recognition, Principle Component Analysis, Viola Jones Segmentation, CLAHE, Probabilistic Neural Network, Grey wolf Optimization, Integral image, Class membership, Fitness function