Bio Inspired Intelligence Techniques and Deep Classifiers for Human Face Shape and Race Classification

Abstract

The facial shape classification is particularly constructive for various purposes, like selecting haircut shape, selecting facial makeup, selecting sunglasses or even for electing suitable shapes. The procedure of classifying facial shapes is done by varied phases like capturing pictures in camera, outlining the face, counting the length and width of the face, cheekbones, jaw, and forehead. The facial shape is separated into six forms, such as: round, oval, rectangle, diamond, square and triangle. In addition, the literatures demonstrate the impairments in face shape classification related with numerous negative consequences, like more time consumption, inaccurate outcomes etc. Up to now, a number of researches have previously existed depending upon face shape classification. This research work contributes three major works. In the contribution 1, a new face shape classification scheme is build up with optimized Convolutional Neural Network (CNN). By CNN, the features are pre-trained and derived by CNN instead of deploying a unique feature extraction technique. In CNN, a novel custom layer is introduced that integrates a dragonfly optimization algorithm. At last, the confusion chart matrices for developed and conventional technique offers the categorized five groups of facial shapes such as quotheart, oblong, oval, round, and squarequot. Finally, the performance of proposed work is compared over extant ones with respect to certain measures. In contribution 2, a novel automated face shape classification scheme is introduced via Detection, Pre-processing extraction of relevant features and Classification. During face detection procedure, major objects in face like nose, eyes, etc are identified that is performed by means of Viola-Jones (VJ) model. In addition, Histogram Equalization (HE) is deployed during pre-processing phase for improving the image contrast. The categorization of facial shapes is done via a hybridized classifier, which connects CNN and Neural Network (NN). For carrying out classification via CNN, t