Assessment Of Facial Beauty For Oriental Female Based On Oriental Aesthetics

Since aesthetics has become a discipline,it has not really absorbed a large number of deep and wise thoughts in eastern aesthetics.The automatic evaluation of Oriental facial beauty is urgent for people to dig.Therefore,only to speed up the study of facial attractiveness of the eastern face can make facial beauty estimation more complete.In this paper,we present a complete facial evaluation algorithm based on Oriental women.The contributions of this paper are summarized as follows:(1)As a necessary precondition of facial beauty estimation,the collection of oriental facial beauty images is of good importance.We collected a dataset including oriental facial images and scores.Then we conduct facial beauty estimation.Because the images in the database are not standard facial images,there is variations in head pose and illumination etc.Therefore,before the feature extraction,we need to preprocess the input images to highlight the face region,so as to get the normalized images.Image enhancement methods are used to highlight the human face details and 83 facial landmarks are detected using face++.(2)Oriental facial beauty can not be calculated directly with computer.We collect a set of evaluating rules of oriental facial beauty and extract global and local facial informations according to these rules.We first extract geometric informations to get various local ratios of facial image.Then we extract the texture and color informations on the global perspective.Thus we get two sets of vectors to express facial beauty information.(3)In order to solve the problems such as large time consuming and poor generalization ability,a new method for estimating the facial beauty of Oriental women based on Extreme Learning Machine and double-deck decisions is proposed.Experimental results show that our method is 3.8-11.8%higher than traditional classifiers including BP and SVM.(4)To deal with the problem that facial beauty estimation result is difficult to evaluate.We compare the experimental performance of our method with others through Pearson's,MAE and pairwise comparison.Experimental results show that the proposed method achieves better performance which Pearson Correlation Coefficient is at most 0.68,MAE is within 5.5 and 88.82%accuracy in pairwise comparison.