Exploring Attractive Faces: General Versus Personal Preferences

Human beings have never stopped on the pursuit of beauty since ancient times. "What is beauty? How to become beautiful?" is always the investigated topic of aes-thetic researchers. Along with the rapid development of image processing technol-ogy, it becomes possible to evaluate facial attractiveness based on the machine learning method. Base on the facial attractiveness evaluator, researchers have proposed larges numbers of algorithms to enhance facial attractiveness. But most of these algorithms just learn a average beautification model, which can only returns one beautified result, and cann’t meet different aesthetic standards. That is to say, the existing method just considers about the generality aesthetic standards, but lacks to study the personality aes-thetic standards. Based on the existing facial geometry enhancement algorithm which just study the generality aesthetic standards, this paper do a further research on person-ality aesthetic standards.This paper’s main works is showed as follows:Firstly, introduce what is involved in the existing facial attractiveness enhance-ment, which contains face image database collection, human rating, face key points extraction, Support Vector Regression (SVR) which used as a study algorithm, face de-formation, beauty uplifting and so on. This paper constructs a facial image database, and does human rating for it. To acquire more accurate facial feature, this paper proposes a facial key points extraction method with higher precision than existing methods.Then, analyze the shortage of existing facial attractiveness enhancement, and cre-atively propose two methods to extract the generality and personality aesthetic stan-dards. One is based on the idea of low-rank sparse matrix decomposition, which de-compose the score matrix into two part:low-rank and sparse, low-rank is corresponded to generality, and sparse is corresponded to personality. The other is based on the idea of recommender system, which also decompose the score matrix into two part:low-rank and low-rank, the first low-rank part is corresponded to generality, while a linear combi-nation of the low-rank part is corresponded to personality. Compared with the existing method on the pearson correlation coefficient standard, this paper draws a conclusion that aesthetic need to consider the personality.Finally, implement a real time facial attractiveness enhancement system which is based on generality and personality aesthetic standards, and users can explored their sat-isfactory results in real time based on the generality and personality aesthetic standards basis which had been studied.Facial attractiveness enhancement has an expansive developing foreground. Nowa- days, smart phones are very popular, many people like to take picture of themselves and upload to WeChat circle of friends. If the pictures can be beautified, they may attract more people’s attention. Facial geometric beautification technology is corresponded to plastic surgery in practical life, this technology can provide reference to the plastic surgeon, and plastic surgery before cosmetic surgery. In making animation or game characters, facial beautification can also provides producers with auxiliary reference.