Analysis Of Geometric Characteristics Of The Human Face And A Beautiful Score Calculation

Analysis of facial aesthetics has become a new research field and has received more and more attention. People desire to pursue beauty. Beauty is everywhere and can not be avoided in our daily life. Therefore, the research on facial aesthetics has become a meaningful and challenging work.In this thesis, according to the facial beauty scores were calculated by this new problem, from the perspective of machine learning to explore the beauty of face geometric features, contribution to the facial beauty degree, predicting face beautiful fraction based on geometric features, including:1) Proving the concept of "the facial beauty score is quantifiable". The experiment shows that the facial beauty score can be obtained by quantitative analysis of the characteristics and features of a human face expression and appropriate prediction machine learning algorithms.2) Analysis of the facial geometries characteristics. That is how the abstract facial beauty quantifiable characteristic expression. We can propose three methods about extracting the facial geometric features based on traditional aesthetic, including the18-dimensional distance characteristics,68feature points based on AAM and the triangle area features, and their combinative features. Striving to be more accurate and complete extraction of influential facial beauty geometric information.3) Study on the prediction model of facial beauty scores. That is how to use the machine learning theory and method for constructing the reasonable prediction model. This paper studies several commonly used machine learning methods to learn and predict the concept of facial beauty scores, including the K nearest neighbor regression model, support vector regression model, multilayer perception neural network regression model, Gaussian Process Regression model, semi-supervised regression model based on Hessian energy.4) Verifying the relationship between the facial beauty scores and facial expression. The experimental results show that:when someone is happy, the geometric beauty scores are relatively high, and when angry, beauty scores are relatively low.