| With the development of social diversity, people have a higher pursuit for the beauty of face. The research of facial beautification has been paid more and more attention recently from the last century 90’s. At present, the facial beautification has attracted more and more researchers from the different fields, such as cognitive science, psychology, biology, artificial intelligence and so on. We analyze a lot of domestic and foreign periodicals and books from the related fields and then find that facial beautification has the following three difficulties and hot spots. Firstly, the combination of shape and texture is difficult. The change of the texture may change the appearance. The face may be the different person after the facial beautification. So the beautification about texture is a difficulty. Secondly, we must keep the balance of the similarity and the facial beautiful degree. The similarity is the facial similarity before and after the facial beautification. The beautiful degree is the beautiful score rated by volunteers that we invite. Finally, facial perception is easy but it is difficult for machine. So it is difficult to be characterization and identification by machine. Based on the difficulties and hot spots above, the main work of this paper is as follows.1. We apply active appearance model to the field of facial beautification. In the shape, we combine active appearance model parameters and the K nearest neighbor (KNN) method. Then the local KNN method is used to do fine tuning. The final face is more attractive after our method. In the texture, we apply the 2D Principal Component Analysis (PCA) to the active appearance model, improving the efficiency and reducing the complexity. The model uses the information of facial shape and texture at the same time and keeps the balance of similarity and beautiful degree.2. We apply Support Vector Regression (SVR) method to facial beautification. We create a standardized image database with feature points and score. The photos in this database are trained by the method of machine learning and we obtain the facial beautification energy function. Then the method of minimizing is used to calculate the beautification face. Support vector regression method is proved by the experiment that it can be applied to the facial beautification. Finally, we compared the two facial beautification algorithms introduced above.3. We apply the local beautification algorithm to color mouth texture. Firstly, we build a template for lips. Secondly, we turn non-rigid transform between lips and template into rigid transform of the triangular subdivision and fit lips edge profile at the same time to make the curve on the edge of the lip smoother. Finally, we complete the lip make-up. |
