Face Beauty Analysis Method Based On Geometry And Texture Features

With the development of computer technology, more and more scholars used computer technology to research the face aesthetics and tried to find the "password" of the human facial beauty and obtain quantitative criteria of human beauty face. In the prediction of facial beauty, the scholars mainly acquired facial geometrical or texture features, and then combining with some machine learning models to study. But there was little research on the fuse with geometric and texture features for face aesthetics. In the face beautification, the digital image processing technology was used, and they achieved some research results, but researchers established the face aesthetics analysis systems were relatively simple on functions. In this dissertation, it was research the method about facial aesthetic analysis based on geometrical and texture features. The main research work was as follows:In the prediction of facial beauty, there were mainly includes two contents which were about extraction of the effective features to describe facial beauty and establishing a learning model for predicting facial beauty. In our database, based on the ASMs model, the geometric features of human face were constructed and the 21 geometrical features of the face were obtained by cross validation. Then based on these geometric features and using KNN learning model to analyze facial beauty, the good correlation coefficient value was obtained. For prediction of facial features based on texture features, a simple and efficient block LBP model(BLBP) had been proposed, and we obtained the optimal sub block number was 64 by experiments. In this parameter, the better correlation coefficient value was reached. Final, based on the fusion with geometric and texture features to analysis facial beauty, the correlation coefficient value was further improved.In facial beautification, this dissertation mainly researched facial skin and facial shape beautification. For the skin beautification, the improved multi-level median filtering model has been used. The experimental results showed that this model could be effective in the facial wrinkles, spots, acne, etc. And for the shape beautification, the improved algorithm based on the moving least square method have been proposed, which could adaptively find similar faces in the training sample, and then established the corresponding model to achieve shape beautification.Finally, the PC platform based on the face esthetics analysis system has been designed and implemented, the function of this system included face detection, facial beauty prediction, facial skin and shape beautification. The reliability of the results of the system was tested in the end. The reliability of face beauty prediction was achieved by 89.7%. The reliability of the face shape beautification was achieved by 80.0%. In the end, the reliability of the face texture beautification was more than 95.0%.Experiments showed that this dissertation has achieved some research results in the face aesthetics analysis. In the prediction of facial beauty, the fusion of the geometry and texture features could promote the prediction of the facial beauty. In the facial beautification, the improved algorithm based on moving least square method could make the result face more realistic. Final, this dissertation realized a complete facial beauty analysis system on PC platform.