Research On Personalized Facial Attractiveness Evaluation Algorithm Based On Multi-dimensional Feature Fusion

The human face is one of the main carriers of aesthetic evaluation between people.In social situations,the human face undoubtedly plays a very important role.In this thesis,face attractiveness refers to the degree to which others like this person because of f'acial features.Consensus face attractiveness represents the unified aesthetic standard of the public,and personalized face attractiveness represents each person's different aesthetic standards.The research on the evaluation of personalized face attractiveness can be applied to personalized social recommendation,beauty beautification,image optimization processing and human-computer interaction.It has important practical significance and research value.Through the analysis of the current research on the attractiveness of face,it is found that the main problems in the field are:the accuracy of the existing f-ace attractiveness evaluation model is low,and the feature extraction method that effectively represents the attractiveness of the face needs to be further studied;There are few training samples for personalized face attractiveness evaluation,and direct use of deep learning methods cannot achieve satisfactory results.Aiming at the problems and challenges in the current personalized face attractiveness evaluation research,this thesis has proposed a personalized face attractiveness evaluation algorithm based on multi-dimensional feature fusion,and compared relevant experiments.The main contributions are as follows:First,in terms of feature extraction,this thesis has proposed a new feature extraction algorithm,which extracts the typical global and local features of face attractiveness,such as global 83 key points of face,17-dimensional geometric features,Gabor texture characteristics,etc.They are combined by feature fusion algorithm and finally combined with the face features automatically extracted by the convolutional neural network.The effectiveness of the face attractive feature extraction algorithm proposed in this thesis is verified by experiments.Second,the consensus facial attractiveness evaluation features were introduced into the model,and a consensus and personalized evaluation feature fusion algorithm was proposed to improve the performance and training efficiency of the model.Experiments were carried out on the SCUT-FBP-500,FaceScrub,and SCUT-FBP-5500 databases,and compared with the existing research results of face attractiveness evaluation.The proposed algorithm achieves better evaluation performance.The Pearson correlation coefficient between the output of the personalized face attractiveness evaluation model and the real evaluation data reached 0.91,and the mean absolute error was 0.17.