A Multi-graph Fusion Based Manifold Embedding for Face Beauty Prediction

Automatic facial beauty prediction is an interesting research topic in computer vision and aesthetic medicine. Most of the existing FBP methods rely on supervised solutions based on geometric features or deep features. Recently, multi-graph fusion techniques have been used to construct more accurate graphs which better represent the data. In this work, we propose a semi-supervised manifold embedding method in which multiple graphs with geometric features, deep features and label information are constructed. The proposed method fuses the geometric features with deep features to generate a high-level representation of a face image. Moreover, our method incorporated the label space information as a new form of graph, namely the Correlation Graph, with other similarity graphs. Furthermore, it updated the correlation graph to find a better representation of the data manifold. The experimental results on the SCUTFBP-5500 face beauty dataset demonstrated the superiority of the proposed algorithm compared with other state-of-the-art methods.