Facial beauty prediction fusing transfer learning and broad learning system

Abstract Facial beauty prediction (FBP) is an important and challenging problem in the fields of computer vision machine learning. Not only it easily prone to overfitting due lack large-scale effective data, but also difficult quickly build robust facial evaluation models because variability appearance complexity human perception. Transfer Learning can be able reduce dependence on large amounts data as well avoid problems. Broad learning system (BLS) capable completing building training. For this purpose, was fused with BLS for FBP paper. Firstly, a feature extractor constructed by way CNNs based transfer extraction, which EfficientNets are used paper, features extracted transferred FBP, called E-BLS. Secondly, basis E-BLS, connection layer designed connect BLS, ER-BLS. Finally, experimental results show that, compared previous methods existed, accuracy improved E-BLS ER-BLS, demonstrating effectiveness superiority method presented, widely pattern recognition, object detection image classification.