基于双激活层深度卷积特征的人脸美丽预测研究

目前，人脸美丽预测存在数据规模小、分类难度大、深度特征研究不足等问题.为此，本文提出基于双激活层深度卷积特征的人脸美丽预测研究的解决方案.首先，采用数据增强和人脸对齐方法来增加训练集的样本数量和提高数据库的数据质量.其次，提出一种双激活层改进CNN模型，使其更适合人脸美丽预测应用.实验结果表明，本文所提方法在分类和回归预测方面均大幅度优于传统人脸美丽预测方法；同时，在主流的CNN模型中取得了较好的实时性和准确性，基于2000测试集的分类准确率达到61.1%，回归相关度达到0.8546.因此，双激活层在深层人脸美丽特征学习中发挥了重要作用，可广泛应用于人脸图像识别与处理.

Research of Facial Beauty Prediction Based on Deep Convolutional Features Using Double Activation Layer

At present,facial beauty prediction is facing the problems,in which data is insufficient,the face image is hard to classify,and the deep feature lacks research.To solve these problems,a solution to facial beauty prediction research based on double activation layer depth convolution feature is proposed.Firstly,we use the method of data augmentation and face alignment to increase the number of samples in training set and improve the data quality of database.Secondly,we propose a double activation layer (DAL) to design a CNN model that is more suitable for facial beauty prediction.Experimental results based on 2000 test set show that the method proposed is superior to the traditional method of facial beauty prediction both in classification and regression.In addition,the proposed method achieves better results and real time performance than the state-of-art CNN model,in which rank-1 recognition rate is 61.1% and the Pearson correlation coefficient is 0.8546.Consequently,the DAL method plays an important role in deep facial prediction learning,which can be widely used in face recognition and image processing.