深度自编码器用于人脸美丽吸引力预测的研究

为了挖掘人脸美丽的内在本质,本文提出了基于深度自编码器的人脸美丽吸引力预测模型:首先利用大量无标签人脸图像数据对深度自编码器进行预训练,然后结合Polak-RibierePolyak共轭梯度反向传播算法对深度自编码器的权值进行微调,从而建立深度自编码器的人脸美丽特征提取模型.最后经过支持向量机(SVM)分类器对人脸图像进行美丽预测.实验结果显示SVM分类器预测的平均识别率为77.3%,表明深度自编码器用于人脸美丽吸引力预测是有效的.

A Study of Facial Beauty Attractiveness Prediction Based on the Deep Autoencoder

To explore the inner essence of facial beauty, this paper proposes a facial beauty attractiveness prediction model based on Deep Autoencoder. This study pretrains the Deep Autoencoder with a great deal of unlabeled facial image data, then fine-tunes the Deep Autoencoder with some labeled facial image data in the light of the Polak Ribiere Polyak Conjugate Gradient Backpropagation, builds up a facial feature extraction model for the Deep Autoencoder, and finally predicts the beauty attractiveness of human facial images with a SVM classifier. Experimental results show that the average recognition rate of the SVM classifier is 77.3%, indicating that the Deep Autoenceder is effective for predicting human facial attractiveness.