基于深度学习的脸部年龄预测算法

Facial Age Prediction Algorithm Based on Deep Learning

为提高脸部年龄预测的准确性,在深度学习的基础上提出一种可有效预测脸部年龄的算法。通过对人脸图像进行预处理,获取左眼、右眼、鼻子和嘴巴四个部分的局部图像,利用迁移TensorFlow深度学习库中的Inception V4模型,提取脸部图像四个部分的多尺度局部特征,并将提取的局部特征使用串联方式相连接以得到融合特征,再将不同年龄的融合特征输入双向长短期记忆中,以学习不同年龄融合特征间的相关性,进而完成年龄预测。在公开数据集FG-NET和MORPH上的实验结果表明,该算法通过利用脸部多尺度融合特征和不同年龄融合特征间的相关性,能够显著提高年龄预测的准确性和鲁棒性。

In order to improve the accuracy of facial age prediction,this paper proposes an algorithm for effective age estimation based on a deep learning.Based on the preprocessed face images,the images of the parts around the left eye,right eye,nose and mouth are obtained.Then by using the Inception V4 model in the deep learning database of TensorFlow,the multiscale local features of these four face parts are extracted and connected in series to obtain the fused features.The fused features of different ages are input into bidirectional Long Short-Time Memory(LSTM)to learn the correlation between the fused features of different ages,completing the age prediction.The experimental results on the open data sets FG NET and MORPH show that the algorithm can significantly improve the accuracy and robustness of age prediction by using the correlation between the fused multiscale facial features and the fused features of different ages.