用于人脸识别的多年龄人脸图像合成

Synthesis of Face Image with Age Variations

为了解决人脸识别中识别率随年龄变化急剧下降,提出了一种多年龄人脸图像合成方法。首先将测试图像表示为形状向量和纹理向量,之后用渐进智能神经网络(IINN)估计测试图像的年龄,然后利用事先训练的查找表计算目标年龄的形状向量和纹理向量,最后利用基于三角形的仿射变换把目标年龄的形状和纹理结合起来,生成目标年龄的人脸图像。实验结果表明,该方法有效地由1幅图像重构不同年龄时期的人脸图像,有效“改变”人脸图像的年龄。

The performance of most face recognition system degrades significantly with age changing. In order to solve this problem,an automatic face image synthesis with age variations is proposed. First,a shape vector and a texture vector are extracted to represent a facial image by projecting it in the eigenspace of shape or texture. Then an intelligence increasing neural networks（IINN） is used to estimate the age of the test image. After this,the synthesized feature vectors at target age are generated by the estimated age and the lookup table of feature vector and age. Then the shape and texture are reconstructed in eigenspaces. At last the facial image at target age is synthesized by combining the shape and texture using triangular based affine transform. Experiments show that the proposed method can effectively ＂change＂ the age of face images ,and synthesize the face image at different ages.