基于对抗样本防御的人脸安全识别方法

Secure Face Recognition Method Based on Defense Against Adversarial Examples

本文使用一个像素攻击法和通用扰动生成攻击法生成对抗数据集,通过人脸识别模型对对抗数据集进行微调训练,使其对一个像素攻击法生成的对抗样本具有较高的正确识别能力和对通用扰动生成的对抗样本的具有完全正确的识别能力,实现人脸安全识别。

In this paper,the one-pixel attack method and the general disturbance generation attack method are utilized to generate a dataset of adversarial examples.After trained by the dataset,the face recognition model is slightly adjusted and exhibits a higher correct recognition rate to the adversarial examples generated by the one-pixel attack method and a completely correct recognition rate to the adversarial examples generated by the general disturbance method,realizing secure face recognition.