【摘要】 随着人脸识别、人脸检测技术的发展，恶意用户批量收集人脸图像并进行身份识别、人脸替换等操作严重侵犯用户的个人权益。针对现有隐私保护方案对图像改动大且图像质量低的现状，提出一种去身份人脸图像隐私保护方案，通过建立人脸属性模板库，根据隐私保护强度筛选模板图像进行人脸融合，达到降低人脸识别正确率、去除人脸身份信息的目的，同时保持图像质量在较高水平。方案没有对攻击模型做过多假设，以LFW数据集为测试集进行隐私保护处理，并调用百度AI人脸对比接口进行人脸相似度检测。统计结果证明，方案能够降低人脸识别算法正确率，在去身份隐私保护的同时保持图像质量在较高水平。更多还原

【Abstract】 With the development of face recognition and face detection technologies, malicious users collect face images in batches and perform operations such as identity recognition and face replacement, which seriously infringing the users’ personal rights. For the current situation of considerable image modification and low image quality in existing privacy protection schemes, a de-identification privacy protection scheme for face images is proposed, where a face attribute template library is established and template images are filtered according to the privacy protection strength for the face fusion to reduce the face recognition accuracy and remove the face identity information as well as keep the image quality at a high level. In this scheme, the attack model is not assumed excessively and the LFW data set is the test set for privacy protection. Baidu AI face comparison interface is invoked for the face similarity detection. Statistical results indicate that the proposed scheme could reduce the accuracy of face recognition algorithm, and maintain the image quality at a high level while removing the identity and privacy protection.更多还原