基于深度学习的人脸图像加密算法研究

Research on Face Image Encryption Algorithm based on Deep learning

摘要:

图像作为视觉传达的重要信息载体，以一种直观、形象的方式向受众传递信息。但是，图像会在不知不觉中带来个人隐私信息泄露等安全隐患。本文从保护图像中隐私安全角度出发，深度融合人脸检测、人脸对齐方法以及混合混沌序列的图像加解密算法，提出了一种基于深度学习算法的人脸图像信息加密算法，即FIIE（Face Image Information Encryption ）算法，用于保护图片中的面部核心部位隐私信息。FIIE算法的具体描述如下：首先，采用WLDER FACE数据集中的人脸图像对MTCNN模型展开训练，并利用训练好的模型根据人脸特征点获取图像中人脸所在的矩形框坐标；然后，通过上述人脸区域坐标生成掩膜，运用生成的掩膜使原图与Logistic混沌序列做位运算，最后，对图像中人脸特定区域的加密。通过实验表明，本算法可以准确识别图像中人脸信息特定区域，实现对图像中面部信息的有效加密，保障用户的隐私安全。

Abstract:

As an important information carrier of visual communication, images convey information to the audience in an intuitive and vivid way. However, the image will unknowingly bring personal privacy informatidisclosure and other security risks. From the perspective of privacy protection in images, this paper deon eply integrates Face detection, face alignment methods and Image Encryption and decryption algorithms mixed with chaotic sequences, and proposes a face image Information encryption algorithm based on deep learning algorithm, namely FIIE (Face Image Information Encryption) algorithm. Used to protect the privacy of the core parts of the face in the picture. The specific description of FIIE algorithm is as follows: First, the FACE image in the WLDER FACE data set is used to train the MTCNN model, and the trained model is used to obtain the rectangular frame coordinates of the face in the image according to the face feature points. Then, a mask is generated by the above coordinates of the face region, and the generated mask is used to perform bit arithmetic between the original image and the Logistic chaotic sequence. Finally, the specific region of the face in the image is encrypted. Experiments show that this algorithm can accurately identify specific areas of facial information in images, realize effective encryption of facial information in images, and ensure the privacy and security of users.