Face privacy protection method based on generative adversarial network

Abstract

The face privacy protection method based on generative confrontation network of the present invention is characterized in that: the face de-recognition method based on generative confrontation network is respectively loaded in the workstation and the robot platform, and the feature model is trained on the workstation W ; When the camera on the robot platform captures the face image, for the face image that needs privacy protection, the robot platform applies the face recognition method based on the generative confrontation network to perform face de-identification, protect the privacy features of the face image, and ensure that the user visual privacy is not violated. The face recognition method based on generation confrontation network includes a generator for improving U-Net network G and 2 discriminators D 1 , D 2 , where the discriminator and generator consist of convolutional layers, residual blocks and self-attention layers. The invention has the characteristics of being able to reduce or eliminate the problems of model collapse and overfitting in the training process, improve the quality of generated images, and visually protect the privacy of images.

Description

technical field [0001] The invention relates to the field of information security protection, in particular to a face privacy protection method based on a generative confrontation network. Background technique [0002] In recent years, a large number of photos and videos have been recorded, stored and processed with the widespread use of mobile phones, tablet computers and other imaging devices. Although these visual devices bring convenience to people, unprotected images or videos will lead to privacy leaks and pose serious challenges to privacy protection. Face de-recognition is an important first step in visual privacy protection, so the problem of face de-recognition has received a lot of attention recently. To preserve visual face privacy, many researchers make the modified face images ineffective for face recognition methods by replacing or changing face regions in the images. [0003] Traditional face de-recognition methods mainly focus on removing identities from i...

AI Technical Summary

Problems solved by technology

But studies have shown that pixelation and blurring do not ensure the effectiveness of correct face de-identification

[0004] In recent years, neural networks can be used for privacy de-identification, however, methods based on convolutional neural networks (CNN) have two limitations: 1) The quality of generated images is not realistic enough

Therefore, it needs some labels to train the network, but it takes a lot of time and cost

[0005] At present, Generative Adversarial Networks (GAN) provide a new direction for privacy de-identification, however, the facial features of the images generated by this method are not obvious, and the value of the image cannot be preserved effectively

There are three limitations of GAN-based privacy de-identification techniques: 1) Most GAN-based privacy de-identification methods are semi-supervised algorithms that require a small number of labels, but labeling training networks requires a lot of time and cost

2) In the confrontation training of GAN, the generator and the discriminator can easily lead to the risk of mode collapse, overfitting and model failure; 3) The quality of the image generated by this algorithm is not realistic enough, and in the process of de-identification Does not preserve image properties

Images