In an era where digital communication and image sharing have become ubiquitous, people post images with strangers identities on their social media all the time. Imagine yourself appearing in a social media post and you don’t like the way you are represented, how would you feel about it? We think it is very important to discuss about the protection of data privacy for those experience passive exposure on social media. This problem has two layers:
1) Is this legal to take photographs of strangers on the street?
Short answer is YES> The privacy rules surrounding photography generally boil down to freedom of expression and are covered by The First Amendment in the US and the European Human Rights Act in the UK and Europe. These rules allow you to take pictures for news and artistic or editorial expression of anyone in a public setting (excluding crime scenes) as long as you are not violating any other laws or posing a safety risk.
2) Is this appropriate to post these photos later on social media?
There is no short and clear answer to this question. Some people may feel more comfortable to share expose their faces or other identities on social media while others may be very concerned about it due to the data privacy considerations. There is a probability that the strangers in your photo don’t like posting any their selfies on any social media but you post it on your social media. As long as the photo is not a criminal scene and not inside a private space, there shouldn’t be any legal issues. But it is still unfair to these strangers, especially if the photos look not very “pleasant” . Inside the street photograph community, there is a debate on whether they should ask for permission when they take photos of homeless people or other vulnerable people. The passive exposure to the social media may bring extra harm to them.
SecurePix
To echo this concern on sharing images with strangers’ faces, we developed SecurePix. SecurePix leverages cutting-edge technologies such as deepfake, face recognition, and Generative Adversarial Network (GAN) algorithms to establish a robust framework aimed at safeguarding the identities of individuals portrayed in images shared across social media platforms. Our primary objective is to advocate for and contribute to the enhancement of data privacy by deploying advanced techniques to anonymize the involuntary identity exposure associated with the proliferation of personal images online.
The SecurePix image process pipeline is mainly consisted of three modules.
1) The first module is the face recognition module where we use Arcface model to detect faces in the source image and use another CNN model to recognize the gender and age range of the face.
2) The second module is the face generation module. We use pretrained StyleGAN3 model to generated faces. To evaluate whether the face has the same gender and age range as the detected face from the source image, we paraphrase a short text attribute with the gender and age range, and then use CLIP model to examine the matching score between the image and the text attribute. We will save the faces with high score for later use.
3) In the third module, we apply a pretrained face swapping algorithm to replace the original faces with the matched generated faces.
Examples
