For some time now, police forces have been using facial recognition tools to locate people in photos posted online. This is particularly the case during demonstrations where these images can turn out to be real bombs to relate for some participants.
We thought that wearing a mask could paradoxically help by hindering the identification of people, but that does not stop everything. We explained to you in particular how Chinese company Hanwang Technology Ltd would be able to recognize people when they wear a mask that hides half of their face. For now, the idea is to help the authorities fight against the Coronavirus pandemic, but we also know that the regime uses facial recognition to suppress demonstrations.
A not infallible tool but promising
To prevent the faces of participants from landing in the basespolice data, researchers at Stanford University specializing in machine learning have developed a real anonymization tool called BLMPrivacyBot. This device is of course created as part of the Black Lives Matter (BLM) movement currently underway in the United States.
To do this, the system covers faces with a fist emoji which is also l emblem of BLM. Concretely, the tool detects faces but does not identify people. This model was trained to recognize faces from a crowd count data set of approximately 1.2 million people, which seriously improves its efficiency.
Need to protect protestors? Tweet at me to anonymize BLM protest photos! Original by @Official_TNIJ pic.twitter. com / Cx4mtvRlYo
- #BlackLivesMatter (@BLMPrivacyBot) 15, 2020
Scientists want to make this tool accessible to the general public and just tweet an image on Twitter to the @BLMPrivacyBot account to make it invisible a face with the emoji. As The Next Web points out, the device is not foolproof either. For example, a person can be identified from their clothing and it is very difficult to make all the images collected during a demonstration unusable. Eventually, the researchers even hope to see a platform like Twitter integrate this type of tool in an automated way among its functionalities.