According to Microsoft, facial recognition is getting better at recognizing people with dark skin tones. The number of wrong face recognition has been reduced by 20 times for men and women with dark skin. Also for women in general, the facial recognition makes nine times fewer mistakes.
Face detection in dark skin
Testing facial recognition systems from Microsoft, among others, showed that 35 percent of dark-skinned women were labeled by the system. How is this possible? This is due to limited datasets. Such a system learns to recognize faces by practicing. If there are relatively few dark female faces in a dataset, it is more likely that the system links the wrong gender to this. The system is as good as the data used to teach the system something. If the data, the faces to train the system, contain bias, the system will also take over this bias.
Improvement facial recognition
The system for recognizing faces has been greatly improved. This is because the system has been trained with more diverse datasets. Different skin tones, but also factors such as hairstyle and jewelry are included. This way the system learns to distinguish more different skin types.
The improvement is technically an achievement, but there is also criticism and skepticism about the system. Where is this going to be applied? Microsoft cooperates with ICE (US Immigration and Customs Enforcement), and people wonder how facial recognition of the skin color plays a role in this collaboration. Microsoft has stated that “it is not applied to separate children and parents at the border”. What it is used for is a mystery.