Chinese researchers test point cloud-based facial age estimation process

A team out of Peking University has published a paper in Science China Life Sciences on  “establishing a deep learning model for age estimation using non-registered 3D face point clouds.”

A point cloud is a discrete set of data points in 3D space. A summary of the paper in TechXplore says the researchers trained their model on over 16,000 instances of 3D face point cloud data, to investigate “an approach for facial data masking that preserves age-related features using coordinate wise monotonic transformations.” The algorithm can “isolate age-related facial features from identifiable human faces” and “recognizes the rotational invariance of human faces.”

In other words, the algorithm can distort faces without changing the relative positions of basic elements, but in such a way that deep learning models can still accurately and consistently perform successful age estimation on faces in various scenarios. In testing, the method achieved an average absolute error of 2.5 years.

“Despite the immense value, the human face, as a hard biometric, is easily accessible and counters the hurdle of data security once collected or shared,” says the paper. “The occurrence of facial data leakage and spoofing events has prompted intense ethical security and privacy concerns.” The research notes that, “to date, there have been limited attempts to apply deep learning directly to 3D face point cloud data in the field of face recognition and facial expression detection.”

The team’s findings have led them to propose a facial data protection guideline, which “aims to provide a theoretical foundation for managing facial data centers or public datasets.” In conclusion, they say the study “leads to a facial data protection guideline that has the potential to broaden public access to face datasets with minimized privacy risks.”

Article Topics

3D point clouds  |  age estimation  |  biometrics  |  biometrics research  |  deep learning  |  face biometrics