Researchers find image restoration can retain biometric recognition value

It is widely thought that facial recognition can only work on subjects where some minimal amount of the face is visible.

Three researchers from Guangdong Polytechnic Institute in China explicitly embedded “identity-preserving constraints” into a GAN framework to improve the recognition rate of restored facial images.

The researchers developed a facial image processing framework based on MS-CNN.

The partial or occluded face image is fed into a convolutional network, and resolution improved with upsampling.

With these conditions, facial recognition accuracy in noisy environments was 71.3 percent.