Improving Occluded Face Recognition with Image Fusion

Abstract

The application of vision technology has penetrated into many fields such as tourism, education, government affairs, housing, Internet applications, etc. Among them, face recognition technology has been closely related to our lives, such as mobile phone unlocking, attendance check-in, residential access control, etc., resulting in an increasing demand for high success rate face recognition. The recent new crown virus epidemic has caused many people to wear masks to go out, and the masks covering the face will reduce the effective recognition rate. In order to improve the inconvenience caused by masks to the application of face recognition technology, we try to apply image fusion technology to improve the success rate of face recognition. We compared the recognition rates of six common image fusion technologies. These technologies use the Labeled Faces in the Wild database as the data source and the Iterative Weighted Regular Robust Coding (IR3C) algorithm as the evaluation method. The final results show that the use of these image fusion technologies can greatly improve the recognition success rate of images occluded by masks. In particular, the weighted average technique and principal component analysis give a stable performance on recognition accuracy of 99.6% and 99.0% in a relatively small-scale data set, while the LFW simulated masked face dataset provided by Wuhan University that did not perform image fusion Only has a recognition rate of 85.4%. The application of image fusion can solve some limitations in the use of face recognition and bring more convenience to various fields.