A shielded face recognition method and device

Abstract

The invention relates to a shielded face recognition method and device. The method comprises the following steps of training a face recognizer by utilizing a face image with an identity mark; traininga face generator, a global face discriminator and a local face discriminator in the generative adversarial network framework by utilizing the non-occluded face image with the identity mark, the corresponding occluded face image, the corresponding face occluded mark graph, the corresponding face semantic segmentation graph and the pre-trained face resolver in combination with the face recognizer;carrying out restoration processing on the input shielded face image by using a face generator to obtain a restored face image; and performing feature extraction or identity classification on the restored face image by using the face recognizer to obtain face features or identity marks, performing face recognition by using the face features, and identifying the identity information of the face byusing the identity marks. The three-party game is realized by adopting the face generator, the face discriminator and the face recognizer, and the shielded face can be repaired and recognized at the same time.

Description

technical field [0001] The invention belongs to the field of computer vision and deep learning, and in particular relates to a face recognition method and device under occlusion conditions. Background technique [0002] In real-world applications such as video surveillance and unmanned driving, face recognition is often disturbed by occlusion problems. Although the current face recognition models represented by deep learning have achieved high recognition accuracy without occlusion, the performance of these models often drops sharply when recognizing occluded faces. The reason is that, unlike unoccluded faces, occluded faces are difficult to recognize due to incomplete visual content and insufficient identity cues. Nevertheless, occluded faces are still recognizable for humans given partial content, suggesting that the human neural system has the ability to repair lost face cues from occlusions. Based on this fact, occluded face recognition methods have been proposed, whic...

AI Technical Summary

Problems solved by technology

[0004] However, in these existing methods, on the one hand, it is difficult to find a more robust solution to restore the lost face clues based on expression-based occluded face recognition methods; on the other hand, reconstruction-based occluded face restoration methods often Restoration and identification processing are separated, and the identity retention ability brought by the method is relatively limited

Images