Near infrared-visible light face image synthesis method based on comparative learning and StyleGAN2

Abstract

The invention discloses a near infrared-visible light face image synthesis method based on comparative learning and StyleGAN2, and the method comprises the steps: 1, collecting paired near infrared-visible light face images, and carrying out the unified face detection and normalization preprocessing, thereby obtaining a data set image; 2, introducing a comparative learning mechanism, and constructing a generative network model comprising a generator based on a StyleGAN2 structure, a discriminator and an image multilayer feature extraction block; 3, combining with a proper loss function and an optimization function, and training by using the training set image to generate a network model; and 4, inputting a near-infrared face image to be tested to test the generative network model, and finally synthesizing a corresponding visible light face image. The synthesized visible light image is closer to a real image, and the facial edge details and skin color information of the face image can be better restored, so that the visual effect of the synthesized image and the cross-modal face recognition performance are improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a near-infrared-visible light face image synthesis method based on contrastive learning and StyleGAN2. Background technique [0002] Near-infrared image sensors are widely used because they can well overcome the influence of natural light, work in various lighting conditions and night scenes. However, in the field of criminal investigation and security, near-infrared face images usually cannot be directly used for face retrieval and recognition, because the single-channel images acquired by near-infrared sensors lack the natural color of the original image, which is very unfriendly to human vision. Compared with real visible light face images, the face recognition performance of near-infrared face images is also poor. Therefore, converting the near-infrared face image into a visible light face image and restoring the color information of the face image will h...

AI Technical Summary

Problems solved by technology

[0004] However, near-infrared face images are different from other near-infrared images. If details such as face contour and facial skin color are distorted during image conversion, the visual effect and image quality of the synthesized face image will be greatly affected.

Images