Face illumination processing method based on Retinex decomposition and generative adversarial network

Abstract

The invention discloses a human face illumination processing method based on Retinex decomposition and a generative adversarial network. A framework comprises an illumination decomposition module, a human face reconstruction module, a discriminator module and a human face verification module. The illumination decomposition module is composed of a convolutional neural network, inputs a pair of faceimages, and decomposes the face images into reflection components and illumination components through unsupervised learning; the face reconstruction module is composed of a coding and decoding convolutional neural network, inputs a reflection component, an illumination component and a target illumination level label containing a low-illumination face image, and can adjust the illumination component of the low-illumination image to a target illumination level; the discriminator module discriminates the authenticity of an input face image through adversarial learning and classifies illuminationlevels; the face verification module comprises a pre-trained face classifier to ensure that the generated face image and the target face image have the same identity information. The method is high in robustness and good in face reconstruction effect, and can be suitable for face illumination processing under the condition of dark illumination at night.

Description

technical field [0001] The invention belongs to the field of image processing and pattern recognition, and relates to a face illumination processing method based on Retinex decomposition and generation confrontation network. Background technique [0002] Face recognition has long been one of the hot research topics in the computer field. In recent years, this technology has been widely used in social multimedia, public safety, intelligent monitoring and other fields. The current face recognition algorithm still has some difficulties in practical application that still need to be solved, among which the illumination change seriously affects the robust performance of face recognition, especially in the scene with dim light at night, the existing algorithm is difficult to recognize Face identity. Therefore, the face lighting processing algorithm has important research significance and application value. [0003] The existing face illumination processing algorithms can be mai...

AI Technical Summary

Problems solved by technology

[0004] (1) The performance is not good under harsh lighting conditions at night, the processing effect on local shadows is poor, and the lighting effect cannot be completely eliminated

[0005] (2) The processed face image is prone to distortion and distortion, and the recovery effect of face detail texture features is not good

[0006] (3) The processed face image loses part of the face identity information, so that the existing face recognition algorithm still cannot recognize the processed face image

Images