Face illumination migration method based on generative adversarial network

Abstract

The invention discloses a face illumination migration method based on a generative adversarial network. The method comprises the following steps: (1) acquiring sample data of a front face image; (2) selecting two images from the sample data as a target face image and a reference illumination image respectively, wherein the classifier is used as the input of the generator, the generator outputs a re-illumination image, the discriminator feeds back the error of the real image and the re-illumination image to the generator, and the classifier feeds back the error of the identity information of the target face image and the re-illumination image to the generator; the generator, the discriminator and the classifier carrying out repeated adversarial training to obtain an optimal face illumination migration model; and (3) performing face illumination migration. According to the method, a generative adversarial network architecture and a loss function are adopted, and the model can effectivelyprocess local illumination details by utilizing an attention mechanism. 3D information of a human face does not need to be used in network training, the human face does not need to be aligned, end-to-end training is carried out, and a good re-illumination result is achieved.

Description

technical field [0001] The invention relates to the field of human face illumination migration and generation network, in particular to a method for human face illumination migration based on generative confrontation network. Background technique [0002] The lighting effects of face images are widely used in film production, virtual reality, game entertainment, etc., but it takes years of practical experience to master the lighting processing technology of images. Although image editing techniques can also be used to achieve face lighting processing, professional image editing tools will inevitably involve many complicated operations. Therefore, it is necessary to develop a more intelligent face lighting processing method. [0003] One processing method in the face lighting processing method is face lighting transfer, that is, transferring the lighting effect on the reference image to the target image. Aiming at the problem of face light migration, many scholars have prop...

AI Technical Summary

Problems solved by technology

[0004] The existing face light migration methods still have the following shortcomings: face alignment or 3D information of the face is needed to solve the problem, which greatly increases the complexity of the problem

Moreover, for face alignment and the 3D structure of the face, it is necessary to detect the key points of the face, and the algorithm may fail under some extreme lighting conditions, which is also a point to be considered

Images