Large-pose facial makeup transfer based on generative adversarial network combined face alignment and face parsing

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Facial makeup transfer is a specialized area of image style transfer.
• Generating high-quality makeup transfer results for images with large pose variations remains a significant challenge.

Purpose of the Study:

• To propose a generative adversarial network (GAN)-based algorithm for large-pose makeup transfer.
• To improve the quality of generated images in facial makeup transfer, particularly for challenging large-pose scenarios.

Main Methods:

• Introduction of a face alignment module (FAM) for key facial point localization (eyes, mouth, skin).
• Design of a face parsing module (FPM) and associated losses to extract facial features.
• Integration of facial features with extracted makeup style codes for transfer.
• Collection and construction of a dedicated large-pose makeup transfer (LPMT) dataset.

Main Results:

• Experimental validation on both traditional makeup transfer (MT) and the new LPMT datasets.
• Demonstrated superior image quality compared to existing state-of-the-art methods for large-pose makeup transfer.

Conclusions:

• The proposed GAN-based algorithm effectively addresses the challenges of large-pose makeup transfer.
• The method achieves improved image generation quality, outperforming previous approaches in complex pose scenarios.