【摘要】 针对现有去妆算法研究没有考虑到皮肤纹理信息,导致去妆人脸不真实,细节模糊等问题,本文提出一种基于条件生成对抗网络的面部去妆方法。该方法通过使用人脸位置对应的配对数据,引入L1损失函数保留身份信息,并且在训练过程中加入VGG16网络提取图片特征,引入感知损失函数,确保皮肤的纹理真实相近。实验结果表明,与其他图像转换生成网络对比,该方法能够生成更加清晰、高质量的人脸图像。更多还原

【Abstract】 In response to the existing research on makeup removal algorithms that do not consider skin texture information, resulting in unrealistic and blurred details of the removed faces, the paper proposes a facial makeup removal method based on conditional generative adversarial networks. The method uses the paired data corresponding to the face position, uses the L1 loss function to retain the identity information, and adds the VGG16 network to extract the image features during the training process and introduces the perceptual loss function to ensure that the skin texture is realistic and similar. The experimental results show that the method can generate clearer and higher quality face images compared with other image transformation generation networks.更多还原