个性化虚拟化妆效果迁移

Personalized Virtual Makeup Transfer

为克服当前虚拟化妆迁移中只能迁移单一化妆效果的问题,提出一种个性化虚拟化妆迁移方法.给定目标图像与多(单)个化妆后的样例图像,首先对脸部进行区域分类,将样例对象脸部与目标对象脸部进行匹配;然后通过二尺度分解技术获得脸部结构层、纹理层及色彩层,并使用梯度域技术修复样例对象纹理层,去除样例的脸部瑕疵;最后将每个子区域的化妆信息迁移至目标对象的对应区域,并无缝融合区域间的信息.实验结果表明,该方法不仅可以快速生成高质量的基于多个化妆样例的个性化虚拟化妆效果,同时具有和谐克隆的特性,可以获得良好的图像和谐克隆结果.

We propose an example-based personalized virtual makeup transfer approach for digital facial images in order to improve the flexibility of the traditional virtual makeup transfer method. First, given a target image without makeup and several example images with makeup, the algorithm defines different face sub-regions and warps example faces to the target face. Second, the algorithm uses a two-scale decomposition method to obtain the facial structural layer, texture layer, and color layers, and then modifies texture layers of examples to remove blemishes of example faces by using a gradient field method. Third, makeup information of each layer of examples is transferred to the corresponding layer of the target image in each sub-region through different ways to obtain a seamless result . Experimental results show that our approach is capable of not only generating high quality personalized makeup results based on multi-examples, but also being applied on harmonization cloning to obtain visually plausible cloning results.