一种高保真人脸图像妆容移植方法

摘要

为满足虚拟化妆体验的需求，提出一种利用抑制光照和边缘自适应滤波实现数字人脸图像妆容高保真移植的方法。通过增加图像光照抑制的预处理提高颜色的连续性，采用快速三角形坐标变换实现特征归一化，使人脸几何结构与纹理信息的分离。通过颜色空间变换将纹理信息分解成亮度和颜色等图层。根据人脸特征区域将亮度图层划分成不同融合比例的区域。对亮度图层混合时，结合边缘自适应滤波器，进而实现针对低频信息的自适应融合。颜色图层则采用线性叠加方式处理。对处理后的图层合成，得到化妆人脸图像。实验结果表明所提方法的妆容细节保留完整，可以克服利用Poisson方程融合所产生的拼接效应。%To satisfy the requirements of virtual face makeup, we proposed a high-fidelity transplant approach for digital makeup face image achieved by using illumination suppression and self-adaptive edge filtering.This method firstly improves the continuity of image’ s colour by adding the pre-processing step of image illumination suppression, then achieves the feature normalisation by using fast triangular coordinates transformation and this separates the face geometric structures from texture information.It decomposes the texture information into the layers of brightness and colour through colour space conversion.And the brightness layer will be divided into several regions with different fusing rates according to face feature region.When mixing the brightness layers, this method will combine the self-adaptive edge filter, thus realises the self-adaptive fusion for low-frequent information.The colour layers will be processed with linear superposition approach.By mixing the processed brightness layers and colour layers the makeup face image can be got.Experimental results showed that the proposed method preserved complete details of the makeup, and had no stitching effect found in Poisson clone.