An Effective Method for Child Picture Synthesis
-
摘要: 提出了一种儿童照合成的方法. 分别给定男性、女性和原始儿童的图片, 该方法从人脸轮廓和肤色两个层面上融合生成目标儿童图片. 针对直接线性合成的人脸轮廓存在五官不成比例、扭曲等问题, 利用基于统计的方法得到标准儿童模板, 并用该模板对人脸轮廓进行约束, 进而求解得到最终的儿童轮廓. 此外, 本文重新构造了基于高斯径向基的变形函数, 消除因仿射变换而导致的图像拉伸和扭曲. 最后, 本文采用交叉融合的方法合成儿童图像, 并提出了用滤波器去除额头区域可能存在的虚影. 实验证明, 利用本文方法融合得到的儿童不仅具有父母亲的特征, 而且具有较高的视觉美感.Abstract: This paper proposes a method for child picture synthesis. Given a man's picture, a woman's picture, and a child's picture, our method generates the child picture by blending these facial shapes and skin colors. To solve the problem caused by direct linear combination, i.e., the facial shape suffers from disproportion and distortion. We get the final shape of the child with the help of a standard and statistical template of the child. In addition, we reconstruct the Gaussian radical base function to remove the image's stretch and distortion caused by affine deformation. At last, we adopt the cross-dissolves method for the child picture synthesis and propose using the filter to remove the forehead's artifacts. The results indicate that the child face generated with our method not only owns the father's and mother's features, but the picture also shows visual aesthetic feeling.
点击查看大图
计量
- 文章访问数: 1196
- HTML全文浏览量: 39
- PDF下载量: 619
- 被引次数: 0