结合深度图像和强度图像的人脸浅浮雕生成算法  

The Bas-Relief Generation Method of Human Faces from 3D Depth Images and 2D Intensity Images

作　　者：陆巧[1,2] 王璐[1,2] 孟祥旭[1,2] 王文佳[3] 

机构地区：[1]山东大学计算机科学与技术学院,济南250101 [2]数字媒体技术教育部工程研究中心,济南250101 [3]山东师范大学生命科学学院,济南250101

出　　处：《计算机辅助设计与图形学学报》2015年第7期1172-1181,共10页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金面上项目(61472224;61472225);国家"八六三"高技术研究发展计划(2012AA01A306);国家自然科学基金重点项目(U1035004);泰山学者建设工程专项经费

摘　　要：针对单纯基于深度图像压缩方法生成的三维浮雕模型容易存在大量噪声,并且易丢失细节的问题,为了生成特征表达准确且具有艺术美感的人脸浅浮雕,提出一种基于三维深度图像和与之对应的二维强度图像的混合人脸浅浮雕的生成算法.首先对三维深度图像的高度场进行压缩生成浮雕基网格;然后提取二维强度图像的灰度信息、梯度信息和显著度信息,并基于显著度信息对基网格的高度场相应地叠加灰度和梯度信息,以保持脸部五官的细节;最后根据显著度信息使用Laplace算子对网格进行光顺处理,使脸部光滑.实验结果表明,与单纯使用基于深度图像压缩方法相比,文中算法生成的浅浮雕模型可以更好地保持五官的细节特征;与基于图像的浮雕生成方法相比,该算法避免了复杂交互操作,可以更好地保持人脸的整体轮廓和形状.Using three-dimensional distance scanner can obtain depth images as well as the corresponding inten- sity images. Most of the depth images are rough, rich of noise and even incomplete. Therefore, it is easy to gener- ate a 3D bas-relief which has a lot of noise and less of details using depth images-based compression method. In this paper a hybrid algorithm for bas-relief generation of human faces from 3D depth images and 2D intensity images is presented. Firstly, the height field can be compressed to generate the base mesh of bas-relief. Secondly, the intensity images can be used to extract the information of gray level, gradient and saliency, by using saliency information, the gray and gradient information can be added into the height field of base mesh of relief to pre- serve the facial details. Lastly, smooth the mesh of relief with the Laplace operator on the basis of saliency to make the face smooth. The experimental results show that compared to the depth images-based compression me- thod, the bas-relief generated by our method can keep the facial detailed features much better and compared to the image-based method, our method will avoid the complex interactive operation and keep the whole contour and and shape of human face better.

关 键 词：人脸浅浮雕 深度图像 强度图像 基于上下文的显著度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]