基于特征的B样条拟合在三维人脸重建中的研究

发布时间：2018-07-13 16:46

【摘要】：随着社会信息化进程加快和计算机识别等技术的迅速发展,如何创建更具有真实感的三维人脸模型成为了一个非常具有挑战性的问题。三维人脸模型的重建在虚拟现实、视频监控、三维动画和人脸识别等领域都有着越来越多的应用。在身份识别方面,与其他生物识别相比较,人脸识别具有采集方便,可用性强等显著优势,受到大量关注。相对于二维人脸图像更容易受外部因素干扰,三维人脸模型不易受到外界光照条件、和化妆等因素的影响。因此,基于三维人脸重建模型的人脸识别技术能够更好的提高识别的准确度。在三维动画和游戏建模方面,创建更有真实感的模型成为了一个热点方向,现在,三维人脸重建己经逐步成为计算机视觉与计算机辅助设计等领域中备受关注的热点研究问题。不管是从理论研究还是从实际应用的方面来看,三维人脸重建都是值得深入研究的,它对于推动计算机视觉和计算机辅助设计研究的发展都有着重要的意义。但是目前的曲线曲面重建算法仍然存在很多问题:(1)当对拟合精度要求低的时候,很多局部上的极小值点就会很容易被忽略,会导致最后得到的曲线在该点出现形变;(2)当对拟合精度要求高的时候,计算量往往会变得十分庞大。针对上述问题,本文的研究内容主要包括以下几个方面:(1)点云数据预处理。三维人脸数据通常都是通过激光扫描仪获取的大量点云数据。这些数据不可避免地会受到噪声的污染,所以在进行曲面重建之前对点云数据进行简单的降噪处理。(2)提出一种新的曲线重建算法。通过对经典重建算法的分析,我们依据曲线的几何特征点和受误差约束的次要关键点在曲线重建中的重要程度,对这两类关键点进行分层,进而在保证精度的前提下提高了算法的工作效率。(3)曲面重建算法及在三维人脸重建中的应用。基于点、线、面的重要思想,对具有行列特征的数据点进行分割来获取曲面的轮廓线的数据点列,再采用特征分层的曲线重建算法,在斜高差,弓高差的约束条件下进行曲面拟合。并将此曲面重建算法应用于三维人脸曲面的重建。
[Abstract]:With the rapid development of social information technology and computer recognition technology, how to create a more realistic 3D face model has become a very challenging problem. 3D face model reconstruction has more and more applications in virtual reality, video surveillance, 3D animation and face recognition. In the aspect of identity recognition, compared with other biometrics, face recognition has many advantages, such as convenient collection, strong usability and so on. Compared with two-dimensional face images, 3D face models are more susceptible to external interference, and 3D face models are not easily affected by external illumination conditions, makeup and other factors. Therefore, face recognition based on 3D face reconstruction model can improve the accuracy of recognition. In the field of 3D animation and game modeling, creating more realistic models has become a hot topic. Now, 3D face reconstruction has gradually become a hot research issue in computer vision and computer-aided design and other fields. From both theoretical and practical aspects, 3D face reconstruction is worthy of further study. It is of great significance to promote the development of computer vision and computer-aided design (CAD) research. However, there are still many problems in the current curve and surface reconstruction algorithms: (1) when the precision of fitting is low, many local minimum points are easily ignored. The resulting curve will deform at this point; (2) when the precision of fitting is high, the computation will become very large. To solve the above problems, this paper mainly includes the following aspects: (1) Point cloud data preprocessing. 3D face data are usually a large number of point cloud data obtained by laser scanner. These data will inevitably be polluted by noise, so the point cloud data is simply de-noised before surface reconstruction. (2) A new curve reconstruction algorithm is proposed. Based on the analysis of the classical reconstruction algorithm, we stratify the two key points according to the importance of the geometric feature points of the curve and the minor key points constrained by errors in the reconstruction of the curve. Furthermore, the efficiency of the algorithm is improved on the premise of ensuring accuracy. (3) Surface reconstruction algorithm and its application in 3D face reconstruction. Based on the important idea of point, line and surface, the data points with column and column feature are segmented to obtain the data points of the contour line of the curved surface, and then the curve reconstruction algorithm of feature stratification is adopted, and the deviation of oblique height is obtained. The curved surface fitting is carried out under the constraint of bow height difference. The surface reconstruction algorithm is applied to 3D face surface reconstruction.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】