【作者】 魏昇；

【导师】 雷蕴奇；

【作者基本信息】 厦门大学 ， 计算机应用技术， 2008， 硕士

【摘要】 人脸识别是基于生物特征的认证技术中具有挑战性的领域之一,也是本世纪有良好发展潜力的技术之一。作为自然而友好的身份识别方式,人脸识别已经成为模式识别和图像处理中的重要研究热点。使用二维图像人脸识别方法,由于受到光照、姿势、表情变化的影响,其识别的准确度受到很大限制。迄今为止,建立一个鲁棒的人脸识别系统仍然是一个很困难的问题。由于3D数据本身具有显式的几何形状信息,因此3D人脸识别更具克服姿态和表情困难的潜力。本文主要针对三维人脸识别,在以下几个方面展开了研究工作:1、在初始三维数点云数据量足够大的情况下,尝试使用B样条曲面拟合生成的网格控制顶点模拟三维点云数据。这个方法提高了点云数据的规格化程度,并大大减少了数据量,提高了算法效率。2、确定了三维人脸坐标系,并结合深度信息特点提取轮廓线,进行了曲率计算和分析,进而提取鼻子距离特征和人脸中分轮廓线分段曲率特征用于识别。降维处理简化了算法复杂度。3、分析特征向量的特性,利用欧式距离法和互相关函数进行样本间相似性度量,完成了人脸识别算法。4、在理论研究的同时,我们采用ViusalC++6.0以及SQL数据库后台设计实现了实验性的三维人脸识别系统平台。该系统能提取人脸轮廓线,并进行曲率计算和分析,从中提取人脸特征向量组,通过欧式距离法和互相关函数相似度比较实现三维人脸识别。试验结果验证了算法的可行性。更多还原

【Abstract】 Face recognition is one of the most active and challenging technologies. As a natural and friendly way, automatic face recognition has become an important part of the researches of image processing and pattern recognition.Because of the influence of illumination, pose variation and expression, the improvement of recognition accuracy of 2D face recognition is greatly impeded and it is difficult to build a robust face recognition system. Due to its richer information contained for facial surface, the 3D face data has more promising potential to conquer the change of pose than 2D images.This thesis addresses to study the 3D face recognition algorithms. The main contributions of the work are as follows:1、If the data of initial 3-D point-cloud is enough , the grid control points can be used to simulate the point-cloud data which is created by B-spline surface fitting. We standardize the point-cloud data to reduce the quantity of point-cloud data to raise efficiency of our algorithm.2、Contour lines are extracted according to depth information feature. Then, features of nose and the profile subsection curvature are obtained by analyzing curvatures of contour lines.3、The characteristic of eigenvector is analyzed, and the similarity between face samples is measured using Euler distance and cross correlation function.4、The 3D face recognition system is developed using visual C++6.0 and SQL Server 2000. The contour lines of the face are found and their curvatures are computed first. Then, by analyzing curvatures the eigenvectors of face are extracted. Finally, the similarities between face samples are measured to recognize the face. Experiments have been conducted to show the feasibility of our algorithm.更多还原