基于多脸数据融合的人脸识别

论文编号:DZXX199  论文字数:12190,页数:33

 摘要

   人脸识别是计算机视觉和模式识别领域的一个活跃课题，有着十分广泛的应用前景。提出一种新的人脸识别方法，该算法采用K-L变换来提取最终特征值，采用了近邻法设计分类器，这样可以有效的避免了利用每一类的“代表点”设计分段线性分类器问题，减小了识别的错误率。在windows XP 平台和MATLAB 7.0 软件环境下，采用上述方法编制了人脸识别系统，仿真结果表明了该方法的有效性。

关键词：人脸识别，特征提取，K-L变换，线性分类器

ABSTRACT

 The face recognition is an active subject in fields of computer vision and pattern
 recognition, which has a wide range of  potential applications. In this paper, a method
face recognition is presented, this algorithm extracts the final features by utilizing the techniques of the simulative K-L transform. The near neighbor method based algorithm was utilized for classification, the problem and difficulty can be effectively avoided, which were faced when piecewise linear classifier was designed using each of the classes of "representative spots" piecewise, and the error rate of the recognition could be reduced. The face recognition system was developed and implemented on Microsoft Windows XP platform and Matlab 7.0 software environment using the above method. The effectiveness of the approach is experimentally demonstrated.
 
 Keywords: Face recognition, Feature extraction, K-L transform, Liner classifer
目录

摘要 I
ABSTRACT II
第1章 引言 1
1.1 背景 1
1.2 研究内容 2
第2章 人脸检测系统设计方法 3
2.1 肤色匹配 3
2.1.1 YCbCr颜色空间的选取 3
2.1.2 基于YCbCr彩色空间的肤色检测 5
2.2 人脸图像的标准化 6
2.3 基于K-L展开式的特征提取 6
2.3.1 K-L展开式 7
2.3.3 K-L坐标系的产生矩阵 9
2.3.4 奇异值分解（SVD） 10
2.4 特征向量的选择 11
2.5 算法实现流程 12
第3章 人脸识别系统设计方法 14
3.1 近邻法原理及其决策规则 14
3.2 最近邻法决策规则 14
3.3 K-近邻法决策规则 14
3.4 算法实现流程图 15
第4章 系统的实现 16
4.1 摄像头的描述 16
4.2 图像的读取 17
4.3 系统实现流程图 17
4.3.1系统训练过程 17
4.3.2 系统测试过程 18
第5章 结果与分析 20
5.1 举例说明运行过程与成功识别的完整过程 20
5.2 识别结果的准确度 22
5.3 结论 23
第6章 结束语 24
致谢 25
附录 部分源代码 27