【作者】 蒋桂莲；

【导师】 徐蔚鸿；

【作者基本信息】 长沙理工大学 ， 计算机应用技术， 2010， 硕士

【摘要】 人脸识别技术涉及到图像处理、模式识别和人工智能等多门学科,已成为计算机视觉和模式识别领域中一个富有挑战性的课题,在国家安全部门和银行密码系统等领域具有广泛的应用背景。支持向量机(SVM)算法由于过学习问题而导致其泛化性能降低,结合粗集对不精确数据的处理能力,提出了一种基于粗集边界和V-支持向量机(RSM-V-SVM)混合分类算法。该算法先在训练前采用粗集理论边界区域的不确定性预选出边界集,替代原始样本作为训练集,减少训练集的数目;然后在V-SVM算法的基础上引入了粗集理论上下近似集概念改进V-SVM算法,使其训练边界集。实验结果表明,该算法在分类正确率不受影响的情况下,大大缩短样本的训练时间,从而提高了改进的V-SVM的泛化性能和分类速度。为了降低冗余和进一步简化输入空间的维数达到减少算法求解计算量及处理时间,同时引入了粗糙集的属性简约方法对数据进行约简。在人脸识别过程中,人脸图像的特征向量采用核主元分析(KPCA)和属性约简方法联合进行提取,从而该特征向量作为RSM-V-SVM输入。用RSM-V-SVM算法对人脸图像样本进行训练,生成RSM-V-SVM分类器对人脸进行分类识别。针对ORL人脸数据库进行了实验,表明联合特征提取和RSM-V-SVM的人脸识别方法在识别率不变的情况下,具有很强的泛化性能。更多还原

【Abstract】 Face recognition technology involves many fields, including image processing, pattern recognition ,artificial intelligence and so on., which has become challenging subjects in computer vision and pattern recognition fields. So it has wide applications prospect in many fields, such as in department of national security and bank password system.The generalization ability of SVM algorithm is decreased due to over-learning, in the paper, a hybrid classification algorithm based on margin of rough sets and V-support vector machine(RSM-V-SVM) was proposed which combining imprecise data ability of rough set. Firstly, the algorithm get the boundary set using uncertainty properties of margin region of rough set theory before training, which substitute the original inputs as a training subset, and the size of the training set was shorten. Then, the concept of upper and lower approximation set of rough set was introduced for improving the V-SVM, which was based on the V-SVM algorithm. Experimental results show that the algorithm can’t influence recognition rate and shorten training time improved V-SVM while keeping the speed of classification and the performance of generalization. Algorithm attribute reduction (AR) of rough set theory was introduced to reduce data’s features, so the computation and the time complexity is decreased.In face recognition process, Extract the features of face images with the kernel principal component analysis (KPCA) and AR, then get the eigenvectors of face images as the input of RSM-V-SVM algorithm, produce RSM-V-SVM classifier last. Face images can be recognized with RSM-V-SVM classifier. The experiments on ORL face database show that face recognition process of combined feature extraction and RSM-V-SVM has strong generalization performance while keeping recognition rate.更多还原