基于人脸信息识别的信息安全技术研究-机械设计及理论专业论文.docx

摘要， 摘要 ， 认脸识别是模式识别领域中一个富有挑战性的课题，有着重要的理论 研究价值和应用价值。J—厂 本文首先介绍了国内外人脸识别的研究现状、方法和发展方向，然后 分别讨论了人脸图像的预处理、人脸特征的抽取和分类器设计等内容。 在图像预处理过程中，首先利用数码相机获取人脸数字图像，然后将 24位图像转换成256级灰度图像，并利用中值滤波的方法对图像进行去噪 处理，接着对图像进行二值化处理，最后采用积分投影的方法将人脸分割 出来并进行归一化，最终得到标准人脸图像。 在人脸特征提取过程中，对经过预处理的标准人脸图像，以类间散布 矩阵为产生矩阵，通过K—L变换降维并结合奇异值分解来提取人脸代数 特征。 在分类器设计过程中，考虑到单一分类器的识别率不是很高，本文将 最小距离分类器与模糊神经网络分类器结合起来构成一个组合分类器，以 期提高人脸识别率。 最后，本文给出了人脸识别实验结果，并对实验结果进行了分析／实 验结果表明，本文所提取的人脸特征是有效的，分类器的设计也是合趣而 高效的。P一7 关键词：人脸识别，代数特征，BP算法，模糊神经网络，最小距离分类器 ABSTRACTFace ABSTRACT Face recognition is a challenge subject in the field ofpattern recognition and it has important theory and appliance value． In this paper,we first introduce the current research，methods and trend on face recognition Then we respectively discuss face image preprocessing，face feature extraction and the design ofclassifiers． In the course of image preprocessing，digital face images are captured with a numeral csanera．and these images 8Je transformed t0 8 bit from 24 bit．Then the noise in face images are removed by using the method of median filter and the face images are binarized．Finally the face in each face image are segmented with the method of integration projection，then standardized．Standard face images are formed throuIgh the above-mentioned processing． During the feature extraction，for those standard face images，regarding the between-class scatter matrix as generating matrix，we extract the algebraic features of face images through K—L transform and singular value decomposition． In the course of classifiers design，considering that the single classifier has not high recognition rate。we construct a combining classifier with a minimum distance classifier and a fuzzy NN classifier to improve the recognition rate． Finally,the experiment results offace recognition are presented and analyzed． The results indicate that the extracted face features are valid and the design of classifier iS sound and efficient． Key words：face recognition，algebra feature，BP algorithm， fuzzy ne