【作者】 马晓燕；

【导师】 杨国胜；

【作者基本信息】 河南大学 ， 应用数学， 2006， 硕士

【摘要】 通过某种算法提取人脸的面部特征,然后进行特征匹配以确定人脸的归属,这就是所谓的人脸识别。它包括人脸跟踪、人脸检测、面部特征点检测、人脸识别、表情分析等多种内容。本文以河南省自然科学基金攻关项目“双目视觉监控系统研制”和河南省高校杰出科研人才创新工程项目“基于PGF和模糊技术的分布式图像分割”为依托,对人脸识别中的若干问题,特别是人眼定位,进行研究。主要工作归纳如下:第一,针对复杂背景下灰度人脸图像中人眼定位的问题,本文提出了一种改进的基于阈值的人眼定位算法。首先对基于均衡处理后的图像估计出初始阈值;其次采用阈值递增法逐次二值化图像,并对二值图像中的黑块逐一标记。然后用人眼位置判定准则确定可能的人眼区域,用相似度准则从中标记出两眼黑块。实验结果说明了该算法的有效性。第二,利用Gabor小波良好的生物特性和二维主元分析的降维能力,本文提出了一种基于Gabor小波和二维主元分析的人脸识别方法。这一方法和基于Gabor小波和主元分析的人脸识别方法的仿真比较实验,证明了基于Gabor小波和二维主元分析的人脸识别方法具有较好的识别性能。第三,细致分析了基于Gabor小波和支持向量机的人脸识别算法,归纳出该算法在实际应用中所遇到的主要问题,即累积贡献率选择、多项式核函数阶数的选择、以及决策函数的确定,给出了累积贡献率和多项式核函数的阶数选择规则,提出了支持向量机和最大值相结合的分类决策方法。通过仿真实验,确定了累积贡献率和多项式核函数的阶数选择范围,证明了支持向量机和最大值相结合的分类决策方法的有效性。更多还原

【Abstract】 The so-called face recognition is the face classification by firstly using certain algorithm to extract face features and then matching face features, which includes face tracking, face detecting, face feature detecting, face recognition, emotion analysis, an so on. Taking the key scientific and technological project of Henan province“The development of the binocular vision surveillance system”and Henan innovation project for university prominent research talents“The distributed image segmentation based on the PGF and fuzzy technology”as the research background, this thesis devotes main efforts to the research of some problems of the face recognition (especially the problem of the eyes location), mainly including:(1) Considering the eyes location in gray image with complex background, an improved method of human eyes location is presented based on the threshold. First, the initial threshold is obtained from the balanced-processing image. Second, the binarization of the balanced-processing image is done by increasing the threshold gradually, and the black blocks appearing in the binary image are marked one by one. Third, the possible areas of eyes are determined according to the rule of eye’s position. At last, the two marked black blocks representing the eye’s areas are fixed by making use of similarity. Simulation results illustrate the effectiveness of the algorithm.(2) By combining the excellence biology property of Gabor wavelet and the decreasing dimension ability of 2DPCA, a new method of face recognition is put forward based on the Gabor wavelet and 2DPCA. The comparison simulations between the method based on the Gabor wavelet and 2DPCA and the one based on the Gabor wavelet and PCA show that the former has the good recognition performance for the face image.(3) The method of face recognition based on the Gabor wavelet and support vector machine has been analyzed in details, which results in that some problems, such as the selection of the cumulative contribution rate in the principle component analysis, the choice of the order of polynomial kernel function, and the determination of the decision更多还原