【作者】 周鹏；

【导师】 孙长银；

【作者基本信息】 东南大学 ， 控制工程（专业学位）， 2017， 硕士

【摘要】 随着模式识别和计算机视觉的发展,人脸识别的应用越来越重要。人脸识别技术因为它的稳定性和可靠性,成为识别技术中比较理性和可靠的方法。它已经成为安全以及其相关领域的较为高精尖的技术,并且已经快速地进入市场,深入到各行各业的应用中。人脸识别的流程一般分为:人脸检测,人脸预处理,人脸特征提取,人脸识别与匹配。其中人脸特征提取是人脸识别整个过程中最为重要的问题,提取有效的特征对识别效果有至关重要的影响。在某些特定的环境下,提取到的人脸特征的识别效果依然不能取得十分理想的结果,特别是因为图像采集过程中,光照,噪声,姿态和表情等因素产生变化造成的影响。通过相关适当的方法,把人脸图像进行变换,将原始的人脸图像映射到一个特征空间,降低这些外部因素的影响,是提高人脸识别效果的有效途径。本文主要的研究内容是针对人脸识别中特征表达的关键技术展开研究,提出识别性能更加良好的人脸特征。本论文的主要工作如下:1、基于局部双交叉三值模式特征描述子(LDCTP)的人脸识别方法:提取有效的特征是图像表示和识别中的基本问题,然而特征的提取很难在描述能力和鲁棒性之间实现合理的平衡。LDCTP是受到人脸图像的特殊纹理结构的启发提出的。在相同条件下,LDCTP的识别性能优于其他相关特征描述子,但是它花费时间稍微多一些。2、基于D-PSIFT的局部主方向直方图的人脸识别方法:结合SIFT和D-HLDO的各自优点,提出了一种新型的人脸特征提取方法,该方法在保证识别效果的前提下,大大减少了计算量,使得计算速度快了接近3倍。3、对特征融合和多核学习方法进行了简介。核方法在计算机视觉,特别是视觉对象识别中越来越受重视,核方法的主要思想是引入非线性决策函数,将原始的特征映射到一个更高维度的空间。通过多核学习将LDCTP与D-PSIFT进行特征融合,并验证了融合之后的特征的识别性能。更多还原

【Abstract】 With the development of pattern recognition and computer vision,the application of face recognition is becoming increasingly important.Face recognition technology,because of its uniqueness and stability,has become rational and reliable method.It has become a sophisticated technology in the field of security,and is rapidly entering the market.The face recognition system usually consists of face detection,preprocessing,feature extraction,face recognition and matching.Feature extraction is the most important issue.In some specific circumstances,the identification rate of the extracted facial features is still difficult to achieve the desired requirements,especially influenced by changes of pose,illumination,noise and expression.Through the appropriate method,mapping the original face image to a feature space is an effective way to improve identification rate which can reduce the impact of these factors.The main research focuses on the face feature extraction,which can overcome the negative influences of noise,expression and illumination in face recognition system.The highlights of the main research are as follows:1.A novel face feature extraction method named local dual-cross ternary pattern(LDCTP)is proposed:Extracting effective features is a fundamental issue in image representation and recognition.However,it is difficult to achieve a reasonable tradeoff between discriminative power and robustness.LDCTP is a feature representation inspired by the sole textural structure of human faces.Under the same conditions,LDCTP has better recognition performance than other related feature descriptors,but it takes a little more time.2.A novel face feature extraction method named face recognition based on D-PSIFT local principal orientation histogram is proposed:Combining the advantages of SIFT and D-HLDO,presents an efficient and simple method for feature extraction of face images,which greatly reduces the computational complexity and makes the speed of calculation increased by nearly 3 times.3.The basic theory of feature integration and multiple kernel learning is introduced.Kernel methods have become popular in computer vision,particularly for visual object recognition.The key idea of kernel methods is to introduce nonlinearity into the decision function by mapping the original features to a higher dimensional space.The integration of LDCTP and D-PSIFT by the method of multiple kernel learning is proved to be effective.更多还原