【作者】 王冬欣；

【导师】 冯浩；

【作者基本信息】 浙江工业大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 用计算机进行人脸识别是当今的一个研究热点和难点。本文基于人脸识别算法并以VC++为平台,开发了一个无人监考系统。该系统首先从3台USB摄像头中实时采集考试人员的三个不同方位的正面人脸图像,从图像中检测到人脸区域,再对图像中的人眼进行几何定位,通过特征提取对图像进行降维之后再进行考试人员人脸确认。经实验验证本系统算法简单,眼睛定位准确,识别率较高,适用于无人监考系统。本系统包括视频捕获模块,人脸检测模块,特征提取模块及人脸识别模块。论文首先介绍了几种常用的人脸识别方法,并对它们的优劣性进行了对比分析。再对各模块的实现进行了研究。视频捕获利用了微软开发工具VC++提供的VFW(Video For Windows)数字视频软件包,它能够灵活的实现从模拟视频采集数字视频信号,并将其以BMP格式存储到文件中进行后续处理。在人脸检测时,文中首先利用直方图对采集到的图像灰度进行预处理,使图像变得清晰。再利用肤色模型进行肤色分割,将肤色与背景区分开来。最后利用眼睛具有脸部最大复杂度的特点找到眼睛的几何位置。鼻子的位置可根据眼睛的位置轻易地找到。由于通常得到的图像空间维数都非常高,为了对图像进行降维,文中详细介绍并使用了傅里叶变换方法。通过傅里叶变换对检测到的眼睛和鼻子进行特征提取,图像的维数大大降低,计算复杂度的降低使之可以很好地应用于实时系统。最后通过计算特征向量在空间中的几何关系进行人脸识别。本课题最后完成了一个实时的无人监考系统,并通过实验对系统的性能进行了分析。实验证明该系统速度快、准确率高,对表情变化不敏感,具有较好的实用性。更多还原

【Abstract】 The face recognition by computer is a hotspot and difficulty nowadays. This paper develops a non-invigilation exam system based on face recognition and VC++ platform. First, this system collects the examinee’s front face image of different orientations from three USB cameras in real time. Detect the face area from the images and then locate the geometry position of eyes. At last, reduce the dimension of the image through face feature extraction to conform the face. The experimental result shows that this method is simple to realize, accurate to locate the eyes’ position and has high recognition ratio. With the advantages mentioned above, it is fit with the non-invigilation exam system.This system includes video capture module, the face detect module, the feature extraction module and the face recognition module. Firstly, this paper introduces some typical methods of face detection and face recognition, and analyzes their merits and demerits. Then the paper researches the realization of every module.The video capture is using the VFW (Video For Windows) data video software package which is provided by the Microsoft development tool VC++. It could aquire data video signal flexibly from simulation video signal, and save images to file as BMP format for the later processing. During the face detection, the paper first preprocesses the gray of the collected images by histogram to make them clear. Then distinguish the face area from background area by the complexion module. Last, locate the eyes’ geometry position based on the theory that the eyes have the maximum complexity on the face, and the nose’s position could be acquired easily from the position of the eyes.Usually, the space dimension of the image is high. In order to reduce the dimension, this paper particularly introduces and uses the Fourier transform. Through the transform to extract the feature of the detected eyes and nose, the dimension is reduced greatly. The reduction of the calculation complex make it fit for the real time system. At last, examinee face could be conformed through calculating the geometry relation of eigenvector in the space.This paper finally completes a real time non-invigilation exam system, and analyzes the system capability by the experimental result. It proves that this system has rapid running time and high recognition ratio. And it has a thick skin to the change face and good practicability.更多还原