【作者】 李杰；

【导师】 阮秋琦；

【作者基本信息】 北京交通大学 ， 模式识别与智能系统， 2008， 硕士

【摘要】 人脸识别是计算机模式识别领域非常活跃的研究课题。在生物特征识别中,人脸识别占有极其重要的地位,它在访问控制、司法、电子商务和视频监控领域有十分广泛的应用。人脸图像预处理和特征提取是人脸识别系统的重要组成部分。本论文主要研究了小波变换在图像预处理中的应用和Fisher脸方法在人脸特征提取中的应用,并对其进行了创新和改进,并且将本文的方法应用在实时的人脸识别系统当中。主要内容如下:(1)小波变换在人脸识别中得到了越来越广泛的应用,然而大多数应用小波变换对人脸图像进行预处理时,都仅仅采用了对图像进行小波分解后的近似图像,其实在进行小波分解过程中得到的图像的水平细节和垂直细节表现了眼睛,鼻子,嘴等细节,对人脸的表示也起到了一定作用。因此,本文提出了基于加权小波法来对图像进行预处理的方法,将小波分解得到的近似图像,水平细节,垂直细节按一定权重进行加权,实验结果表明,该方法比单独使用近似图像有更好的识别率。(2)介绍了Fisher脸方法的基本原理,针对Fisher脸法受特征维数限制的问题,引进了一种利用迭代法来增加特征维数的RFLD方法。但RFLD方法在大样本集的情况下,计算量比较大,本文针对RFLD方法的这点不足对其进行了改进。通过实验,我们发现,经过改进的方法不仅减少了计算量,并且保证了识别率,甚至提高了识别率。(3)将本文的算法应用在实时的人脸识别系统当中,通过USB摄像头进行实时人脸识别,验证了算法的可行性,获得了理想的效果。本文对人脸识别的中的图像预处理和特征提取方法进行了详细的阐述,对传统的方法进行了改进,通过实验,分析了本文算法的优点与不足,提出了以后研究工作的方向与目标。更多还原

【Abstract】 The technology of face recognition is an active subject in the area of pattern recognition. As a biometrics technology, FRT has numerous applications such as access control, law enforcement, e-commerce, video surveillance and so on.Face image preprocessing and face feature extraction is an important component of the face recognition system. In this paper, we focus on the application of wavelet transform on face image preprocessing and the application of Fisher face method on face feature extraction. Based on the classic method, we make some innovation and improvement and use the method in a real-time face recognition system. The principal works are listed as follows:(1) Wavelet transform is used more and more widely in face recognition, but only approximate image was used for feature extraction in most reference. In fact, horizontal details and vertical details also play a certain role in description of human face, especially in eyes, nose, etc. We propose a novel face image preprocessing method based on weight-wavelet which adds approximate image, horizontal details and vertical details with certain weight coefficient. The experiment result shows that this method makes better recognition rate.(2) Expound the basic principal of Fisher face method, and introduce a new method named RFLD which can which can solve the limited of vector number. But it needs large amount of computation. In this paper, we improve the RFLD method. The experiment result shows that the improved method not only reduces the amount of computation, but also ensures the recognition rate, or even increases the recognition rate.(3) Use proposed method in a real-time face recognition system, prove the feasibility of our method and get satisfactory results.(4) In this paper, the image preprocessing and feature extraction methods are described in detail. We improve the traditional methods and analyze the advantage and disadvantage of our methods, and propose the direction and objective in the future.更多还原