【作者】 程雪红；

【导师】 刘志镜；

【作者基本信息】 西安电子科技大学 ， 计算机应用技术， 2006， 硕士

【摘要】 人脸识别技术是近年来图象处理、模式识别、人工智能等领域内最为活跃的研究课题之一。它具有广泛的应用领域和重要的理论研究价值。它主要包括人脸检测和人脸识别两方面内容。本文对彩色图像中正面人脸检测问题和对彩色、黑白和手绘等图片的识别问题进行了详细研究,并实现了一个人脸识别仿真系统。在人脸检测方面,提出了肤色检测与面部特征空间位置验证相结合的方法。首先针对人脸肤色在色彩空间中的分布及其特性展开研究,建立了有效的肤色模型,对图像进行肤色分割;接着,对分割结果依据人脸的矩特征去掉部分非法的肤色区域;最后,在所得区域内检测出面部特征,按照特征之间的空间位置关系对候选脸区的真假进行验证。经实验表明,该方法在保证较高的检测率的基础上,大大降低了误检率。在人脸识别方面,提出了基于小波系数的隐马尔可夫模型的识别方法。对原始图像进行3级小波变换,获得低频平滑、水平细节和垂直细节三个子图像的小波特征,然后在频域上对其进行核主分量分析,对最终获得的三组特征向量进行特征融合,并把它作为隐马尔可夫模型的观察向量进行训练,将优化的模型参数用于人脸识别。此外,本文针对手绘图像识别问题进行了研究,提出了一个将手绘图像转化为照片的方法,尝试将手绘人脸识别问题转化为普通照片识别问题,降低了手绘人脸识别的难度。最后,实现了人脸识别仿真系统,并进行了大量的实验和测试。结论表明该系统不仅适用于多表情的人脸识别,还对不同类型样本选取有很好的鲁棒性。更多还原

【Abstract】 Face recognition technology (FRT) is one of the most active research topics in areas of image processing, pattern recognition and artificial intelligence recent years. It includes two parts: face detection and face recognition. This dissertation focuses on research of frontal face detection and recognition of color, monochrome and sketch images. Furthermore, the author also establishes a face recognition simulation system.In order to detect frontal face, a method which combined skin color detection and validated spatial facial features scope is presented. The distribution and characteristic of skin color is firstly studied in color space to establish effective skin color model and to use it in image segmentation. Then, some unseemliness regions in segmentation are eliminated according to moment features. At last, we locate the facial features based on the acquired skin color, and to validate whether this face region contains face or not according to the relative location among features. The experiment results show that this method decrease false detection rates enormously.A new algorithm for face recognition based on wavelet transform and hidden Markov model (HMM) is proposed. The original image is decomposed into low frequency and high frequency sub-band images by applying three-level wavelet transformation. As a result, three groups of wavelet features which correspond to low frequency, horizontal details and vertical details are obtained. Kernel principal component analysis (KPCA) is then performed on each group in frequency. These three groups after KPCA are combined together by using a feature fusion method, and served as observation vectors of HMM. A set of images representing different instances of the same person are used to train each HMM. Then the optimal model parameters are used in face recognition. Furthermore, the author also works on face sketch recognition. The author presents a novel sketch-to-photo transformation method in order to transform sketch recognition to face photo recognition for face sketch recognition more easily.At last, a face recognition simulation system is established. The experiment results show that it is not only suitable to multi-expression recognition but also robust to different types of images.更多还原