基于特征选择和隐马尔可夫模型的人脸识别-仪器仪表工程专业论文.docxVIP

摘 要 人脸识别技术因为其在信息安全、公共安全、金融等方面的应用前景已经成 为模式识别和机器视觉领域的热门研究课题之一。隐马尔可夫模型（Hidden Markov Model，HMM）作为一种统计的识别方法，不仅能够考虑到各个器官的 数值特征，而且还兼顾了人脸的整体特征，因而可以取得较好的识别效果。一个 人脸识别系统的识别准确率在很大程度上取决于特征提取部分，因此特征提取是 人脸识别中一个至关重要的环节。 本文针对传统的隐马尔可夫模型在特征提取方面的不足，运用了离散余弦变 换(DCT)、奇异值分解(SVD)、离散小波变换（DWT）以及局部二值模式（LBP） 四种特征提取方法，分别从人脸辨别和人脸认证两个方面研究了基于不同特征提 取方法的隐马尔可夫模型的识别性能。 本文首先使用不同的特征提取方法优化 HMM 模型的观察矢量，有效地提高 系统的识别性能；其次，将以上几种算法应用到人脸辨别和人脸认证两个领域， 在不同人脸库中进行仿真，并且引入了多种性能评价指标来全方位的评价算法的 性能，本文工作为今后研究不同条件下人脸识别的特征选取提供了参考；最后， 针对人脸认证中“相似度偏移”的问题，本文提出了用自适应阈值来取代传统的 统一阈值人脸认证方法，实验证明，这一改进明显地改善了系统的识别性能。 关键词：隐马尔可夫模型 人脸辨别 人脸认证 特征提取 阈值 Abstract As one of the most challenging problems in the fields of pattern recognition and machine vision,Face recognition has a wide range of promising application, such as information security, public security and finance. As a statistical method, Hidden Markov Model (HMM) can take into account not only the characteristics of the various organs, but also the overall characteristics of the human face. Therefore, good results can be achieved. Extracting face features from facial images is the most important part in the face recognition system． Firstly ,some improvements are made on the shortcomings of feature extraction of HMM in this dissertation. More robust feature vectors can be extracted from the original images, using Discrete Cosine Transform(DCT), Singular Value Decomposition (SVD), Discrete Wavelet Transform(DWT) and local binary pattern (LBP). Then these vectors can be utilized as the observation vectors for HMM training and recognition.The experiment results show that the presented approach improve the recognition rat e effectively. Secondly, Face recognition includes two fields: face identification and face verification. This dissertation discusses the performance based on the different feature extraction methods of HMM in both fields.A comprehensive performance analysis of face recognition with several different fusion methods is given under different test conditions ． T