基于wld特征的综合人脸识别方法分析及其软件设计-analysis of synthetic face recognition method based on wld features and its software design.docxVIP

摘要摘要自动人脸识别技术是模式识别、人工智能领域的研究热点。经过近半个世纪 的发展，已经提出了很多行之有效的人脸识别方法。然而，识别效果仍然受到光 照、姿态、伪装等外部因素的严重影响。由于局部特征具有提取方便，细节表征 能力强的特点，已经得到了了广泛的应用。韦伯描述子(WLD 特征)是一种强大的 局部纹理信息描述子，它不仅识别效率高，而且对光照、姿态变化有着良好的鲁 棒性。本文主要研究基于 WLD 特征的人脸身份、表情和性别识别方法。首先介绍 了 WLD 特征结合 Gabor 滤波器的人脸身份识别，并在 ORL 数据库上进行测试。 然后探讨了 WLD 特征的人脸表情表征能力，利用 WLD 特征结合 HMM 分类器 进行人脸表情识别，并在 JAFFE 人脸表情库上做测试。对 WLD 特征进行改进， 依据人脸不同区域对于识别效果的不同贡献，对不同区域设置了各自的 WLD 特 征权重，并结合 SVM 分类器在 FERET 数据库上进行人脸性别分类识别。实验表 明，WLD 特征在身份、表情和性别识别三个领域都取得了良好的识别效果。基 于前面所作的工作，本文还设计并开发了一个人脸综合识别软件，该软件集成了 自动人脸检测技术、LBP 与 WLD 特征提取和几种主要模式识别分类器，可以在 现有主流 Windows 操作系统下运行，能够进行人脸的身份、表情和性别识别，具 有较高的识别率。关键词：WLD 特征人脸身份识别表情识别性别识别ABSTRACTABSTRACTAutomatic face recognition is a quite active research field in modern pattern recognition and artificial intelligence, and there are numerous face recognition methods presented during past decades. Methods based on local texture features have been broadly applied in face recognition because of their convenience and detail characterization in feature extraction. Weber Local Descriptor (WLD) has been proved to be a powerful local texture descriptor, which is robust in complicated illumination and posture conditions and can get high recognition rate.In this paper, the WLD is improved and adopted in face recognition, face expression recognition and gender identification. We firstly combine the WLD features with Gabor features for face recognition, and take experiments on ORL database. Then the WLD features with HMM as classifier are used to face expression recognition on JAFFE database. However, the WLD often pays more attention to the regions of high variances in images. In consideration of the discrepancy of different regions in an image, We bring weights to WLD features of different regions, and apply it in gender identification with SVM classifier on FERET database. Finally, a set of integrated software is designed and developed for face identification, expression and gender recognition. Experiments manifest that WLD and its improvements are q