【作者】 曾华；

【导师】 娄联堂；

【作者基本信息】 武汉工程大学 ， 计算机应用技术， 2011， 硕士

【摘要】 人脸识别技术具有直观性、被动性及非侵犯性,且人脸的识别以携带方便、识别经济及准确的特点,是现在模式识别领域的专家和学者研究的热点之一。人脸的红外图像可以用于人脸识别的原因是红外人脸图像是由被测对象的人脸皮肤组织和结构的红外热辐射形成的,可以独立于外部光照。又由于热红外图像与人脸面部的血管分布有关,具有唯一性、抗干扰性、具有防伪装防欺诈性。因此,红外人脸图像识别受到越来越多的重视。论文首先研究了红外人脸图像的特点及特征向量的提取,利用红外人脸图像血管交叉点提取出特征向量,利用特征向量来匹配红外人脸图像。然后研究了基于混合高斯分布的红外人脸分割,该方法先用最小错误率的贝叶斯人脸分割算法将人脸的红外图像手动分割成人脸部分和背景部分,若这二部分分别服从不同的高斯分布,分别使用EM算法初始化混合高斯分布的参数,求出人脸部分和背景部分的概率分布。再用各向异性扩散滤波器对人脸图像平滑滤波提取特征向量,增强人脸血管边缘的对比度,再利用高帽分割可以得到红外人脸图像图像中的人脸血管。最后使用特征向量匹配算法来识别红外图像。通过红外人脸图像数据库算法验证。该文提出的算法识别率较高,具有可行性和实用性。更多还原

【Abstract】 Because of portability, economic and accurately of the recognition, furthermore, the face recognition technology has the features of intuition, passivity and non-infringe , which makes it one of the hot spots of the pattern recognition. The important feature of face recognition through infrared image of face is that infrared image of face being composed of infrared radiation of structure of the skin , it is independent of external light. Also the infrared image is in connection with vascularity of the face, which is unique, strong in anti-interference performance and anti-fraud. So more and more attentions are payed on this field.In this paper, firstly the character and eigenvector were extracted of the infrared image of face, the eigenvector was collected through the intersection point of the blood vessel of the infrared image of face, moreover the infrared image of face was matched by eigenvectors. Secondly the partition of the infrared image of face was studied based on mixed Gaussian distribution. The infrared image of face was divided to segment of face and background by hand based on Bayes algorithm of division of face with minimum error rate, supposing these two parts obey different Gaussian distribution. Parameters of Gaussian distribution were initialized through EM algorithm separately, so the probability distributions of segment of face and background were obtained. the eigenvector were extracted on smoothing filtering of image of face by anisotropic diffusing filter, contrast was strengthened on the fringe of blood vessel of face, then the blood vessel in the infrared image of face was achieved through top-hat segmentation. In the end, infrared image was recognized by the eigenvector matching algorithm. Validated by the database algorithm, the algorithm suggested in this paper has a high recognition rate and feasibility.更多还原