【作者】 张淑艳；

【导师】 周春光；

【作者基本信息】 吉林大学 ， 软件工程， 2005， 硕士

【摘要】 人脸识别在基于生物特征识别技术的身份认证中是最主要的方法之一。基于人脸识别的自动身份认证具有重要的理论意义和应用价值,早在六七十年代就引起了研究者的强烈兴趣,对人脸自动识别方法的研究已成为当前模式识别和人工智能领域的一个研究热点。但是阻碍人脸识别技术应用到实际中的瓶颈之一——光照问题,一直没能得到很好的解决。一般的人脸检测和识别算法是假设待处理图像是在均匀的光照条件下获得的,而实际上光照往往是不均匀的。偏光、侧光导致的高光和过亮、过暗以及阴影等都会使人脸检测和识别率大幅度下降。本文中,将对人脸识别领域的光照影响进行分析和研究,从高光检测和校正以及基于纹理的光照恢复两个方面,进行探讨。针对彩色图像,本文进行了高光检测和恢复方法的讨论与分析,使用K-L 模型和Agent 模型给出人脸的粗定位算法,使用人体皮肤的反射光谱和双色反射模型进行了高光区域的检测以及辐射校正方法。对于灰度图像的光照恢复方法,本文从人脸光照的阴影区域估计开始,介绍参考人脸模型即人脸图像预处理方法,并使用粒子群优化算法进行纹理恢复参数的优化处理,最后进行了基于人脸纹理的人脸光照恢复操作。更多还原

【Abstract】 With the rapid development of computer and the need of the security, biological authentication about person identify has applied broadly in lifes. Different from other biological authentication methods, face detection and recognition is much more researched as it is simple, convenient, untouched and infringed. But the research about face image illumination is little. A general method of face detection and recognition is got in suppose that the image was shot in equally illumination. In fact, the illumination is almost unequally. The light of image is more highly, low, shade caused by sidelight will drop the corrected rate of face recognition. FERET test and FRVT test all indicate the illumination change is still one of the bottlenecks of practical face recognition systems of people. This paper discusses and analyses high light detection and restoration in color image. We locate the rough face using K-L and Agent model. Then detects the high light area of face image and corrects it with human skin reflection spectrum and double colored reflection model. As to the illumination restoration of a gray image, the paper estimates the shade area in the image first. Then introduces the reference face model and image preprocessed method. Optimize the texture parameters by PSO. Finally restore the face illumination by the face texture. The content of this paper is arranged as: Chapter one is a introduction the technology of the paper, including of the research content、advantages and difficulties of face detection and recognition, the opened problem in face recognition is that the influence of illumination and the methods to resolve the light problem. Chapter two is about the high light area detection is a colored image. In this part, we give some prepare knowledge such as some color model used in detection based in color. Then we introduce the method of rough locating of face based K-l and Agent model. The human skin reflection model and double colored model is used to detect the high light area and correct it. Chapter three researchs the image’s illumination restore of uncertain point light source. First provides the preparation knowledge including the light and shade model of picture, image strengthen, POS methods use for optimizing the parameters. Then it introduces the reference face model, face alignment and the estimation of shade in the face. At last it provides the method that optimizing the parameters of the face texture by PSO and restores the image illumination. Chapter four summarizes the whole paper and gives the research direction of the face illumination in the future. The illumination process of face image is explored and discussed primely. But there is a distance to use face illumination process in life as it is an open problem and is difficult to resolve in face recognition. We will make a deep and detailed analysis and research in the future.更多还原