可变光照和遮挡条件下的人脸识别技术研究及其应用

【作者】 魏道敏；

【导师】 茅耀斌；

【作者基本信息】 南京理工大学 ， 控制理论与控制工程， 2012， 硕士

【摘要】 人脸识别研究是近年来模式识别领域的一大研究热点,具有广阔的应用前景。本文主要致力于基于静态图像的可变光照和遮挡条件下的人脸识别方法研究,重点研究了基于子空间、基于LBP纹理特征和基于稀疏表示的人脸识别方法。本文研究内容主要包括以下几个方面：(1)研究了基于子空间的人脸识别方法的原理和在人脸识别应用,该方法通过将图像从数据空间映射到特征空间,达到数据降维的目的同时提取出有利于识别的信息,具体分析了PCA、Fisherface、SLPP、2DPCA、2DLDA、KPCA和KFDA七种子空间方法的优缺点及内在联系。(2)针对光照变化问题,本文提出了分块LBP纹理特征的人脸识别方法,首先提取每个窗口内的LBP纹理特征直方图,然后将每块的直方图叠加,最终获得人脸描述特征。实验结果证明了该方法对光照变化稳健并且计算量低。(3)针对带遮挡或受噪声干扰的正面人脸图像,本文给出了一种自动人脸识别方法,即基于稀疏表示的人脸识别方法。该方法将识别问题当作多元线性回归模型中的一种分类问题来研究同时有关稀疏信号恢复的新理论对阐述这个问题起了关键作用。利用L1范数最小化获得的稀疏表示,本文得到了一种用于物体识别的统一算法。本文对公开的人脸数据库做了大量的实验,证明了该方法的有效性。最后,在实验室内自然环境下,本文采用基于LBP纹理特征的方法搭建了一个实时的人脸识别系统并且取得了良好的识别效果。更多还原

【Abstract】 As one of the most heated research spot in the field of pattern recognition, the human face recognition applies to tremendous aspects and has a promising future. This dissertation focuses on face recognition algorithms based on statci image under variation of illumination and occlusion, special attention has been paid to the subspace, LBP, as well as spare representation.The main research results are as following:(1) Considering that, by casting image from data space to feature space, subplace based face recognition methods can absorb information that is useful for recognition from reducted date dimension, this dissertation studies its application in face recognition. The merits and demerits of seven subspace based methods are presented and the relationship between the seven methods are disclosed.(2) To alleviate the impact of variation of illumination, we bring about the method based on LBP texture feature of sub windows. Firstly we extract the LBP texture feature histogram of every window, and then we pile up every histogram, at last we acquire the characters of human face. The result confirms that this method has a steady effectiveness to the variation of illumination and requires low computational complexity.(3) This dissertation finds the method of automatically recognizing human faces from frontal views with disguise or occlusion, this is face recognition based on spare representation. The method casts the recognition problem as one of classifying among multiple linear regression models and argue that new theory from sparse signal representation offers the key to addressing this problem. Based on a sparse representation computed by L1-minimization, we propose a general classification algorithm for (image-based) object recognition. We conduct extensive experiments on publicly available databases to verify the efficacy of the proposed algorithm.At the same time, under natural environment, we build a real-time face recognition system using LBP texture feature which achieves good results in the laboratory.更多还原