【作者】 王永军；

【导师】 赵丽红；

【作者基本信息】 东北大学 ， 电路与系统， 2014， 硕士

【摘要】 人脸识别技术是最近几十年生物特征识别领域的研究热点,它融合了计算机图形学、数字图像处理、计算机视觉、模式识别和人工神经网络等多个学科的方法和理论。人脸识别技术在公共安全和军事安全领域有着十分广阔的应用前景。本文在查阅相关文献和对人脸识别系统深入研究的基础上,利用北京瑞泰创新公司的ICETEK-DM6437-B-KIT作为硬件开发平台,设计实现了基于遗传算法的人脸识别系统。系统包括对视频图像的光照补偿、特征提取、特征选择、分类识别等功能,并能将识别结果显示。本文主要完成的工作如下：(1)图像光照补偿。在实时视频采集系统中,光照强度对系统性能影响很大。本文通过实验对比分析了三种光照补偿算法,最后确定本文采用的光照补偿方法。(2)人脸检测与定位。本文在YCbCr颜色空间采用肤色检测方法对人脸肤色进行聚类分析,完成了人脸区域的检测,并通过人脸形状特征参数来排除脖子等非人脸区域。(3)人脸特征提取与选择。本文采用8×8分块LBP(Local Binary Patterns)算子提取人脸纹理特征,共提取出640维特征向量。采用遗传算法进行特征选择,并根据基本遗传算法的理论知识,提出了一种改进的遗传算法,主要改变了适应度函数、选择算子、交叉概率和变异概率等参数。实验结果表明,本文提出的改进遗传算法在全局收敛性、迭代时间和识别率上都优于基本遗传算法。(4)人脸分类识别。本文采用最近邻分类器和支持向量机(Support Vector Machine,SVM)分类器相结合的方法进行人脸分类识别。在处理前端用最近邻分类器进行粗分类,得到与测试样本距离最小的两类,再对这两类用SVM进行细分类。(5)在ICETEK-DM6437-B-KIT硬件平台上实现了系统的功能。通过在CCS(Code Composer Studio)中采用C语言编程,实现了视频人脸识别功能。通过测试,本文所设计的人脸识别系统识别率达到90%以上,具有较好稳定性和很强的实用性,为进一步研究人脸识别系统打下了很好的基础。更多还原

【Abstract】 Face recognition technology has become one of the hot research topics in biometrics recognition, and it integrates computer graphics, digital image processing, computer vision, pattern recognition and artificial neural network. Face recognition technology has a broad prospect in public security and military security.Based on literature review and in-depth study of the face recognition system, this thesis designed a face recognition system based on genetic algorithm. The system used the Beijing Ruitai innovative company’ICETEK-DM6437-B as a hardware development platform. This system included video image light compensation, feature extraction, feature selection, classification and identification and it displayed the recognition results. The main achievements can be described as follows.(1) Image light compensation. In the real-time video acquisition system, light intensity has a great influence on the performance of the system. In this thesis, through comparative analysis of the experiment of three kinds of illumination compensation algorithm, this thesis drew the illumination compensation method.(2) Face detection and location. In the face detection stage, using the clustering of human skin color in YCbCr color space, this thesis chose skin color detection method to complete the candidate face region detection and used the feature parameters of face shape to exclude some non-face region.(3) Face feature extraction and selection. The thesis used the 8×8 block Local Binary Patterns(LBP) algorithm for feature extraction and extractd 640 dimensional feature vectors. This thesis used genetic algorithm for feature selection. According to the basic theoretical knowledge of genetic algorithm, this thesis presented an improved genetic algorithm. The main changes of the control parameters were the fitness function, selection operator, crossover probability and mutation probability. The experimental results showed that the improved genetic algorithm proposed in this thesis had better global convergence, the less iteration time and higher recognition rate than basic genetic algorithm.(4) Face recognition. This thesis used support vector machine classifier combined with nearest neighbor classifier to match the facial feature. In front, the nearest neighbor classifier was used for coarse classification to obtain two classes of the minimum distance with the test sample, and then it used the support vector machine fine classification.(5) On the ICETEK-DM6437-B-KIT hardware, this thesis achieved the function of the system. By using C language, it achieved the video face recognition in CCS.The experiment showd that recognition rate of the system designed in this thesis was beyond 90% and the system had a certain degree of stability and a strong practicality. It laid a good foundation for the further study of face recognition.更多还原