人脸识别研究及在高校迎新系统中的应用

发布时间：2018-05-27 20:37

本文选题：人脸识别 + 光照模型　；参考：《贵州民族大学》2017年硕士论文

【摘要】：人脸识别技术作为身份识别的主要内容,是目前数字图像处理及模式识别研究的热点问题。其通过计算机分析采集到的人脸图像,提取出有效的识别信息,用于自动“辩认”身份,解决“他是谁?”的问题。但由于人脸图像采集过程中,受到复杂外部环境和自身属性的的影响,如光照、阴影、化妆、配戴首饰(头饰)等的影响,同一人脸获取的图像会有较大的差别,这些使得用机器来正确识别人脸变得艰难。本文从人脸图像的稳定特征分布出发,利用人脸类别分布的先验知识,提出一种自适应的人脸图像标准化方法和光照不变性的特征提取算法,通过对这些特征的分析来实现对人脸图像的识别。主要工作如下:1、使用改进高斯肤色模型进行人脸检测与分割,在朗伯光照模型、伽马校正、对数变换等图像处理基础上结合同态滤波,提出了一种自适应的人脸图像标准化方法。2、对传统的LBP、多级LBP、Gabor小波等光照不变性的特征提取算法进行了深入的研究,在二维Gabor小波算法基础上融合LBP思想,获得降低特征维数的特征抽取方法。3、设计统一标准的分类方法,分别使用最近邻分类和LibSVM方法进行分类识别。并对实验结果进行比较。4、在visual studio 2015平台上,将本文算法集成应用于高校迎新实验系统,对新入校的学生进行身份识别,获得较好的效果。
[Abstract]:As the main content of identity recognition, face recognition is a hot issue in the research of digital image processing and pattern recognition. Through the computer analysis of the collected face images, extract effective recognition information, for automatic "identification" identity, to solve "who is he?" The question of However, due to the influence of complex external environment and its own attributes, such as illumination, shadow, makeup, wearing jewelry (headdress) and so on, the image acquired by the same face will be quite different in the process of face image acquisition. This makes it difficult to correctly identify human faces with machines. Based on the stable feature distribution of face image and the prior knowledge of face category distribution, an adaptive method of face image standardization and a feature extraction algorithm based on illumination invariance are proposed in this paper. The recognition of face image is realized by analyzing these features. The main work is as follows: 1. Using improved Gao Si skin color model for face detection and segmentation, combining homomorphic filtering with image processing such as Lambert illumination model, gamma correction, logarithmic transformation, etc. In this paper, an adaptive face image standardization method .2is proposed. The traditional feature extraction algorithms such as LBP and multilevel LBP- Gabor wavelet are studied deeply, and the LBP idea is fused on the basis of two-dimensional Gabor wavelet algorithm. The feature extraction method. 3 is obtained to reduce the feature dimension. A unified classification method is designed. The nearest neighbor classification method and the LibSVM method are used for classification and recognition respectively. The experimental results are compared with .4. on the platform of visual studio 2015, the algorithm of this paper is integrated into the experimental system of welcoming new students in colleges and universities, and the identification of new students is carried out, and good results are obtained.
【学位授予单位】：贵州民族大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】