基于改进LDP特征的人脸识别方法

发布时间：2018-06-07 23:57

本文选题：人脸识别 + Adaboost　；参考：《南京理工大学》2017年硕士论文

【摘要】：人脸识别具有操作简单、表现稳定的优点,在近年来获得了日益广泛的应用,这也对人脸识别方法的成功率与运行速度提出了更高的要求。本文在理解了人脸识别基本原理与难点、研究了国内外相关文献的基础上,按照图像预处理、人脸检测、特征提取、完整实现、结果验证的脉络进行了改进人脸识别方法的研究。首先对人脸识别的发展历程进行了概述,明确了从局部特征方法着手的研究方向。其次分别介绍了直方图均衡化、空间域滤波以及图像锐化算法,完成了对图像预处理的研究。然后研究了 Adaboost算法,利用OpenCV平台提供的Haar特征分类器,验证了一种有效的人脸检测方法,并引入积分图与级联分类器模型优化了人脸检测速度。之后对包括基本LBP、改进LBP与LDP在内的局部特征提取方法进行研究,重点提出了一种改进的LDP特征提取方法,改进内容包括:将待识别人脸图像切割成相同大小的子区域;在得到中心像素的特征编码时将负数梯度返回值纳入考虑;对各个子区域按照结构对比信息的不同赋予不同权值。最后在ORL与Yale人脸数据库上验证并对比了 LBP、LDP与改进LDP方法的人脸识别效果,通过子区域直方图加权匹配保证了不同子区域在特征直方图匹配时发挥不同的作用。通过实验结果的对比证明了基于改进LDP特征的人脸识别方法与传统的LBP、LDP方法相比,具有更高的人脸识别成功率,证明了本文提出的改进方案是有效的。
[Abstract]:Face recognition has the advantages of simple operation and stable performance, which has been increasingly widely used in recent years, which also put forward higher requirements for the success rate and running speed of face recognition methods. Based on the understanding of the basic principles and difficulties of face recognition and the study of the related literature at home and abroad, this paper is based on image preprocessing, face detection, feature extraction, and complete realization. The results show that the improved face recognition method is studied. Firstly, the development of face recognition is summarized, and the research direction of local feature method is clarified. Secondly, histogram equalization, spatial filtering and image sharpening algorithms are introduced respectively. Then, the Adaboost algorithm is studied, and the Haar feature classifier provided by OpenCV platform is used to verify an effective face detection method, and the integration graph and cascade classifier model are introduced to optimize the face detection speed. Then the local feature extraction method including basic LBPand improved LBP and LDP is studied, and an improved LDP feature extraction method is put forward, which includes: cutting the face image to the same size sub-region; The negative gradient return value is taken into account when the feature coding of the center pixel is obtained, and different weights are assigned to each sub-region according to the different structure contrast information. Finally, the face recognition effects of LBP LDP and improved LDP are verified and compared on ORL and Yale face databases. The weighted matching of sub-region histogram ensures that different sub-regions play different roles in feature histogram matching. The comparison of experimental results shows that the face recognition method based on improved LDP features has a higher success rate than the traditional LBP- LDP method, and proves that the improved method proposed in this paper is effective.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】