基于高鉴别力SIFT和LGQP的人脸识别研究

发布时间：2018-04-26 16:48

本文选题：图像自适应分块 + SIFT　；参考：《上海师范大学》2017年硕士论文

【摘要】：随着科技的进步和多媒体技术的迅速发展,人脸识别技术在各个场景中的应用越来越多,特别是在安全管理领域的应用,更加凸显了人脸识别技术的重要性。人脸识别技术的关键是从人脸图像中提出有效的特征信息,这些有效的特征信息是实现人脸匹配识别的重要因素之一。因此,如何在信息量庞大的人脸图像中提取出最有效的特征信息,成为人脸识别技术中研究的热点。本文主要研究基于高鉴别力SIFT(Scale-invariant feature transform)和LGQP(Local Gabor Quaternary Pattern)的人脸识别,对传统的特征提取方法做出了相应的改进。本文研究的问题主要由以下三个方面构成:1.提出一种新的图像自适应分块策略,进一步提高识别算法的效率。该方法根据人脸图像中SIFT特征点的分布情况,对人脸图像分块,使人脸五官分别位于同一个子块内,提高子块的可识别力,降低头部姿势变化和面部表情变化对识别结果的影响。本文选取了ORL、YALE和JAFFE人脸库作为实验对象,实验结果证明了该方法的有效性。2.在传统LBP(Local Binary Pattern)算子的基础上,利用局部区域的均值和标准差对周围像素进行灰度变化运算,提出一种四值模式的LQP(Local Quaternary Pattern)算子,并将该算子与Gabor滤波结合,用于在人脸图像中提取LGQP特征。实验结果表明LGQP特征具有比LGBP特征和LGTP特征更加稳定和可靠的特性。3.借鉴Fisher线性判别分析的思想,本文提出一种利用类间与类内相关系数计算SIFT关键点鉴别力的算法。使用该算法计算出SIFT关键点的鉴别力之后,再计算出该点在不同方向、不同尺度的LGQP特征,并将LGQP特征与SIFT特征结合在一起,完成最终的人脸匹配识别。
[Abstract]:With the rapid development of science and technology and the rapid development of multimedia technology, the application of face recognition technology in every scene is more and more, especially in the field of security management, which highlights the importance of face recognition technology. The key of face recognition technology is to put forward effective feature information from face image, and these effective feature letters are the key to face recognition technology. Interest is one of the important factors to realize face matching recognition. Therefore, how to extract the most effective feature information in a large face image has become a hot spot in the research of face recognition technology. This paper mainly studies people based on high discriminability SIFT (Scale-invariant feature transform) and LGQP (Local Gabor Quaternary Pattern). Face recognition has made a corresponding improvement to traditional feature extraction methods. The main problems studied in this paper are mainly composed of the following three aspects: 1. a new image adaptive partitioning strategy is proposed to further improve the efficiency of the recognition algorithm. The method is based on the distribution of SIFT feature points in the face image, block the face image and make face five. In this paper, ORL, YALE and JAFFE face database are selected as experimental objects. The experimental results show that the effectiveness of the method.2. is based on the traditional LBP (Local Binary Pattern) operator and uses the Bureau. The mean and standard deviation of the region are calculated by the gray change of the surrounding pixels, and a LQP (Local Quaternary Pattern) operator of the four value mode is proposed, and the operator is combined with the Gabor filter to extract the LGQP features in the face image. The experimental results show that the LGQP feature has a more stable and reliable specificity than the LGBP and LGTP features. .3. draws on the idea of Fisher linear discriminant analysis. This paper presents an algorithm for calculating the discriminability of SIFT key points using the correlation coefficient between classes and classes. After using this algorithm to calculate the discriminability of the key points of the SIFT, the LGQP characteristics of the point in different directions and different scales are calculated, and the LGQP features are combined with the SIFT features. Final face matching recognition.

【学位授予单位】：上海师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】