基于级联回归和LBP的人脸识别算法研究
发布时间：2018-10-19 20:51
【摘要】：人脸识别技术是计算机视觉和模式识别领域的重要研究内容之一。由于人脸识别问题的复杂性,且受到诸多外在因素的影响,要在识别率与稳定性上达到较高的标准比较困难。本文提出的人脸识别方法首先用级联回归算法对人脸关键特征点进行精确定位,用定位好的关键特征点对人脸进行校正,求取关键特征点局部邻域内的LBP直方图统计特征进行人脸建模,最后采用SVM分类器对人脸模型进行分类与识别。本文的人脸识别方法有效提高了人脸识别的准确率和识别效率,在GT人脸数据库上进行的实验中看出,本文人脸识别方法识别率最高可以达到99%。本文主要研究工作包括:(1)深入研究基于级联回归算法的人脸定位。级联回归算法采用迭代回归求解的方式优化人脸定位的准确性,本文在LBF特征的训练过程中,采用三种阈值计算方法改进了RF算法中弱回归树节点的训练。在LFPW数据集上的训练和测试证实改进算法能有效提高LBF特征的稳定性,提升模型整体的预测精度。(2)提出局部LBP人脸建模算法。全脸LBP人脸建模割裂了特征之间的关系且引入较多噪声。本文提出的局部LBP人脸建模算法,在降低噪声的同时保留了人脸的关键信息,为人脸识别器的训练提供了更好的特征数据。(3)构建基于SVM的人脸识别器。相比于直接的欧氏距离计算,SVM算法可以从人脸建模数据中找到关键区分样本,不仅识别效果更好,而且提高了时间上的响应效率。
[Abstract]:Face recognition technology is one of the important research contents in the field of computer vision and pattern recognition. Due to the complexity of face recognition and the influence of many external factors, it is difficult to achieve high recognition rate and stability. The method of face recognition proposed in this paper firstly uses cascaded regression algorithm to accurately locate the key feature points of the face, and corrects the face with the key feature points of good location. The LBP histogram statistical features in the local neighborhood of the key feature points are used to model the face. Finally, the SVM classifier is used to classify and recognize the face model. The method of face recognition in this paper can effectively improve the accuracy and efficiency of face recognition. The experiments on GT face database show that the recognition rate of this method can reach 99%. The main work of this paper is as follows: (1) the face localization based on cascade regression algorithm is studied in depth. The concatenated regression algorithm optimizes the accuracy of face location by iterative regression solution. In this paper, three threshold calculation methods are used to improve the training of weak regression tree nodes in RF algorithm during the training process of LBF features. Training and testing on LFPW datasets show that the improved algorithm can effectively improve the stability of LBF features and improve the overall prediction accuracy of the model. (2) A local LBP face modeling algorithm is proposed. Full-face LBP face modeling splits the relationship between features and introduces more noise. The local LBP face modeling algorithm proposed in this paper not only reduces the noise but also preserves the key information of the face and provides better feature data for the training of the face recognizer. (3) A face recognizer based on SVM is constructed. Compared with the direct Euclidean distance calculation, the SVM algorithm can find the key discriminant samples from the face modeling data, which not only has better recognition effect, but also improves the response efficiency in time.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
[Abstract]:Face recognition technology is one of the important research contents in the field of computer vision and pattern recognition. Due to the complexity of face recognition and the influence of many external factors, it is difficult to achieve high recognition rate and stability. The method of face recognition proposed in this paper firstly uses cascaded regression algorithm to accurately locate the key feature points of the face, and corrects the face with the key feature points of good location. The LBP histogram statistical features in the local neighborhood of the key feature points are used to model the face. Finally, the SVM classifier is used to classify and recognize the face model. The method of face recognition in this paper can effectively improve the accuracy and efficiency of face recognition. The experiments on GT face database show that the recognition rate of this method can reach 99%. The main work of this paper is as follows: (1) the face localization based on cascade regression algorithm is studied in depth. The concatenated regression algorithm optimizes the accuracy of face location by iterative regression solution. In this paper, three threshold calculation methods are used to improve the training of weak regression tree nodes in RF algorithm during the training process of LBF features. Training and testing on LFPW datasets show that the improved algorithm can effectively improve the stability of LBF features and improve the overall prediction accuracy of the model. (2) A local LBP face modeling algorithm is proposed. Full-face LBP face modeling splits the relationship between features and introduces more noise. The local LBP face modeling algorithm proposed in this paper not only reduces the noise but also preserves the key information of the face and provides better feature data for the training of the face recognizer. (3) A face recognizer based on SVM is constructed. Compared with the direct Euclidean distance calculation, the SVM algorithm can find the key discriminant samples from the face modeling data, which not only has better recognition effect, but also improves the response efficiency in time.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 王李冬;;一种新的人脸识别算法[J];计算机技术与发展;2009年05期
2 桂荣;沈荣鑫;詹泳;;基于模糊神经推理的人脸识别算法研究[J];华南师范大学学报(自然科学版);2012年04期
3 苏士美;王燕;王明霞;;基于加权小波分解的人脸识别算法研究[J];郑州大学学报(工学版);2014年01期
4 魏冬梅;周卫东;;采用压缩感知的人脸识别算法[J];计算机工程;2011年18期
5 陈皓;霍星;;视频监控中人脸识别算法稳定性的改进[J];工程图学学报;2011年06期
6 李美瑾;;人脸识别算法综述与远程考试系统应用研究[J];广西广播电视大学学报;2012年01期
7 盛朝强;王君;;煤矿井下人员签到系统人脸识别算法研究[J];计算机技术与发展;2012年07期
8 朱学毅;王崇骏;周新民;张W，
本文编号：2282324
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2282324.html
