基于LBP算法的人脸识别研究

发布时间：2018-03-17 10:38

本文选题：人脸识别　切入点：特征提取　出处：《安徽理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着计算机和信息技术的快速发展,人脸识别技术越来越受到重视。本文主要研究了人脸在不同光照、不同表情下的特征提取与识别的一些关键问题,提出了一些改进方法,并通过实验进行了可靠性验证。针对LBP算法提取人脸图像的表情特征信息时会丢失特殊的特征信息的缺点,本文提出了多重局部二值模式的人脸表情识别方法(Multiple Local Binary Patterns,MLBP),该方法在保持LBP算法优点的前提下,通过增加一位二值编码,利用中心像素点作用以及邻域像素点灰度值之间的关系,得出特征向量图。实验结果表明MLBP算法比LBP算法描述的表情纹理图像更加均匀,且识别率约提高10%。针对人脸表情图像进行纹理特征提取时的模块大小划分问题,本文提出将MLBP算法与Harr小波分解相结合,该方法首先将表情图像进行Harr小波分解,得到四幅不同频率的子图像,然后对其中三幅图像进行MLBP特征提取,并将得到的特征值串联形成表情图像的特征向量。实验结果表明该方法比MLBP方法直接提取表情特征所产生的特征向量的维数减少了 25%,特征提取和识别的速率提高了,其中识别率约提高了 9%。人脸识别研究中的识别率容易受光照强度的影响。针对MLBP算法在光照变化时具有旋转不变性,以及Gabor小波能提供空间位置、空间频率的特性,本文提出了多重局部 Gabor 二值模式方法(Multiple Local Gabor Binary Pattern,M LGBP),该方法先对人脸图像使用Gabor小波进行变换处理,保留受光照影响较小的高频部分,然后再采用MLBP算法对Gabor提取后的图像采用分块编码,最后得到联合直方图序列,获得丰富的局部特征信息。实验结果表明了该算法有效的降低了光照对识别率的影响,提高了光照不均匀时的人脸识别率,且在特征提取方面比Gabor等算法更加有效。
[Abstract]:With the rapid development of computer and information technology, more and more attention has been paid to face recognition technology. In this paper, some key problems of feature extraction and recognition under different illumination and expression are studied, and some improved methods are proposed. The reliability of the algorithm is verified by experiments. The LBP algorithm will lose the special feature information when extracting facial expression information from the face image. In this paper, a multiple Local Binary pattern algorithm for facial expression recognition is proposed. The method preserves the advantages of the LBP algorithm by adding a bit of binary coding. Using the relationship between the central pixel and the gray value of the neighboring pixel, the eigenvector graph is obtained. The experimental results show that the MLBP algorithm is more uniform than the expression texture image described by the LBP algorithm. And the recognition rate is increased by about 10%. Aiming at the problem of module size partition in facial expression image texture feature extraction, this paper proposes the combination of MLBP algorithm and Harr wavelet decomposition. This method firstly decomposes facial expression image by Harr wavelet decomposition. Four sub-images with different frequencies were obtained, and three of them were extracted by MLBP. The experimental results show that this method reduces the dimension of feature vectors generated by MLBP method and improves the speed of feature extraction and recognition. The recognition rate is increased by about 9%. The recognition rate of face recognition is easily influenced by illumination intensity. Aiming at the rotation invariance of MLBP algorithm when illumination changes, and the feature that Gabor wavelet can provide spatial position and frequency, the recognition rate of face recognition is easy to be influenced by illumination intensity. This paper presents a multiplex local Gabor binary mode method called multiple Local Gabor Binary pattern (MGBP). In this method, the face image is first processed by Gabor wavelet transform, which preserves the high-frequency part which is less affected by illumination. Then the MLBP algorithm is used to code the extracted Gabor images in blocks. Finally, the joint histogram sequence is obtained, and the local feature information is obtained. The experimental results show that the algorithm can effectively reduce the influence of illumination on the recognition rate. The face recognition rate of uneven illumination is improved, and the feature extraction is more effective than Gabor algorithm.
【学位授予单位】：安徽理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】