基于改进PCA和多特征分类器集成的掌纹识别

发布时间：2018-10-20 13:29

【摘要】：掌纹识别是一种新兴的生物特征识别技术。掌纹识别是用掌纹特征来进行身份鉴别的一种方法。主成分分析(Principal component analysis,PCA)是一种经典掌纹特征提取方法,但它也存在一些不足:丢失部分信息,受旋转、光照强度的影响较大等问题。这在一定程度上影响了PCA的效果。本文针对上述问题提出两种改进的PCA算法,它们分别侧重掌纹的全局特征和局部特征。在对掌纹识别系统的研究中发现,全局特征和局部特征在掌纹感知过程中所起的作用不同,所以综合这两种特征的识别算法的研究是十分有意义的。针对上述问题,本文主要的研究工作如下:(1)针对PCA算法中存在的问题:只保留原始图像的主要信息,会丢失部分信息。结合原始掌纹图像的强弱对比不明显的特点,提出基于傅里叶与PCA特征相结合的掌纹识别方法。首先对掌纹图像进行傅里叶变换,然后对变换后的图像进行主成分分析,最后利用LS-SVM分类器进行识别。实验验证了所提方法的有效性。(2)针对PCA算法存在的问题:受旋转、光照强度变化的影响较大。提出基于分块PCA(ModulePCA)和LS-SVM相结合的掌纹识别方法。该方法特征提取时将一整幅掌纹图像分为4×4个子块图像,在此基础上应用主成分分析。在单样本掌纹识别问题中优势明显,是因为采用分块PCA方法变换后的图像与原有掌纹图像共同作为新的训练集进行训练,这样有效的扩充了样本数量。实验验证了该方法的有效性。(3)傅里叶变换使掌纹图像的主要信息更集中且分块PCA会更好的保留局部信息,使图像鲁棒性增强。由于掌纹图像不仅包含全局特征也包含丰富的局部特征,综合应用两种上述方法,本文设计基于全局和局部特征相融合的掌纹识别系统。掌纹图像经过傅里叶变换得到全局特征;经过Gabor变换,再运用分块PCA算法得到局部特征。在识别阶段,采用由粗到精的串行集成策略来集成全局和局部分类器。实验验证该系统可以兼顾掌纹识别的速度与精度,增强掌纹识别系统的实用性。本文实验所使用的数据集来自香港理工大学掌纹数据库(Poly U)。
[Abstract]:Palmprint recognition is a new biometric recognition technology. Palmprint recognition is a method of identification using palmprint features. Principal component Analysis (Principal component analysis,PCA) is a classical palmprint feature extraction method, but it also has some shortcomings: loss of some information, rotation, light intensity and other problems. This affects the effect of PCA to some extent. In this paper, two improved PCA algorithms are proposed, which focus on the global and local features of palmprint, respectively. In the research of palmprint recognition system, it is found that global feature and local feature play different roles in palmprint perception, so it is very meaningful to study the recognition algorithm which synthesizes these two features. The main research work of this paper is as follows: (1) aiming at the problem in PCA algorithm, only the main information of the original image will be retained, and some information will be lost. A palmprint recognition method based on the combination of Fourier and PCA features is proposed. First, the palmprint image is transformed by Fourier transform, then the transformed image is analyzed by principal component analysis (PCA). Finally, LS-SVM classifier is used to recognize the palmprint image. The experimental results show that the proposed method is effective. (2) the problem of PCA algorithm is that the variation of light intensity is greatly influenced by rotation. A palmprint recognition method based on block PCA (ModulePCA) and LS-SVM is proposed. In this method, an entire palmprint image is divided into 4 脳 4 sub-block images, and principal component analysis is applied. The advantage of single sample palmprint recognition is due to the use of block PCA transform image and the original palmprint image as a new training set, which effectively expands the number of samples. Experimental results show that the method is effective. (3) Fourier transform makes the main information of palmprint image more centralized and block PCA can better preserve local information and enhance the robustness of the image. Since palmprint images contain not only global features but also rich local features, a palmprint recognition system based on the fusion of global and local features is designed in this paper. The global feature of palmprint image is obtained by Fourier transform, and the local feature is obtained by using block PCA algorithm after Gabor transform. In the recognition phase, a coarse-to-fine serial integration strategy is used to integrate global and local classifiers. Experiments show that the system can give attention to the speed and precision of palmprint recognition and enhance the practicability of palmprint recognition system. The data set used in this experiment is from the palmprint database (Poly U). Of the Hong Kong Polytechnic University.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】