图像特征提取算法在掌纹识别中的应用研究

发布时间：2018-05-18 19:07

本文选题：掌纹识别 + 特征提取　；参考：《吉林大学》2017年硕士论文

【摘要】：作为生物特征识别的一种,掌纹识别是通过提取手掌的纹理特征来进行身份识别的一种方法。与当前常用的指纹等识别方法相比,掌纹图像采集更容易被人们接受,且掌纹面积比较大包含更多的识别信息。因此,掌纹识别同其他身份识别相比将具有更广阔的应用前景。此外,掌纹采集对设备要求低,不需要高分辨率的成像设备,是当前研究的又一热点方向。当前已有多种掌纹识别方法,如:使用掌纹纹理的识别方法,掌纹主线的识别方法,掌纹方向的识别方法,特征相关的识别方法和掌纹图像子空间的识别方法等。掌纹识别过程主要包括:掌纹图像获取、预处理、特征提取、分类决策等几大部分。其中特征提取是掌纹识别的关键,在整个识别过程中占据着非常关键的位置。特征提取的好坏直接决定掌纹识别结果的优劣,如何有效的对掌纹图像进行特征提取是进行掌纹识别的关键。随着图像识别技术的进步,一些对掌纹原始图像特征进行识别操作的算法被提出。这些算法往往具有识别速度快,计算量小等特点。常用的图像特征,包括图像的点、线、面等特征,还包括图像的统计特征、图像的颜色特征和纹理特征等。当前经典的掌纹识别算法中既有基于点、线、面的特征提取算法,也有基于傅立叶变换和Gabor变换的频域特征提取算法等。其中,基于图像的点、线、面结构特征的方法是比较传统的特征提取方法;使用统计特征的方法常常使用灰度图像的统计特征,如方差、标准差等对图像原始特征进行定义和衡量;常见的空域-频域变换的特征提取,主要的变换方法有Fourier变换、Window Fourier变换和Wavelet变换等;基于子空间的特征提取又称为基于图像表征的识别方法,通过获得图像的局部特征,利用图像的这种局部特征构造特征向量进行图像的识别操作,实现从N维到M维的转换,常见方法有主成分分析和费舍尔线性判别等。在图像特征提取领域SIFT(Scale Invariant Feature Transform)算法具有较好的性能,但是其在掌纹图像识别中的研究较少,因此本文提出使用该方法进行掌纹特征提取。使用SIFT算法的关键是找到特征点,并生成特征点的特征描述子。该领域中SIFT算法有比较成熟的特征子生成方法,本文中通过对掌纹图像进行特征向量生成,然后利用欧式距离进行特征匹配计算,取得了较好的图像匹配效果。因此,本文对掌纹识别的图像特征提取关键技术和常用的匹配算法进行了详细的介绍,并对典型的图像特征提取算法进行了全面的探究,找出了各种算法在掌纹图像特征提取中的优劣之处,在这些算法的理解基础之上提出了基于SIFT特征提取的掌纹识别方法。本文介绍的用于掌纹图像特征提取技术能够为本领域开发人员提供技术参考,通过系统的阐述该领域用到的关键技术将为下一步相关应用的研发打下基础。
[Abstract]:As a kind of biometric recognition, palmprint recognition is an identification method by extracting the texture features of the palm. Compared with the commonly used fingerprint recognition methods, palmprint image acquisition is more easily accepted, and the palmprint area is larger than the palmprint area contains more recognition information. Therefore, palmprint recognition will have a wider application prospect than other identification. In addition, palmprint acquisition requires less equipment and does not require high resolution imaging equipment. At present, there are many palmprint recognition methods, such as: palmprint texture recognition method, palmprint main line recognition method, palmprint direction recognition method, feature related recognition method and palmprint image subspace recognition method. The palmprint recognition process mainly includes: palmprint image acquisition, preprocessing, feature extraction, classification decision and so on. Feature extraction is the key of palmprint recognition and occupies a very important position in the whole recognition process. The quality of feature extraction directly determines the result of palmprint recognition. How to effectively extract the palmprint image feature is the key to palmprint recognition. With the development of image recognition technology, some algorithms for recognition of palmprint original image features are proposed. These algorithms often have the characteristics of fast recognition speed and less computation. The commonly used image features include the points, lines and surfaces of the image, as well as the statistical features of the image, the color features and the texture features of the image. The current classic palmprint recognition algorithms include feature extraction algorithms based on points, lines and surfaces, and frequency-domain feature extraction algorithms based on Fourier transform and Gabor transform. Among them, the method based on the points, lines and surface structure features of the image is a more traditional feature extraction method, and the statistical feature of the gray image, such as variance, is often used in the statistical feature extraction method. The standard deviation defines and measures the original features of the image, and the main methods of feature extraction in spatial and frequency-domain transform include Fourier transform, window Fourier transform and Wavelet transform, etc. The feature extraction based on subspace is also called the recognition method based on image representation. By obtaining the local feature of the image and using the local feature vector of the image to construct the feature vector of the image to carry out the image recognition operation, the transformation from N dimension to M dimension can be realized. Common methods include principal component analysis and Fisher linear discriminant. In the field of image feature extraction, SIFT(Scale Invariant Feature Transform) algorithm has better performance, but its research in palmprint image recognition is less, so this paper proposes to use this method for palmprint feature extraction. The key of using SIFT algorithm is to find feature points and generate feature descriptors of feature points. In this field, SIFT algorithm has more mature feature generation method. In this paper, the palmprint image is generated by feature vector, and then the Euclidean distance is used for feature matching calculation, and a better image matching effect is obtained. Therefore, this paper introduces the key techniques and common matching algorithms of palmprint recognition in detail, and makes a comprehensive study of typical image feature extraction algorithms. The advantages and disadvantages of various algorithms in palmprint image feature extraction are found. Based on the understanding of these algorithms, a palmprint recognition method based on SIFT feature extraction is proposed. The feature extraction technology of palmprint image introduced in this paper can provide a technical reference for the developers in this field, and the key technologies used in this field will lay a foundation for the research and development of related applications in the next step.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】