基于特征掌纹的在线掌纹识别方法研究

【作者】 许慧；

【导师】 林家恒；

【作者基本信息】 山东大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 掌纹是一种重要的生物特征，具有采样简单、特征唯一、信息量大等优点，掌纹识别方法的研究具有重要研究价值。本文提出了一种利用最大内切圆确定和分割掌纹有效区域的方法，并尝试使用形态Harr小波和形态中值小波进行非线性多分辨率分析，以及基于傅立叶变换的旋转平移校正等预处理。然后定义了“特征掌纹”，并用于掌纹识别。最后通过实验验证了一系列预处理方法和识别方法的可行性。主要研究内容为： 1．在线掌纹图像有效区域的定位是整个识别系统中的一项重要的基础性工作。本文结合掌纹图像本身的特点，提出了一种利用最大内切圆确定和分割掌纹有效区域的方法。与现有方法相比，该方法不需要经过任何基本图像预处理(去噪声、图像增强、二值化、细化等)、角点检测和边缘检测，鲁棒性好，且不易受掌纹图像质量的影响。 2．使用形态小波对掌纹图像进行非线性多分辨率分析，缩小了原始图像的大小，既抑制了小细节又突出了最主要的特征，进而减小了后续旋转平移校正的计算量。 3．旋转平移校正是特征匹配及识别的关键。在形态小波非线性多分辨率分析的基础上，本文采用了基于傅立叶变换的旋转平移校正，其实质是相位相关算法。该方法对噪声容忍程度高，检测结果基本不受光照条件影响，对图像灰度依赖小。 4．在线掌纹图像中的掌纹特征稳定、表示方法简单，可实现快速匹配和识别，但是对比度低和强噪声也是在线掌纹分析中无法调和的一对矛盾。因此如何表征在线掌纹的特征是掌纹识别系统中关键性的问题。本文将PCA方法应用于掌纹识别，定义了“特征掌纹”，主要目的是选择较少的相互独立的新特征向量来描述原始掌纹数据样本。这样，掌纹识别就可以在降维后的空间上进行。 5．最后对所设计的在线掌纹识别系统进行了一系列实验对比。实验结果表明，本文设计的在线掌纹识别方法具有较高的正确识别率。更多还原

【Abstract】 As one of the most important biometrics features, palmprint with many strong points, such as simple sampling, one and only feature, information in abundance has significant influence on research. According to the character of palmprint, the dissertation proposes a largest inscribed circle based on positioning and segmentation method, which does not need basic preprocess, and does not need any contour detection and key points detection. The dissertation tries employing multiresolution image decomposition based on the 2-D morphological wavelet transform then. After morphological wavelet transform, rotation and translation compensation based on FFT is performed then. The eigenpalmprints are defined and applied them to identifying the online palmprints. At the end of the dissertation, a series of experiments are carried out to prove the feasibility of the preprocessing method and the recognition method. The main contents are as follows:1. The positioning and segmentation in online palmprint system is a basic and important job. According to the character of palmprint, the dissertation proposes a biggest inscribed circle based positioning and segmentation method, which does not need basic preprocess and does not need any contour detection and key points detection. This method is quite robust and can deal with more palmprint distortions. And the palmprint image’s quality can not have influence on the method easily.2. Employing multiresolution image decomposition based on the 2-D morphological wavelet transform , based on the enhancement of the main principal features in a palmprint image, can not only wipe off fine ridges, but also minish the size of the original palmprint image, in order to speed up rotation and translation compensation based on FFT later.3. Rotation and translation compensation is the key issue for the feature matching and recognition. Rotation and translation compensation based on FFT is performed based on the 2-D morphological wavelet transform. In this dissertation, the rotation and translation compensation is an extension of the phase correlation technique更多还原