【作者】 孙艳杰；

【导师】 王科俊；

【作者基本信息】 哈尔滨工程大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 继单生物特征识别技术的广泛应用,多生物特征识别技术又以其高稳定性,高可靠性得到了越来越多的关注。这种多生物特征身份识别技术,可以利用不同生物特征之间的互补信息,弥补单模态生物特征识别技术的缺陷f如特征缺失、损失或质量较差),进而提高身份识别技术的准确率。本课题即是基于指纹与手指静脉双模态生物特征的研究。图像预处理是自动识别过程的第一个处理环节,它的好坏将直接影响到待处理图像在后续的边缘检测、图像分割、特征提取和模式识别的结果。然而在图像获取、图像传输等过程中,却又不可避免地受到各种噪声的干扰,因此,对获取的指纹和手指静脉图像进行预处理,成为一个很重要的研究课题。因为小波变换在时频域具有多分辨率的特性,可灵活地对信号局部特征进行提取和时变滤波,所以,基于小波的图像去噪成为现有去噪算法的主流。但在高维情况下,对于文中要处理的指纹和手指静脉图像所表现出来的丰富的线面纹理信息,给小波变换的应用带来了局限性。特别在噪声的影响下使得指纹图像的脊线结构不清晰,手指静脉的血管脉络模糊,这就限制了单小波理论算法去噪效果的进一步提高。因此我们开始关注具有良好性质的多小波理论,如Ridgelet、Curvelet及Contourlet这些多尺度几何分析方法。本文首先对这些多尺度几何分析变换方法进行详细地研究和讨论,包括原理、构造方法、性能等,在此基础上,以指纹及手指静脉图像为研究对象,利用多小波理论变换将图像变换到多小波域,进行去噪仿真实验研究。并通过信噪比,均方根误差等评判标准作为图像去噪后的质量评价,对指纹和手指静脉图像的去噪结果做了相应的分析。最后,论文对处理后的指纹和手指静脉图像进行了进一步的图像分割,图像二值化和细化等过程。更多还原

【Abstract】 As a single biometric identification technology has been widely used, a multi-biometric identification technology has received more and more concern, owing to its high stability and high reliability. Multi-biometric identification technology will further enhance the recognition rate. It can use complementary information among different kinds of biometric, which make up for the defects of single mode biometric technology. The defects include trauman, loss of characteristics and bad quality of the characteristics. In this paper, the dual-mode biometric based on the fingerprint and finger vein features were studied.Image pre-processing is the first section of the automatic identification process, whose results will directly affect the following section such as edge detection, image segmentation, feature extraction and recognition accuracy. However, in the process of image acquisition and transmission, image will inevitably be subjected to a variety of noise influence. Therefore, the image pre-processing of fingerprints image and finger vein image has become a very important topic.Because of multi-resolution characteristics of wavelet transform in time-frequency domain, it is flexibility enough to extract the local features of the signal and time-varying filter. Therefore, denoising algorithm based on wavelet has become the mainstream of the existing image denoising algorithms. However, fingerprints and finger vein images with high dimension treated in this paper show that the texture information of the noodle-line is rich. It is limited for application of wavelet transform. Particularly, the ridge structure of fingerprint image was not clear and vascular collaterals of the finger vein was fuzzy in the influence of noise, which is limited for the single-wavelet theory algorithm to further improve the denoising effect. Therefore, we focus on the progress of the multi-wavelet, such as Ridgelet, Curvelet and Contourlet. All of these belong to multi-scale geometric analysis.Firstly, multi-scale geometric analysis transformation methods were studied and discussed detaily in this paper, including principles, construction methods, properties and so on. On this basis, denoising simulation of fingerprint and finger vein images was studied. Using multi-wavelet theory, the images were transformed in multi-wavelet domain. Next, we evaluated the denoising quality of image through the criteria such as Signal-to-Noise Ratio, root mean square error etc. and analyzed the corresponding denoise results of the fingerprints and finger vein image. Finally, further image segmentation, binaryzation and thinning processes are employed to the fingerprints and finger vein image processed.更多还原