虹膜定位算法研究及其嵌入式软件平台设计

【作者】 鲁剑；

【导师】 解梅；

【作者基本信息】 电子科技大学 ， 信号与信息处理， 2008， 硕士

【摘要】 虹膜识别技术是一种新兴的生物特征识别技术。相对其他生物特征识别技术(指纹识别,面部识别,声音识别等)而言,虹膜识别技术具有以下几个特点:稳定性、高可靠性和非接触性。这些特点使得虹膜识别具有非常广阔的应用前景。本文在前人研究的基础上,对虹膜识别的一些核心算法、嵌入式虹膜识别系统都做了一定的研究和改进。本文主要的研究工作包含以下几个方面:一、在虹膜图像定位算法中,本文研究了二种新的虹膜定位算法。第一种算法是基于灰度信息和分数阶微分边缘检测的,在内边缘定位上由于内边缘定位比较简单因此采用快速的基于灰度信息的方法,在外边缘定位上采用分数阶微分边缘检测的方法。另外一种算法是基于SUSAN边缘检测算子的,主要是侧重于外边缘的定位。由于SUSAN边缘检测算子具有积分特性,不会对虹膜的纹理细节信息造成大幅度的衰减,并且对局部噪声不敏感,因此非常适合在虹膜外边缘定位上的运用,取得了很好的定位效果。二、在虹膜归一化算法方面,本文研究了一种基于极坐标变换的虹膜归一化算法。该归一化算法采用同心圆的方式进行极坐标转换以方便运算,针对归一化图像对比度比较低的情况进行了光照估计。采用了双三次插值和直方图均衡化的方法对虹膜归一化图像进行了增强。三、阐述了嵌入式虹膜识别系统的相关硬件,针对项目情况设计了基于TI DSP平台的DSP/BIOS嵌入式系统,在系统中进行了任务设计调度以及任务间的通信和同步,给出了算法在TI TMS320C6713b上的移植和优化方法。本文研究的虹膜识别核心算法在CASIA虹膜数据库(version 2.0)及实验室虹膜数据库中进行了测试实验,并得出了较好的识别效果。更多还原

【Abstract】 Iris recognition is an emerging biometrics technology. Compared with the other biometrics technologies (fingerprint recognition, facial recognition voice recognition, etc.), iris recognition has the following characteristics: stability, high reliability, non-contact. Therefore, it will be widely used in the future.This thesis made research and improvement of the iris recognition kernel algorithms. The main tasks of this thesis include:First, at the iris image localization stage, two new localization methods are proposed. The first method is based on gray information and fractional differential. In iris inner localization, compare with iris outer localization, iris inner localization is simple, so use the fast method based gray information. In iris outer localization, use fractional differential edge detection. The second method is based on SUSAN edge detector. This method is emphasized on iris outer localization. The SUSAN edge detector has the integral characteristic, so it will not depauperate much iris texture. Also, SUSAN edge detector will not impacted by local noise.Second, at the normalization stage, we propose a polar coordinate transform resolution. Use concentric circle as the normalization style. Take illumination estimation to change the low contrast of mormalized image. Also, to enhance image, we use bicubic interpolation and histogram equalization.Third, introduce the hardware of embedded iris recognition system, design DSP/BIOS based on TI DSP. Transplant and optimize the algorithm to TI TMS320 C6713b DSPs platform.The algorithm experiments are done in the CASIA iris database (version 2.0) and Lab iris database, and good recognition results are achieved.更多还原