应用于健康诊断的异常虹膜纹理识别方法研究

An abnormal iris texture may strongly indicate that the body be in a sub-healthy condition or may encountering specific illnesses. So it can be an easy, economic and non-wounding way to examine the patients' health condition by checking their irises. This method is especially suitable for those patients who have been living in a lowly developed area with poor medical conditions and economic situations, as well as those who can not go to the hospital due to timeless or inability for the regular interval body inspections. This research interests also corresponding to the national policy of "vigorously develop the technology of health-state diagnoses" in our nation's Twelfth Five-Year Plan. At present, examining the body situation based on the iris' condition is mainly accomplished by iridologists, which could take long to popularize. Although there are some research reports concentrating on how to automatically recognize the abnormal parts of the iris, their works are unsystematically designed and lacking of experimental evidences. Differentially, this project is aiming to develop an automatic iris texture recognizing system that can examine the abnormal parts of the iris and automatically reveals the overall health condition of the patients by consulting the standard international iris chart. To put this into practice, we mainly consider the methods of how to extract the iris area of the eyeball; examine the existence of the radial canal, the contradiction ditch, the pigmented spots as well as the pit hole and judging their position, size and depth; recognize the density of the iris and examine whether the two pupils are of an abnormal condition as for their size, differences on zooming rates, abnormality of transforming and etc. Based on the theoretical analysis and massive experimental evidences, we can propose effective arithmetics and corresponding mathematical models to form a relatively rounded system on recognizing abnormal textures of the iris. Meanwhile, these efforts can provide supports on realizing the iris based system to diagnose people's health condition automatically.

异常虹膜纹理提示身体很可能处于亚健康或者疾病状态，这种基于虹膜纹理信息的身体健康状况检查方法简便、低价、无创伤，非常适合医疗条件差、经济条件有限、无暇或不便定期体检的人群使用，符合我国十二五规划中"大力发展健康状态辨识技术"的国策。目前主要凭借虹膜师观察做出健康状况诊断，普及速度缓慢。虽有自动异常虹膜纹理识别研究报道，但零星分散缺乏系统性和实验验证。本项目研究异常虹膜纹理的机器自动识别方法，主要研究：从人眼图像中提取虹膜区域，在虹膜区域中检测放射沟、收缩沟、色素斑、坑洞是否存在以及存在的位置、大小、深度，检测虹膜纹理的密度，检测瞳孔是否存在过大、过小、两孔大小不等、变形等异常形态。借助国际标准虹膜分区图，给出身体健康状况综合评价。通过理论分析和大量实验给出上述内容的有效实现算法，建立相应的数学模型，形成一套较完整的异常虹膜纹理识别系统，为实现基于虹膜信息的健康状况自动诊断系统提供理论支撑。

虹膜图像中的异常特征包括很多种，其主要包括：典型的环状特征（角膜老年环，K-F环，神经压力环，卷缩轮）、线状特征（阳光放射线）以及点状特征（色素斑，坑洞）等。对这些异常特征的检测不仅仅可以用来分析人体的健康状况还可用于不同个体间身份信息的识别。本项目对于计算机辅助虹膜图像中的异常特征检测若干问题进行了研究，旨在辅助提高虹膜识别的准确率以及为计算机辅助虹膜诊断系统的研究提供解决方案。本项目针对应用健康诊断虹膜图像采集设备光源太强导致被采集者不适的问题，设计开发了两种虹膜图像采集设备；针对被采集者在自然睁开状态下采集到的虹膜图像容易受到睫毛及眼睑遮挡问题，提出了基于改进蚁群算法的睫毛检测与基于并查集算法的虹膜图像有效区域分割算法；针对于对于虹膜偏色图像校正，提出了一种基于直方图匹配的虹膜偏色图像校正方法，此方法可以实现将认为偏色的图像颜色与目标图像颜色匹配，得到与目标图像颜色接近的虹膜图像。.本项目提出一种基于统计规律与基于梯度极值的卷缩轮提取算法，通过对卷缩轮的提取为后续实现虹膜异常特征的计算机自动提取与识别以及病灶特征的匹配与分析提供了技术基础。针对虹膜块状纹理的检测问题，提出三种实现虹膜图像块状纹理检测的方法，这些方法能够实现仅存在色素斑的简单背景虹膜图像和多种纹理并存的复杂背景虹膜图像中块状纹理的检测；针对虹膜色素斑块检测问题，提出三种方法；针对虹膜坑洞的检测问题，提出四种方法。这些方法适用环境不同，检测正确率和算法复杂度各具优劣。算法遵循由粗到精的思路，最终得到用于虹膜识别和虹膜诊断的块状纹理特征参数。.针对角膜老年环与K-F环的粗分类问题，提出了基于HIS色彩空间的K-F环粗分类方法，提出了基于多尺度颜色替换的角膜老年环粗分类方法。针对这两种环状特征的检测，提出了基于梯度响应最优的环状特征边界提取方法并通过定义的宽度特征以及颜色特征定量的对环状特征进行量化分析。提出一种应用于虹膜图像粗分类的自仿射拟合纹理分割算法，实现虹膜肠环外边界检测。该方法在多尺度下拟合图像局部到整体信息的自仿射性，有效地在小区域内描述肠环内外纹理的差异，通过分析差异的变化规律确定肠环区域。此外对于线状特征的检测也取得了很好的效果，该方法在不需要手动去掉卷缩轮内区域的情况下，能够在多种虹膜纹理并存的复杂背景下检测出虹膜线状特征。