基于统计特征的非理想虹膜图像分割方法

Segmentation of non-ideal iris image based on statistical features

由于退化条件的存在,非理想虹膜识别的关键在于正确分割虹膜区域,这一区域包含能够用于个体识别的纹理。本文提出了一种基于统计特性的非理想虹膜图像分割方法,包括内边界定位、外边界定位和眼睑检测3个阶段。在内边界定位阶段,通过高斯混合(GMM)模型及多弦长均衡策略,实现对瞳孔及虹膜中心的精确定位;在外边界定位阶段,利用简化的基于区域信息的曲线演化方法,将其与序统计滤波(OSF)结合,以确保曲线收敛至虹膜外边界;在眼睑检测阶段,利用二次曲线对眼睑进行建模。对多个数据库进行实验的结果表明,本文方法能够有效克服反光、睫毛和眼睑遮挡、外边界模糊等不利因素的影响,精确实现了非理想虹膜图像的分割。

Since the presence of the degraded conditions such as illuminative variations, eyelashes or eyelids occlusions, ambiguous outer boundary, etc, the key of recognition for non-ideal iris in real application is to correctly segment iris region which contains texture features distinguishing a person from another. In this paper, we propose the segmentation method for non-ideal iris based on statistical features of images. It consists of three phases,i, e. , inner boundary localization, outer boundary localization, and eyelids detection. In inner boundary localization, this method localizes pupil and iris center accurately by exploiting Gaussian mixture model （GMM） and multiple strings equilibrium. By GMM,multiple Gaussian distributions are evolved to fit image histogram. For this reason, GMM is adaptive among iris images in different databases. In outer boundary localization, we present the simplified region-based eurve evolution which is combined with order statistical filters （OSFs） to guarantee its convergence to exterior iris boundary. Finally in eyelids detection we employ parabola to model iris eyelids. By evaluating the data bases, this method can segment non-ideal iris accurately by eliminating undesirable reflections and eyelash /eyelid occlusions.