Abstract: Defocus assessment of iris images is crucial for iris recognition system. Traditionally, the spectral power in high frequency band is adopted to measure the iris image quality. However, these methods are easily affected by the illumination variation and outlier regions in iris images, such as eyelash or eyelid regions. This paper proposes a two-step framework for the iris image defocus assessment. In the first step, the traditional iris image defocus metric is introduced to identify severely defocus iris images. In the second step, the quality features of iris images based on steerable pyramid decomposition are extracted. Then, the radial basis network is adopted to formulate the relationship between iris image quality features and iris image quality levels on the synthetic database. Finally, the model trained on the synthetic database is directly used to predict the iris image quality. The experimental results conducted on Clarkson database demonstrate that the proposed iris image defocus assessment method not only distinguishes the clear ones from the defocus iris images, but also is more relevant to perceptual quality than state-of-the-art methods.

Key words: iris recognition, iris image quality assessment, defocus

摘要： 离焦模糊评价在虹膜识别系统中尤为重要。传统的方法是通过频谱分析测量虹膜图像的离焦模糊程度，这类方法容易受到光照变化以及睫毛和眼皮等噪声区域的影响。提出了一种由粗到精的虹膜图像离焦模糊评价方法。第一步，通过频谱分析去除严重模糊的虹膜图像，进行虹膜图像离焦模糊粗分类。第二步，通过方向金字塔分解，提取虹膜图像的质量特征。在人工合成的离焦模糊虹膜图像数据库中，利用径向基神经网络建立起质量特征与质量等级间的对应关系。通过建立起的模型进行实际的虹膜图像离焦模糊等级预测，以及虹膜图像离焦模糊精分类。在Clarkson数据库上的实验结果证明了该方法不仅可以准确区分清晰图像和离焦模糊图像，而且相比于传统的虹膜图像离焦评价方法更接近于人的视觉感知。

关键词: 虹膜识别, 虹膜图像质量评价, 离焦模糊