Abstract: There exist frequently different types of bad sample images in an iris identification application system. When these bad images are imported into the identification process, generally it results in increased enrollment failure rate and localization errors or identification errors. According to the articulation and resolution of the iris part, previous image quality evaluation methods estimate whether an image is a bad or not after having calculated the iris location of an input image. So, only part of bad samples can be handled, and it is time-consuming. The reasons and characteristics that bad sample images were analyed. A real-time pre-estimation method for supporting vector machine's associated estimation network was proposed. Before the localization or rough localization process, sample images temporarily saved in memory are detected. According to the output results from pre-estimation network, the system determines re-acquisition or to turn into the next step. The experimental result shows that the method can detect most types of the bad sample images. Detection speed is fast and error rate is comparatively low. The method can satisfy the pre-estimation requirements of a real-time iris identification system.

Key words: Image pre-estimation, Real-time iris identification system, SVM associated pre-estimation network, Bad-image

摘要： 实际的虹膜识别系统会遇到因为各种原因产生的不同类型的坏样本图像，如果它们进入系统的识别进程，常常会增加系统的注册失败率，也会导致定位或者识别的错误。而现有的图像质量评估方法是在完成虹膜定位或者粗定位之后，根据虹膜部分的清晰度和分辨率来判定是否为坏样本。因此实际上只能处理部分类型的坏样本，而且计算耗费大。详细分析了坏样本产生的原因和特点，提出了一种基于支持向量机联合评估网络的实时预评估方法，在定位或粗定位开始之前，检测暂时存储的样本图像，根据预评估网络的输出结果来决定是进入下一步处理还是重新采集。结果表明，该方法可以检测出大部分类型的坏样本， 检测速度快，而且检测的错误率相当低，能够满足实时虹膜识别系统的评估实时性和准确性的要求。

关键词: 图像预评估, 实时虹膜识别系统, 支持向量机联合评估网络, 坏样本