【摘要】 在采集虹膜图像样本时,会由于各种原因而出现不同类型的坏样本。由于现有的图像质量评估方法需要在虹膜定位或者粗定位之后,根据虹膜的清晰度或分辨率来进行判定,因此只能检测出特定类型的坏样本。为了能对各种类型坏样本进行检测,在分析坏样本产生原因和类型的基础上,提出了一种实时预评估网络的检测方法,即在定位或者粗定位处理之前,预先对缓存中的图像进行评估,再根据预评估检测的结果来决定是重新采集,还是进入后续的处理。其目的是,①节省现有的图像评估在定位处理上花费的时间和降低采集失败率,以提高识别系统的友好性;②减少因为坏样本的输入而导致的定位出错,以避免引起误识别的可能;③提高识别率的同时降低识别系统的注册失败率。实验结果表明,这种方法不仅可以检测出多种类型的坏样本,且检测错误率低,同时具有较高的评估速度,可以满足实时虹膜识别系统的要求。更多还原

【Abstract】 There are different types of bad images when an iris identification application system captures iris images.Because previous image quality evaluation methods judge an image whether bad or else by the resolution and the definition of the iris part after the iris localization,they barely can sift a few types of bad images.For improving the performance of image’s real-time estimation,debasing the rate of failure to acquire and decreasing the localization errors possibility resulted from bad images,this paper proposes a new idea of the real-time pre-estimation network which pre-estimates images saved in memory before the localization or rough localization in an recognition process and decides whether it captures an iris image again or turns into the next step by the network output.The experimental result shows that the method can detects most types of bad images with fairly low false rate and little calculation and satisfies the requirement of a real-time iris recognition system.更多还原