A robust model-based iris segmentation

This paper proposes a method to segment iris area from the closed-up eye image. The method is a model-based method. It approximates pupil boundary and iris boundary using two circles, and approximates eyelids using two parabolas. The proposed method utilizes intensity gradient with local refinement to detect pupil boundary. A new concept of using signal to noise ratio (SNR), a ratio of mean and standard deviation of the edge image as a feature to detect outer iris boundary is introduced. Eyelids and eyelashes are located using dark line detector. The detected outer iris boundary and eyelids are modeled using a weighted integral approach. From our experiments, the SNR of an edge image is a robust feature. The proposed iris segmentation method can be applied on different iris databases without parameter tunings. The proposed method is validated using two databases: 250 images from CASIA-IRISV3-Interval database and 250 images from KSIP_DB01R database. Performance of our proposed method is evaluated by comparing the obtained segmented iris with its ground truth image. The ground truths of these 500 images are generated manually. The proposed system yields 92.44% correct segmentation rate (CSR) for CASIA-IRISV3-Interval database and 89.21% for KSIP_DB01R database. The average error rate (AER) of false accept rate (FAR) and false reject rate (FRR) is 5.44% for CASIA-IRISV3 and 8.16% for KSIP_DB01R. Additionally, results of our proposed segmentation method are compared to the results of the open source iris segmentation method. Our proposed system yields half the FRR error of the open source method.