Iris Outer Boundary Location Algorithm And Coarse Classification Algorithm

Iris recognition is a biometric technology, which identifies and verifies individuals by its biometrics character. Iris recognition is more stable and reliable compared with other biometrics technologies, which gives the system a widely application and a business future.Based on the previous research, the dissertation improved pre-processing module and added a rough classification module into the system. The main area has been included in this paper:Firstly, in pre-processing module, image quality evaluation was improved, especially the bright holes filling. The location of bright holes can be calculated in evaluation part. A fast filling algorithm is employed here based on morphological image analysis. A pixel queue will be given after forward and backward global point searching. In principle of first-in-first-out, expand the whole queue. Noises that gray values increase sharply will be removed and iris images without bright holes will be left.Secondly, improve algorithms for outer edge location to be more robust and adaptable. Get an approximate iris area and an eyelash threshold in inner edge location module. Detect iris edge to extract points on the boundaries by pixel gradient. Then classify the set of points to remove most fake edge points based on circle shape of iris'edge. Curve fitting based on improved least square method and get some circles to be collected. At last pick out the real one from the set of circles and output the result as the edge of iris.Thirdly, propose the concept: rough classification of iris recognition based on massive databases, while an algorithm of iris representation and an unsupervised clustering algorithm were employed to realize this function according to the massive databases. This paper analyzes the difference between rough classification and matching and raises three criterions of coding iris images in application to index. According to these requirements, present an algorithm on extracting statistical features from wavelet coefficients. Before matching iris codes, cluster the iris databases by unsupervised learning based on kernel methods. In the end, the clustering algorithm was verified by using SVMs because of its abilities of self-training and fast classification.The outer edge location was test in CASIA version2.0 and version3.0, which has an excellent result. Experiments for rough classification algorithm were carried out with CASIA 2.0 and a set of synthetic data, of which results show that the clustering method has a better performance.