Research On Visible Iris Recognition Based On Multi-View Angle

Iris recognition based on near-infrared light is an identification method with high recognition rate and robustness.With the widespread use of smartphones,there is a greater need to support iris identification on smartphones.Since most smart phones are not equipped with near-infrared camera equipment,there are great theoretical and practical significance to study iris recognition under visible light.Because the resolution of visible iris imaging of smartphones is low and the texture features are not clear,especially in Asians,the visible light transmittance of their iris is lower,so the realization of visible iris recognition on smartphones has difficulties such as low recognition rate,poor robustness,and low iris quality.Aiming at these problems,this paper proposes an iris recognition method based on multi-view iris texture and spot weighting fusion.Using a variety of fine structures presented by iris images at multiple viewing angles,combined with K-Median clustering algorithm and SVM classification algorithm,significantly improved recognition accuracy and reduced false positive rate.The main work and contributions of this paper are as follows.(1)In the iris pretreatment stage,the angle of the camera is unknown for the test,and the iris characteristics of different angles are different.The difference between different viewing angles of the same iris may be greater than the difference of different irises,which affects the recognition accuracy.A color and shape information of visible iris based on different viewing angles is proposed.The gray-scale distribution of hue saturation channel in HSV color space is combined with Hough transform and Canny edge detection to realize iris boundary location and angle localization based on direction code.And tested in our database and UBIRIS database to prove the robustness of the method.(2)In the feature extraction step,the texture and speckle features of the multi-view iris are extracted.According to the heterogeneity of iris texture structure under multiple angles,K-Median clustering and LGBP algorithms were used to extract iris texture features under multiple viewing angles,and using the iris spot information mapping to distinguish the pixel values in the Lab color space.Morphological algorithm combined with regionprops function was used to extract iris spot features at different angles.(3)In the training and recognition stage of iris model,a recognition method based on multi-view iris texture feature and spot feature fusion is proposed.Aiming at the inconsistency between the angle of the iris sample to be identified and the iris angle in the training sample,the training samples are augmented by the neighborhood image interpolation method.Based on the angular positioning of the iris,the test samples and angles are embedded into the feature vector,and the support vector machine method is used for classification training and recognition.Based on the multi-view texture and multi-view spot recognition weights,the texture and spot features are merged to realize iris feature fusion recognition.In order to verify the effectiveness of the method,at 11 different angles,a total of 2,640 iris images were collected from 30 volunteers as a training set.A total of 400 iris images from 5 perspectives of 5 volunteers were used as test sets.According to the experimental results,the multi-view iris texture recognition rate is 76.65%,the iris spot recognition rate is 58.85%,and the recognition rate of the two features is improved to 84.43%,which is obviously improved compared with the traditional method.