Study On Visible Iris Recognition On Mobile Terminal

With the rapid development of economy and technology,mobile smart devices such as smart phones,have been gradually popularized.In order to ensure the information security of mobile smart devices,using biometrics technology to achieve mobile identity recognition has become a research hotspot.Iris recognition technology has become one of the hotspots of biometric recognition technology because of its advantages.In recent years,infrared iris recognition technology has been applied in smart phones.However,infrared image acquisition can cause eye damage and increase the cost of mobile devices.Compared with infrared iris recognition,the iris recognition in visible mobile terminal can use the image acquisition sensor which is safer and cheaper.However,because of the unstable illumination conditions and acquisition equipment,there are high intra-class differences in recognition samples and the image is susceptible to interference,which affect the accuracy of identification.Current recognition methods cannot solve these problems.Therefore,this thesis focuses on the problems encountered in visible iris recognition on mobile terminal.Aiming at iris image preprocessing,feature extraction and matching,two kinds of visible iris recognition methods for mobile terminal are proposed,which are suitable for different illumination conditions and different image acquisition sensors.In the iris image preprocessing stage,in order to reduce the influence of visible light interference on image quality,haze removal algorithm using dark channel prior is used to restore and enhance the normalized iris image.In the feature extraction stage,an operation structure of analogous convolutional neural network is proposed.The structure consists of two convolution layers and two pooling layers.It reduces the image dimension,enhances the iris global texture features.The first method is to build the iris feature vectors based on local block extraction and euclidean distance classifier is used for classification in the matching stage.Collaborative representation is used to extract and match the convolution dimensional reduction results directly in the second method.In this thesis,the algorithm is simulated and evaluated using the MICHE-I iris image dataset,which is taken by the iPhone5 and SamsungGalaxyS4.The experimental simulation platforms for different illumination conditions and different sensor devices are constructed respectively.The first method achieves a recognition rate of 98% when the illumination conditions are the same,and 77.5% when the illumination conditions are completely different.It indicates that the method can be applied to mobile iris recognition registration and actual recognition in different illumination environments.The second method is experimented on the iris images collected by two different sensor devices,namely,the iPhone5 and SamsungGalaxyS4.The second method achieves a recognition rate of 87% when indoor and outdoor illumination conditions change.The iris images collected by the iPhone5 device are used as training samples,SamsungGalaxyS4 as testing samples,and the recognition rate of iris is 85%.The experimental results show that the second method is suitable for different sensor devices and different illumination conditions.The results show that the proposed method can effectively overcome the problem of iris image noise in visible light,and the method has better recognition accuracy,can be applied to a variety of sensors,and has strong stability in different environments.