Research On Unconstrained Iris Recognition System

An unconstrained iris recognition system is the extension of traditional iris recognition system. Unconstrained means under normal working conditions, the users’ behaviors do not require strictly restricted. Meanwhile, the iris recognition system can use a number of acquisition equipment to capture iris images for recognition at a distance, in visible light environment, and different deployment environment. It aims to reduce the environmental requirements of the iris recognition system. From the image information processing perspective, the main difference between the unconstrained iris recognition and traditional iris recognition is the iris images quality of diversity. There will be many cases traditional iris recognition technology cannot handle the non-ideal iris images. In this thesis, the unconstrained iris recognition system is the main line of research. Aiming at non-ideal iris images processing and across-environmental matching problem in the unconstrained iris recognition system research. This thesis mainly falls into the following aspects:1) To solve the quality diversity problem of iris images in the unconstrained iris recognition system. Two quality evaluation algorithms are put forward for defocus blur and motion blur images quality evaluation. The iris image evaluation of defocus blur and motion blur algorithm. In order to evaluate the degree of defocus blur, a defocus blur calculation method with registered image reference is put forward. By comparing with the high-frequency energy of the registered image, the method can accurately evaluate the degree of defocus blur. When the image contains motion blur, iris image will appear of directional changes in texture. Since the Fourier transform is very sensitive to such changes. A method by using Fourier transform twice is proposed to detect direction and degree of motion blur. The method can be directly used to estimate the degree of motion blur. In order to estimate the motion blur of the image more accurately, the motion blur evaluation method of referring to registered image is proposed.2) As for the weak texture problem in iris image, a non-ideal iris image enhancement algorithm based on local standard deviation is put forward. In order to enhance the strength of the texture and improve the image quality of the iris image, adaptive contrast enhancement method is utilized to image processing. According to the texture feature of the iris image, to avoid damaging texture in the iris, the local standard deviation and the global standard deviation of the image are used as the contrast gain parameters. For color images, it transforms RGB color space to Lab color space for image texture enhancement and achieves good results. In order to solve the motion blur problem of the image in unconstrained iris recognition system, an image restoration method based on Richardson-Lucy(R-L) deconvolution algorithm and parameter estimation is proposed. R-L deconvolution algorithm has a beneficial effect on image restoration, but the premise is to clear the restoration parameters of motion blur. Using image evaluation method to estimate the direction and displacement of motion blur. Establishing the correspondence between the motion blur degree and displacement, then it can estimate the direction and displacement of motion blur image. In order to improve the accuracy of image restoration, extend the estimated value from the center to both sides, then multiple image restoration algorithm is implemented. The consequences of multiple image restoration can be matched with the registration sample, which can improve the accuracy of the system identification.3) Built on the structural features of iris images, an iris image segmentation method using watershed and region merging is put forward. Total variation flow model is used to divide the eye image into structural part and texture part. Light spots in the pupil are removed by morphology processing. Watershed and region merging are both applied in the structural part. Structure part is divided into separate parts. The regions may contain iris are labeled. Then, merging iris regions using average gray rule and total pixels rule. The method has a good segmentation effect if the iris image has a clear pupil, and it can accurately segment the image for getting the iris region. A row-sequence edge points detection algorithm for locating iris contour is proposed. The algorithm considers every row-sequence in the image as a processing element, considers the shape of the object as detection feature, then uses a specific sequence detection method to determine the edge points. In image target detecting process, to compare with the background, the locating object has a special shape. The special shape can be regarded as low-high-low brightness structure in the row-sequence. The proposed algorithm can determine whether there low-high-low structure, and where is the position coordinates of the edge points. It can accurately detect the weld location by utilizing edge points detection algorithm. The proposed algorithm is also applicable to locate the unconstrained iris contour. It can obtain the edge points of the iris region, then Hough Circle Transform is used to vote for the edge points for finding a circle as the iris contour. Compared with the traditional edge detection method, the number of the iris edge points is greatly reduced, so the time of Hough Transform is greatly reduced, and the efficiency of the iris contour localization is improved.4) To improve unconstrained iris recognition system performance in different environments, a performance improvement method of unconstrained iris recognition based on domain adaptation metric learning is proposed. A kernel matrix is calculated as the solution of domain adaptation metric learning. Known Hamming distance computing by intra-class and inter-class is used as the optimization learning constraints in the process of iris recognition. An optimal Mahalanobis matrix is computed for the certain cross-environment system, then the distance between two iris samples is redefined. The experimental results indicate that the proposed method can increase the accuracy of the unconstrained iris recognition in different circumstances, improve the classification ability of the iris recognition system.In summary, this thesis takes related problems in unconstrained iris recognition system as the research object. Based on the framework of traditional iris recognition, the extended research is carried out. It provides some effective method for non-ideal image processing and improving performance of cross-environment iris matching in the unconstrained iris recognition system.