The Research On Non Ideal Iris Image Preprocessing Algorithm

It is the inevitable trend of modern society to replace the traditional identity authentication method with biometrics.Because of its stability,uniqueness,security and easy to collect,iris recognition technology has been widely used in many fields such as national defense,financial management,government departments,security and so on.At present,iris recognition system can only achieve a high recognition rate in the laboratory environment,and the accuracy is less than expected.The difficulties mainly come from the iris segmentation of preprocessing part.For the non ideal iris images,such as unconsistent illumination,iris obscured by the eyelids,motion blur,strabismus,traditional edge detection algorithm has some limitations in Iris Segmentation.For example,the adaptability of the algorithm is not strong enough,the gradient threshold should be adjusted manually;Influenced by the texture background,the iris contour can not be accurately extracted.Therefore,how to obtain the accurate edge information in the non ideal iris image has always been the difficulty of iris segmentation.This paper uses the machine learning edge detection algorithm to transform the edge detection problem into the boundary point classification.This algorithm preserves the relationship between the boundary pixel and the surrounding pixels and the structure and context information of the iris.The algorithm has very strong generalization performance and does not need to adjust any parameters manually.First,the iris detection is performed to exclude the non iris or human eyes closed images.This step can coarse local iris region and decrease the images size.And then,we can obtain boundary point by Canny edge detection,and get the image window clipped by this point as the images center.We extract Haar and HOG feature of this windows and use the Probabilistic Boosting Tree classified edge points.By training the pupil,iris,eyelid three classification Trees,we can get accurate iris and eyelid boundaries.Since the boundary points obtained with high accuracy,we can use the simple algorithm such as least squares to fitting iris boundary.The parabolic fitting algorithm is also effective in eyelid location.In order to correct pupil boundary and detect the eyelashes and shadows,this paper uses the local OTSU algorithm to get different thresholds for segmentation.We use the elastic model to normalize the iris,and get the final iris image for recognition.We do the preprocessing algorithm experiments in Casia-irisV4-Thousand data base,and get 0.64%error rate.The 2D Gabor wavelet is widely used for feature extraction in recognization part.The performance of the proposed algorithm is proved by ROC curve parameter of the iris recognition system compared with the experiment results of the open literature.