Research On Iris Recognition Algorithms For Personal Identification

Biometrics refers to a personal identification technology based on the physiological or behavioral characteristics of human beings. As the novel biometrics, iris recognition technology has the physiological advantages of unique, stable, non-intrusive and anti-counterfeit features, so the performance criterions of correct recognition rate and error rate are better than other biometrics. Iris recognition technology is one of the best prospect biometrics, which gets broad international attentions with extensive markets and scientific study values.Aimed at the problems of iris recognition algorithms, in this paper the algorithms of iris localization, iris feature extracting and encoding and iris matching are researched. The main contributions of the work in this thesis are as follow:(1) Aimed at the problems of noise influence, higher time consumption and bad adaptive performance for iris localization, a rapid iris localization algorithm based on the method of iterative pixel ratio of cirque area is proposed. The 5 methods of cutting iris image, image sampling, iterative pixel ratio of cirque area, rapid circle detection of Hough transformation and layered localization theory are used to improve the algorithm speed. The morphological method is used to eliminate the pupil faculae. The parameters, such as the threshold of pupil segmentation, the small ranges of circle center and radius, and so on, are all gotten by computation with good adaptive performance. Four iris databases are applied in experiments. The experimental results show that the accuracy of the proposed algorithm is 97.75%～99.07% and the time consumption of that is 52.847～158.502 ms. It is a robust, rapid, adaptive iris localization algorithm with good comprehensive performance.(2) Traditional iris matching method based on Hamming distance of optimal offset is researched, and the iris matching method based on improved standard deviation of offset Hamming distances is proposed. At first, the iris feature extracting method based on Gabor filter is researched, and the new 16-channel Gabor filters with odd symmetry are gotten to extract iris texture features of different scales and different directions. Then the filter results of sample points are encoded by detecting if zero-crossing. At last, the parameter of improved standard deviation of offset Hamming distances is constructed for iris matching. The experimental results show that compared with traditional matching method, the correct recognition rate of the proposed matching method is higher 0.129% reaching 99.902% in database with noises and higher 0.165% reaching 99.949% in database with little noises. The experiments demonstrate that a good iris recognition method based on Gabor filter is proposed, and the proposed matching algorithm is a better matching method of iris code.(3) Aimed at the problems of bad anti-noise ability, lower recognition speed, lower security and small optional threshold range of linearity classifier for iris recognition, a new iris recognition method based on LBP operator is proposed. The LBP16,4 operator is used to extract iris texture features, and the parameter of improved standard deviation of offset means is constructed for iris matching. The experimental results show that the correct recognition rate of the proposed method can reach 99.976% without noise mask. The time consumption of feature extracting and matching is only 59.902 ms. The recognition result tends to rejection but not acceptance with high security and the threshold range of classifier is large. The proposed method has the advantages of high correct recognition rate, strong anti-noise ability, rapid recognition speed, high security, and large optional threshold range of classifier. The idea is simple and the method is easy to realize.(4) Aimed at the problem of the high storage consumption of the iris feature extracting method based on LBP operator, a compressing algorithm for iris LBP features based on Gaussian pyramid is proposed. The two steps of the compressing algorithms for original image by Gaussian pyramid and for LBP feature image by thresholding save the storage space of iris LBP features 32 times. Furthermore, the proposed method improves the speed of feature extracting and matching. It is a very valid compressing method for iris LBP features. The experimental results show that the correct recognition rate of the proposed method can reach 99.968% without noise mask. The time consumption of feature extracting and matching is only 16.496 ms, and is about 27% of that before feature compressing. It holds the advantages of high correct recognition rate, strong anti-noise ability, rapid recognition speed, and high security.