Font Size: a A A

Methodology On Feature Extraction And Descriptors In Iris Recognition

Posted on:2008-12-03Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z H LiFull Text:PDF
GTID:1118360212497680Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
With the increasing requirements for security, biometrics based personal identification methods have been received extensive attention. Recently, iris recognition is becoming an active topic in biometrics due to its high reliability for personal identification. A great deal of progress in iris recognition has been achieved in the past decade, but there is a lack of methodology to instruct how to extract and describe iris features. In practice, iris features were extracted and described depending on researchers' experience and assumptions, which resulted in the blindness to describe iris features.To get rid of the above trouble, the principles on extraction and descriptor of iris feature are theorized in the thesis to analyse the the merits and disadvantages of the iris feature in theory and application on the choice of multi-scale analysis methods, the information of iris feature used, the range of iris feature represented and the form of iris feature described. The principles are described as follows: (1) the multi-scale analysis method is chose to extract the iris feature, which should have the well combination property of space and frequence, and also have the better selective capability of sensitive orientation. In practice, the multiwavelet, Gabor and steerable pyramid are better chose to analysis the multiscale iris feature. Further more, the steerable pyramid can also provide the translation invariance representation. (2) At the aspect of the information on the iris feature used, the orientation information should be first considered. (3) At the aspect of the iris feature representation's range, the local feature should be first considered. (4) At the aspect of the iris feature representation's form, it is difficult to decide which is better between structural feature and non-structural feature. In practice, the steadable and convenient representation principle should be taked into account. In all, the principles on extraction and descriptor of iris feature are theorized can be used to provide the guidance for extraction and descriptor of iris features.Biomimetic pattern recognition can effectively overcome the trouble of the N+1 category in traditional pattern recognition. According to the theorized principles on extraction and descriptor of iris features, it proposed a new method of multiscale iris recognition based on biomimetic pattern recognition in the thesis. Here, the biomimetic pattern recognition is firstly introduced to study the multiscale iris recognition. Meanwhile, multiwavelets provide more merits than wavelet and wavelet packets to analyze multiscale signature, then the GHM multiwavelets is used to represent the iris multiscale features, and the local orientation feature representation is designed on the selected subbanks. In training, a reason hypersphere point-set covering is constructed by SOM clustering and distance projecting distribution to represent the homology iris feature covering space. In recognition, Fusion of multiscale recognition is used to judge whether the sample belongs to the covering set or not to recognize it. Experimental results on JLU-IRIS databases show that the performance of the proposed method is encouraging, and the method can satisfy with the needs of practical application.Iris features which are invariant to iris imge translation, is an important content in iris recognition. According to the theorized principles on extraction and descriptor of iris feature, it introduced a new iris recognition method of local phase code based on complex steerable pryamid in the thesis, and the proposed method is translation invariance and its' local feature is stable and highly repeatability. The complex steerable pyramid transform provide the phase information of multiscale iris features, and the decidability is used to select the robust and highly dividing subbanks. The each n×n block is looked on as a local feature region to reduce the difficulty of local feature's repeatability detection, and the information of real and imaginary parts in each local block is added respectively. Then the sum of real parts and imaginary parts is encoded respectively, the each block is encoded 2 bits iris local code, so the 1920 bits iris phase code is encode. In recognition, the hamming distance is used to compare the distance between the two corresponding to subank codes, and the fusion of the compare results are considered to judge the iris image whethere comes from a person. Experimental results on JLU-IRIS and CASIA databases show that the performance of the proposed method is encouraging, and the proposed method is suitable for situations with high security requirements.The well software structure can improve compatibility, transplantable and opening of iris recognition system, and reduce the complexity of the iris recognition system. In the thesis, the Object-Oriented Peer-to-Peer architecture design pattern was used to design the iris recognition system software structure based on the Object-Oriented software engineering and 7 layers architecture of network protocols. The software design method provids the opening, compatibility experimental system platform for iris feature extraction and descriptor. Furthermore, the design method can offer reference and guidance for iris recognition system design standardization.The normalized iris image was processed by image capture, image quality assessment and iris image preprocess etc. The JLU-IRIS iris serial database, which iris images were captured from 241 volunteers, can be used for experiments, such as iris serial quality assessment, image auto-collected and multi-sample recognition. Accordingly, the normalized image and JLU-IRIS provide the foundation for iris feature extraction and descriptor. In addition, the iris localization algorithm of the voting optimized and coarse-to-fine strategy has been brought forward. The localization Experimental results on JLU-IRIS shows that the method is better than others and it have the better localization precision and the faster localization speed. Additionally, the local Fourier transform window and local histogram equalization were used to reduce the illumination sensitivity to iris images. According to human visual model, the wavelet transform was used to get the higher quality iris image by score of different resolutions in region of interest. Furthermore, the algorithm of orientation energy maximum and algorithms of eyelash and eyelid detection have been brought forward to reduce the rotation and occluaion influence.As a result, the main contributions in the thesis are that the theorized principles on extraction and descriptor of iris feature is proposed, and under the theorized principles, the new method of multiscale iris recognition based on biomimetic pattern recognition and the new iris recognition method of local phase code based on complex steerable pyramid are put forward. Experimental results on JLU-IRIS and CASIA databases show that performances of proposed methods are encouraging. Accordinly, the conclusions can provide the reference and guidance for extraction and descriptor of iris feature in a certain sense.
Keywords/Search Tags:Iris Recognition, Multi-scale Analysis, Feature Extraction and Descriptor, Biomimetic Pattern Recognition, Steerable Pyramid, Biometrics
PDF Full Text Request
Related items