Research On Iris Recognition Methods Based On Wavelet Packet Neural Network

With the development of the computer technology, communication technology and digital informatization technology, the technology of personal identification based on biological characteristic are promoting the traditional security into a new level, and put forward new requirements for the personal identification at the same time. Iris recognition is regarded as the most reliable and accurate biometric identification system available. The iris recognition is studied further in this thesis.1. The iris recognition system consists of an automatic segmentation system that is based on improved Canny edge detection and the Hough transform, and is able to get the centre and radius of inner circle of iris, occluding eyelids and eyelashes by threshold method, and reflections. Testing results have shown that the proposed method can improved iris location and is not sensitive to noise. In order to remove the effects caused by displacement and zoom of iris image, the extracted iris region was then normalized into a rectangular block with constant dimensions to account for imaging inconsistencies.2. In the light of the methods of forerunner, the processes that iris segmentation, pre-processing, feature extraction, and pattern match have been carried out. Corresponding to the specialist of the distribution of iris texture, the work presented in this thesis involved developing an'weighted statistics'iris recognition method so that more features are extracted in the inner iris region by weighted statistics method. Based on the recent advancements in iris recognition, a recognition approach using the technique of principal components analysis image processing and neural network pattern recognition is presented in this thesis. Some new algorithms and experiment results are also described. The algorithms of iris recognition have also been studied deeply.3. For compact integration between the feature extraction and classification stage, two kinds of algorithms are considered in order to improve the adaptive learning mechanism of the classification of iris recognition. One of them is wavelet packet neural network method and the other is Gabor wavelet network. Gabor wavelet network are used to extract the iris texture features in the space-frequency-scale domain by Gabor-basal function transformation. Finally, lots of transformations are carried out to find the optimum values of the parameters to guide the characteristics extraction and classification. The two algorithms performed with perfect recognition on a set of 420 eye images; however, the experiments implemented on CASIA iris database V3.0 show that, the method based on Gabor wavelet neural network performs very well, with a success rate of 99.3% in recognition rate. Experiments show that the method is valuable in iris recognition that possesses the advantages of high precision, high response speed, and great adaptability.4. In the decision-making of classification, the idea of integrated classifiers-based neural network is proposed for classification. The decision-making rule is built by the combination conclusions of the basal classifiers. Finally, according to the output of the composite indicator, which confirms the robustness of the learning mechanism of classification model.