Defocus Blur Iris Image Quality Evaluation Method Based On Region Of Interest

Iris recognition is one of the most important method of identity recognition, which is widely used in recent years. Iris recognition system mainly includes:iris acquisition, image preprocessing, feature extraction, feature matching, and so on. And iris image preprocessing includes: iris segmentation, image quality evaluation, normalization and image enhancement. Defocused quality is one of the most common interference factors. So it is very important to evaluate the quality of the defocused image accurately before the iris recognition.In the thesis, an evaluation method of defocused blur iris image quality based on region of interest is summarized. The thesis work mainly includes the following 3 aspects:(1) Study on the theoretical basis. In order to understand the method of determining the region of interest, understand the characteristics of the defocused iris image and the knowledge of image quality evaluation, investigate and analyze the all related theories. The research status, development prospect and related research methods of iris image quality evaluation are summarized. Summarize the origin and the influence of the defocused blur iris image. Understand the method of image quality evaluation combined with machine learning. Summarize the relevant theory of SVM. The above works lay the theoretical foundation for the proposed algorithm framework.(2) Algorithm design. According to the analysis and summary of the iris image quality evaluation research status and deficiencies, summed up the algorithm framework. In order to evaluate the quality of iris image more effectively, combined with mathematical morphological processing, locate the gray iris image. Determine the region of interest, the region of interest including the part of the iris that is not covered by the eyelids and eyelashes. Then normalized the ROI area. Obtained the amplitude spectra by using 2D Fourier transform. Then, extract frequency domain feature, determine the quality score. Construct training samples, choose kernel function and parameters. Use the SVM multi-classifier to train and classify samples. Finally, realize the classification grade of the defocused iris image.(3) Algorithm simulation and experimental verification. The experiment was carried out by using the CASIA iris database of Chinese Academy of Sciences. Use the LIBSVM toolbox in the Matlab R2014 a software platform. Write the simulation program, achieve the simulation algorithm. Draw the simulation results and analysis the conclusions. Finally, presente the deficiencies of this algorithm and the future needs to be done in the further.The innovations of this thesis are summarized as the following two aspects.(1) When calculating the iris image quality score, in order to the different defocused blur image quality scores have great disparity, use the ratio of low frequency energy and high frequency energy, by using SVM training and evaluation get more accurate evaluation results of image quality.(2) The quality grades of the defocused iris image is divided into four grades, not the traditional two levels, which provides a great convenience and reference value for the following research.In this paper, the algorithm still exist deficiencies:(1) In the feature extraction, this paper only proposes one quality characteristics, this method can not reflect the direction information of the iris image. It is difficult to get the best model for evaluation by machine learning methods. Therefore, in order to evaluate of image quality more accurately, should extract a lot of different feature vectors.(2) When use the method of OAA, did not consider the effect of SVM multiple-classifier to test error rate. The larger of the training samples are, the more complex to train. In order to reduce the error rate, should take some measures in the further research. It is a worthy to further study problem that ensuring the operational efficiency and improving the accuracy at the same time.