The Research Of Iris Image Quality Evaluation And Acquisition

In this paper, I proposed to divide the iris image quality evaluation process of iris recognition technology into rough evaluation and meticulous evaluation. Firstly, rough locate the iris image which acquired in real-time and then rough evaluate them, based on its main characteristic of fast and real-time, it can effectively evaluate the ROI of the real-time iris images and filter out images that obviously do not meet the recognition requires. Secondly, locate the iris image with non-concentric location algorithm to get iris area; evaluate iris image meticulous with the effective area detection algorithm and the JND information detection algorithm, the evaluation qualified image can do the follow-up feature extraction and recognition. I conduct the recognition experiment to JLU_IRIS1、JLU_IRIS2、CASIA4three iris image database and compare the recognition rate, to verify the validity of the iris image quality assessment algorithm from subjective, objective and recognition rate. This article mainly divided into several aspects as follow:1. Biometric and information security laboratory researched and invented iris recognition device independently and collected the two generations of iris databases. By listing and analyzing factors that influence the quality of iris image during the process of acquisition, proposes the focus and the algorithm to solve problems above in this paper.2. Roughly locate the real-time iris image, get the general areas of iris, set the ROI of image for the for follow-up advance processes. The character of the algorithm is its short execution time, the purpose of the algorithm is to effectively narrow the range of the image for quality evaluation process. Large number of experiments show that the algorithm can accurately locate the outer boundary of the iris, carve out of the target area.3. Rough evaluation phase, I use the standard deviation、gradient、edge strength、 fuzzy entropy、information entropy five factors as evaluation factors, I propose the average error probability algorithm to train the weights of the five factors and get the final quality result of the image, the feature of this algorithm is its speed that made it more suitable for real-time iris image evaluation. Large number of experiments of JLU_IRIS1、JLU_IRIS2、CASIA4three iris databases’s range images show that the algorithm divide the quality of image very clear, I set the threshold0.4, it can effectively filter out the qualified image for the follow-up process, and the execution time of the algorithm is within0.2s.4. In the pretreatment part, I use the non-concentric method to locate iris images, proposed the iris outer circle location algorithm based on the partial average gray gradient, the core idea is to detect iris outer edge of the iris using the difference between the gradation values of iris and sclera, detect the outer circle center in the range of five pixels within the inner circle center, the outer circle radius is between inner circle radius plus60and inner circle radius plus90, scan in the range of about90degrees of the image, detect the circle that with the maximum of the average gray transition, the circle is the iris outer circle. The algorithm effectively avoids and improves the drawbacks of concentric circle location, enhances the accuracy of the information obtain. Finally, normalize and enhance the located sample image,complete the pre-pretreatment phase.5. The meticulous evaluation based on both the geometry and the feature. Proposes two algorithms called effective area quality evaluate algorithm and JND evaluate algorithm. Firstly, morphology enhance the located iris image, prominent iris texture and eyelashes, secondly, conduct Sobel filter to the enhanced image to retain and prominent longitudinal eyelashes, and then remove salt and pepper noise using the regional growth algorithm to minimize the salt and pepper noise made to the detection of eyelashes, finally, using the curve fitting method to fitting the eyelid curve, then calculate the effective area factor using formula. I use the principle of compressed science that information redundancy is the guarantee of information security, calculate the information redundancy beyond the JND threshold, the redundancy is the evaluation result of the JND factor.6. For spatial location、effective regional and the JND three evaluation algorithms, I conduct both subjective and objective verifications on JLU_IRIS1JLU_IRIS2and CASIA4three iris databases, and compare the results of two validations, the result shows that the quality assessment algorithm result and subjective evaluation result consistent. Conduct Gabor feature extraction and recognition experiment to the qualified images of JLU_IRIS1、JLU_IRIS2and CASIA4, experimental result shows that the qualified images makes good experimental result, verify the effectiveness of the algorithm. To sum up, in this paper, researched and invented iris recognition device and collected the two generations of iris databases of laboratory, proposed the rough and meticulous image quality evaluation algorithm and non-concentric iris location algorithm. Verify the validity of proposed algorithm from the subjective and objective experiments and recognition rate experiment.