| With the improvement in intelligent environment requirements and strengthening in security awareness of human beings, the security authentication technology which is based on biological-features has been developed quickly. Iris is not only easy to carry and acquire, but also is safe and stable. So iris recognition technology based on biological-features has become the research hotspot currently. As one of the critical stages of iris recognition, iris acquisition has been the concerned problem of domestic and overseas scholars.Taking the productive of iris recognition as the starting point, this article puts much stress on the research about some key problems of iris acquisition. Firstly, this paper analyses the factors which impact the quality of the iris images during the acquisition, and then we design and realize a new iris-capture-device according to the impact factors. Secondly, we put much stress on the algorithms of auto-acquisition, texture analysis on iris area, quality evaluation based on human visual characteristics. Finally, we used the proposed evaluation algorithm and three different recognition algorithms to evaluate the random samples of JLU-IRIS1, CASIA3 and JLU-IRIS2. Large amount of quality analysis and recognition comparasion of experimental results show that:among all the samples from three iris databases, the ones of JLU-IRIS2 are more suitable for human visual characteristics and recognition. The key problems of iris acquisition studied are as follows:1. It analyzes the factors affecting the image quality during the process of acquisition, designs and implements a new generation iris capture devices. Firstly the article introduces the principle and process of iris acquisition and then analyzes the factors affecting the iris images. Secondly, we choose proper optical lens, image sensor, and design the light source in round shape, user interface to eliminate the noise which affect the image quality. After finishing the iris capture device, we use the DirectShow to capture and save videos about human iris. The design and implementation of iris capture device are very important. They provide all kinds of iris samples for the further study of automatic acquisition and evaluation.2. It proposes the algorithm about automatic acquisition to improve the intelligence of acquisition. According the drawbacks of the traditional iris acquisition, this paper proposes algorithms about automatic acquisition to ease the iris aquisition, so that we can capture the iris samples without assistant. The automatic acquisition of iris images includes three phases:the module of iris image capture, the inspection module of iris, and the image quality evaluation of iris. Firstly the capturing module extracts an iris image from the camera or the video, and then uses the inspection module to check the existence of iris of the image. If the condition is satisfied, we use image quality evaluation based on Grayscale co-occurrence matrix to analyze the texture of the iris. In this way, we can discard or obtain an iris sample by a series of evaluation.3. The algorithms about iris inspection. In the frame of iris automatic acquisition, this paper proposes two different localization algorithms:contour following without light and the vote algorithm with light. If there are no light pots in the iris image, contour following algorithm is selected to locate the internal contour of iris. The processes are as follows: First of all, get the possible edges of the iris and the pupile by binary processing on the iris image. Secondly, use dilation algorithm to offset the noise around the pupil. Finally, locating the inner contour of the iris by contour following algorithm and in this way we can get the real area of the iris in the image. If there are light pots in the image, we firstly use the corner detection algorithm to locate the light spots, and then calculate the possible center location of all the spots; Secondly, we scan the iris image in the horizontal and vertical direction from the possible center of the light spots to find the four gray mutation points on the edge of pupil and iris. As we all known, three different points but not in a line can draw a circle, so in the last we used the vote algorithm to obtain the accurate corrdinate of the iris center.4. Iris image quality evaluation based on human visual characteristics. After using automatic acquisition to establish the JLU-IRIS2 database, firstly the paper analyses the factors affecting the human visual characteristics, and then we use the algorithm based on human visual characteristics to evaluate the JLU-IRIS1, CASIA3, JLU-IRIS2 samples by space position, unified brightness, the noise rate, and the grain clarity of the iris area. A large number of experiment data shows that the samples from the JLU-IRIS2 are more suitable for human visual characteristic than the ones from JLU-IRIS1 and CASIA3.From the result, It can infer that the iris capture device designed by this paper is more validate and reasonable.5. Iris image analyses based on recognition algorithm. In order to prove the samples of JLU-IRIS2 are more validate, this paper constructs six different ideal samples and nine different real samples by random sampling from JLU-IRIS1, CASIA3, JLU-IRIS2. Each group of experiment samples are composed by ten individuals. We used three different feature extracted and classified methods to calculate the recognition rate, which are 2-D gabor and weighted euclidean distance, wavelet zero-crossing and the similarity of structure, ICA/PCA and cosine distance. When the false recognition rate is equal to false reject rate no matter under the ideal or the real environment, the recognition rate of JLU-IRIS2 is much higher than that of other iris database among the same kind of experiments.For example, when we use the samples of JLU-IRIS2, the recognition rate is 97% and 92% by 2-D Gabor in ideal and real condition. It can be proved the samples of JLU-IRIS2 are more suitable for iris recognition; the design of the iris capture device is more validate.In conclusion, the paper puts much stress on the design and implementation of iris capture devices, automatic acquisition without assistant, iris image evaluation based on Grayscale co-occurrence matrix, image analysis based on human visual characteristic.In the end, we use image evaluation algorithm based on human visual characteristic and three different recognition methods which are 2-D Gabor, Wavelet zero-crossing, ICA/PCA to analyse the ideal samples and real samples from JLU-IRIS1, CASIA3, JLU-IRIS2. The results not only from evaluation but also from recognition expriments show that the iris images from JLU-IRIS2 are more validate. The article lays the foundation for further study of iris recognition. |
PDF Full Text Request