Research And Implementation On Iris Recognition Technology

Bio metric technology has been widely used in the current world. And among these, Iris has become a mainstream means of biometrics due to its uniqueness, reliability and non-invasive. As we know, Iris recognition is very common in daily attendance, identity authentication, data encryption, e-commerce, e-government systems, etc. It comes with great convenience to our daily life and also receives a favorable review because of its easy-of-use, fast identification, and low recognition rate of errors. In contrast to other methods, such as face image, voice, and fingerprint and so on, iris recognition has a higher accuracy. Generally, Iris recognition consists of four parts:iris location, feature extraction, coding and recognition.In our thesis, we unfold to discuss mainly from the following aspects. First of all, we propose an iris location algorithm based on an improved Hough transform method in the iris localization part. More specifically, the algorithm is firstly utilized to establish the iris gray histogram image in order to analyze the gray threshold of the iris boundary, pupil image binarization. In this process, corrosion, expansion and region growing have been used to remove the noise for the purpose of obtaining the radius of the inner capsule. After that, we conduct ins localization according to the geometrical feature and gray feature of the human eye image. By narrowing the scope of the search, it can improve the speed as well as the accuracy.Secondly, we have adopted two iris feature extraction methods in the aspect of iris feature extraction and match. The first method is the improved Daugman's iris recognition method, in which we take advantage of sub-block-based texture analysis method for the filter phase encoding under Cartesian coordinate system To be more concrete, we firstly divide the normalized image into equal sub-blocks. For each sub-block, extraction and encoding operation are performed only on the center pixel, and then, the Hamming distance is utilized to match. The second method is the Gabor filtered method, in which multiple filters are built under the polar coordinates with the result of dividing the normalized image into equal sub-blocks. After that, the template convolution operation is conducted on the each sub-block. Take the mean and standard deviation of each filtered sub-block as the characteristic value, and the weighted Euclidean distance is applied to deal with the match operation Altogether, the above two methods both can obtain a high recognition performance and good robustness for real-time systemUltimately, we create a prototype system of iris recognition in order to facilitate the verification algorithm result. Its key functions include iris localization, feature extraction, match operation, result displayed, as well as some data analysis.In spite of the fact that our proposed method has gained good performance, there are still some problems to be solved. Considering the real time and complexity of the algorithm, our method only uses part of the region to do the feature extraction regardless of the extraction of eyelid and eyelash; as a result, it incurs the high rate of rejection of this algorithm. Besides, the robustness of the algorithm is not good for the images that seriously damaged by noise. However, the noise can be eliminated in our experiments to weaken the effect. On top of this, we adopt a general method of Hamming distance to design the classifier. Due to its simplicity, sometimes it cannot obtain satisfactory result. In our future work, the neural networks of Fisher linear discriminate methods can be utilized to improve the recognition rate.