Iris Recognition AlgorIthms Based On DCT Transform And Its Parallelization Study

As the emergence of biological identification, iris recognition technique has received increasing attention in recent years. One of the most promising recognition techniques is the one based on the human iris due to its high reliability for personal identification. It has desirable properties such as uniqueness, stability, noninvasiveness and living-object character over the traditional method.Iris recognition system is made up by iris image preprocessing, iris extraction and classification. In this thesis, based on the development of current iris research results, the iris localization, image preprocessing, feature extraction and iris code classification are researched. Some new algorithm improvements have been proposed and given some experiment results.The pre-process of the Iris localization include image localization, image denoising, normalization and enhancement. In iris procedure of localization, in order to overcome the defect of huge computation of traditional Hough transformation and to avoid blind search, so as to realize fast and precise registration, this thesis proposed a new algorithm of Iris registration which is based on Mathematical Morphology. The threshold of adaptive binarizing and Binary Morphology are used to get the centre andradius of inner circle of the iris. And the Hough transformation and image edge detection based on Mathematical Morphology are used to extract that of outer circle of the iris. Thus we can separate the iris from the eye. The method is proposed for this algorithm to accelerate the localization speed further more. Radon transformation is used to isolate the eyelids. For eliminating eyelashes, the threshold technique is used. In order to eliminate the effect caused by the pan and zoom deformation of iris image, this thesis adopt a kind of iris unitary method based line segment extraction, this method expand the annular iris region into 64×256 block matrix. Compared with the traditional iris unitary method, the theory of this method is simple and directly, and do not need to classified in to 5 parts to analyze and gets high efficiency. For improving the effect of iris recognition and reducing the influence of unsymmetrical illumination, the iris image is transformed by histogram equalization.According to the multi resolution analysis features of DWT and the advantages of Energizing Phase and decorrelation of DCT, a algorithm feature extraction based on the principle of the Discrete Wavelet Transform(DWT) and the Discrete Cosine Transform (DCT) isproposed in this thesis , which is implemented as follows :first, we reduce the data dimension by DWT, then decorrelation by DCT, extracting the feature vector by ZIGZAG, finally. The experiment result shows that: the algorithm in this approach is easy and fast. The thesis analyzed feature extraction algorithm has internal parallelism and realized the parallelization algorithm based on MPI.In the feature matching algorithm, if we adopt the Euclidean distance method directly, some specify features of unitary iris will be ignored, and bring side effect to the efficiency of iris recognition. With the Euclidean distance used as the weighting coefficients, the iris nomralization image are divided 3 parts, each part is matched separately and then 3 matching results are got. Different weighting coefficients are used to multiply the matching result of each part.Good recognition results are achieved after 15000 recognition experiments have been done in the CASIA iris database(version 1.0).