Iris Recognition Based On Independent Component Analysis And Support Vector Machines

In recent years, biometric identification technology has been achieved a significant development and the market share is also increasing every year. Among the biometric technologies, the iris recognition technology has been received increasing attention, and it is considered as one of the most promising biometric technologies because of its high degree of accuracy, uniqueness, stability and non-invasive.Iris recognition process is divided into image preprocessing, feature extraction and matching identification. To solve the feature extraction problem and recognition problem in the process of iris recognition, an algorithm is proposed in this paper, which adopts ICA (Independent Component Analysis) to extract iris feature and SVM (Support Vector Machines) for classification. The normalization and image enhancement is used to process the iris position which was located in the eye images. ICA is used to extract statistical independent feature which is a higher-order statistical methods, emphasizing on the independence between the various components, and it can describe the essential characteristics of the image. SVM is used to classify the iris feature, and it is a learning system using a linear function in high-dimensional feature space, by some nonlinear function mapping the input vector to a high-dimensional feature space. In various fields and applications, the performance of SVM is better than most other learning systems.In the process of image preprocessing, binarization is used to locate iris inner boundary. Canny operator is used for the edge detection and hough transform is used to locate iris external boundary. Polar transformation is used to scale the iris image to a uniform size. Finally, the histogram equalization method is used to enhance the iris image. In the process of feature extraction, FastICA algorithm is used to improve the extraction rate, and PCA algorithm is used to reduce the computational amount. In the process of matching recognition, we select the best parameters to construct SVM model. Experimental results show that the algorithm proposed in this thesis is feasible and suitable for iris recognition.