Iris recognition based on image enhancement using K-SVD dictionary learning

Biometrics is a general term used for computer-controlled, automated and smart systems developed to identify the user by recognizing the physical and behavioral characteristics. In this thesis, an identity verification system implemented with iris recognition method, which is a sub-branch of biometric systems. Iris recognition systems consist of successive stages. Images taken from Casia Iris database were used in the scope of the thesis. In the method, the iris picture is taken from the database and blurred by applying the median filter. Then the pupil boundaries are determined. In the next step, the iris region is determined by the Circular Hough Transform Method and the iris region is separated from the rest of the image. Later, a transition was made from Cartesian coordinate to polar coordinate. In this way, the iris image is transformed into a rectangular form in matrix format. All images converted to polar coordinates were resized as 256x64 pixels and all images were standardized. After going through the preprocessing stages, the images brought to standard dimensions in two-dimensional matrix format were normalized and then the distinctive features of the iris were clarified by enriching the images with the K-Singular Value Decomposition (K-SVD) dictionary learning method over the separated iris region. By this means, comparison is made on iris images with the feature vectors obtained by using Structural Similarity Analysis (SSIM), Hamming Distance (HD), Euclidian Distance (ED) and Mean Squared Error (MSE) and verification and diagnosis systems are established. Segmentation errors, iris recognition success percentages and error percentages were examined in comparison with the literature, and the proposed method presented competitive successes with the methods in the literature. The method developed within the scope of the thesis using Euclidean Distance has shown 99.86% success. While the false acceptance rate was 0%, the false rejection rate was 0.0014.