Please use this identifier to cite or link to this item:
|Title||Enhancing Iris Recognition|
In this thesis, we propose three techniques to increase the iris recognition robustness and accuracy. First, we propose a new segmentation algorithm to handle iris images were captured on less constrained conditions. This algorithm reduces the error percentage while there are types of noise, such as iris obstructions and specular reflection. The proposed algorithm starts by determining the expected region of iris using K-means clustering algorithm, then circular Hough transform is used to localize iris boundary. After that, some proposed algorithms will be applied to detect and isolate noise regions. Second, a study of the effect of the pupil dilation on iris recognition system is performed, in order to show that the pupil dilation degrades iris template and affects the performance of recognition systems. Therefore, a limit of pupil dilation degree is determined. If the degree of pupil dilation exceeds this limit, the iris code will be affected or some of its information will be discarded. This limit can be used to avoid detrimental pupil dilation. Finally, we analyze the iris code bits to determine the consistent and inconsistent bits, and we compare between the inner and outer regions to find which region contains more inconsistent bits. In our experiments, we use three free public iris' databases (UBIRIS v1, CASIA v3 and CASIA v4). Each of our algorithms were implemented in the MATLAB 7.0 software. The environment where the experiments are performed in is Compaq PC, Core 2 Due Intel Pentium Processor (2.00 GHz), 1GB RAM and Windows 7 operating system. Experiments on UBIRIS v1 show that the accuracy of our segmentation algorithm is 98.76%, which significantly improves the performance of iris recognition, and the average execution time of our segmentation algorithm equals 1.68 ms, which is less than the execution time of the other algorithms. Experiments on CASIA v3 show that the effect of pupil dilation can be minimized significantly by excluding irises which have a pupil dilation degrees more than the estimated limit of dilation degree. Finally, experiments on CASIA v4 show the existence of inconsistent bits in the iris code, and also show that the bits of inner regions are more stable than the bits of outer regions. Results show that this work significantly improve the performance of iris recognition, and make it possible to be applied in wide environments, especially in non-cooperative environments.
|Publisher||the islamic university|
|Files in this item|