الفهرس 
Iris OverviewOnly 14 pages are availabe for public view
 |  | from | 188 |  |  |
| | | from | 188 | | |
Abstract
In this thesis, the performance of iris identification/verification systems are described and analyzed. Each system is broken in a similar way in terms of image processing procedures: image segmentation, image normalization, feature encoding, and pattern matching. CASIA version-1 database and a newly constructed database throughout this study using digital simulation techniques and voxel statistics were used to evaluate the performance of a number of preprocessing, encoding and template matching technique. Three encoding techniques are investigated: 2D-discrete cosine transform (DCT), 2D-discrete wavelet transform (DWT), and 1D-Log-Gabor filter. The encoded template consists of a huge number of coefficients to be used as an input to the classifier. Therefore, dimensionality reduction is a necessary step. Three techniques are used for this purpose: DCT zigzag scan, DWT thresholding, and Principal component analysis (PCA). A multi-layer perception (MLP) artificial neural network (ANN) is implemented to perform iris identification. The results have shown a correct classification rate (CCR) of 99.9% using DWT and PCA. For the verification part, two minimum distance classifiers are used: Hamming Distance (HD), and Weighted Hamming Distance (WHD). Two different types of weighted map were used: (i) Personalized best bit map (PBBM), and (ii) Personalized weight map (PWM). As a result of the comparative study it has been found that the HD and DWT encoding gives an optimum value of equal error rate (EER) value of 1.531×10-4. In an attempt to minimize the iris verification systems, data fusion at the matching score level for the two highest performance techniques was performed. Data fusion is evaluated using three different methods: (i) simple sum (SS); (ii) Minimum Score (MIS), and (iii) Maximum Score (MAS). It has been found that the fusion method using MAS gives an optimum EER value of 0.8×10-4. It is concluded from the thorough investigation performed that iris encoding using DWT technique gives the highest performance. For iris identification, the adopted technique of MLP-ANN using DWT encoding gives promising results. On the other hand, the minimum distance classifier using HD and DWT gives the minimum error rates. Finally, the maximum score (MAS) data fusion method plays an important role in minimizing the error rates in iris verification.