The thesis project began by reviewing literature within the area of iris recognition in the visible wavelength to learn what had been done and which challenges that still needs to be solved, before a reliable recognition system can be created for mobile devices by using the visible wavelength. Based on the research articles it became obvious that segmentation methods for iris images in the visible wavelength had been well researched and had a comparable performance to that of iris images on Near-Infrared images. However recognition rates on visible wavelength images were poor so it seemed that the biggest challenge was to get enough discriminative information through feature extraction.
The primary focus of the thesis was to see if different types of feature extractors could be fused together and complement each other to improve recognition performance. For the project three different types of feature extractors were provided to be tested. The types of feature extractors were conventional feature extractors, key point feature extractors and general image descriptors. As a fourth feature extractor type a soft biometric using the iris colour was implemented.
To test the feature extractors and to test the fusion of feature extractors a framework was implemented. This framework would run the feature extraction, comparison, normalization, fusion and evaluate the recognition performance.
The early experiments showed that key point feature extractors and general image descriptors had a comparable performance to that of conventional feature extractors on iris images in the visible wavelength. They also showed that fusing different feature extractor types tended to have a better recognition rate compared to fusing feature extractors of the same type.
Throughout the experiments it was also discovered that fusing a feature extractor from each of the four different types the recognition rate could be improved significantly.