The purpose of this thesis is to explore methods intended for the assessment of image quality in the domain of ocular images captured in visible light. The thesis uses already established methods from the fields of face image quality assessment and general image quality assessment. The main objective of the thesis is to apply these algorithms to sets of ocular images in order to eliminate the worst quality examples. The MOBIUS dataset is used. It uses quality labels for images that are of poor quality from a biometric usability perspective. Standard procedures for testing and evaluating methods are used. Since the assessment of the quality of an image depends on the given task, part of the thesis is devoted to reviewing biometric recognition based on ocular images. In the experimental part of the thesis, adapted facial methods perform better than general methods. However, their success depends largely on the recognition model used. General methods prevailed in less well-trained models.
|
