Study On Methods For Iris Image Prepocessing And Feature Coding

Human iris feature is suitable for identity authentication. However, the speed of current iris image noise detection and iris localization methods is relatively slow and the robustness is relatively poor. These factors have restricted the performance of the iris recognition system. Therefore, the study on iris recognition methods with high speed, high accuracy and good robustness has become an important issue.The methods for iris image noise detection, iris localization, iris feature coding and code registration were improved in this paper. The main works were as follows:(1) Two improved methods for iris image noise detection were introduced in this paper. The adaptive dual threshold detection method was used to detect eyelashes in iris images. Thresholds could be adjusted flexibly according to the different light conditions, so this method had better robustness than traditional ones. The detection method based on least square method was used to detect eyelid in this paper. Combined with the adaptive dual threshold method, this method could detect the eyelids shaded by eyelashes effectively by analyzing the gray scale characteristics of iris images.(2) Two improved methods for iris localization were proposed in this paper. The method based on connected domain detection was used to localize the iris inner boundary. The eyelashes and some other interference could be removed by selecting largest connected domain, so the robustness of this method was good. Also an Iris outer boundary localization method based on the delineation of region of interest was proposed in this paper. A rectangular template was used to delineate region of interest. Then the process of outer boundary localization would be conducted in ROI. The detecting region was smaller in this method than in traditional ones, so the accuracy and speed could be increased by using this method.(3) The iris feature coding method was improved in this paper. The response and the number of the Gabor filter which had maximum amplitude response in the filtering process would be recorded and the other filtering results would be abandoned. Therefore, the scale of feature code could be reduced obviously while texture information was hardly lost. A code registration method based on reference point was provided, which could improve registration efficiency in condition of precisely localization of eyelid.