Image enhancement and unified segmentation approach in motion-blurred image for iris recognition
Lack of user cooperation, poor quality cameras and surrounding conditions during image acquisition are factors responsible for motion-blurred iris images. These factors caused the images to have a high false rejection rate of iris recognition due to variations in an iris pattern caused by the occurr...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2013
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Lack of user cooperation, poor quality cameras and surrounding conditions during image acquisition are factors responsible for motion-blurred iris images. These factors caused the images to have a high false rejection rate of iris recognition due to variations in an iris pattern caused by the occurrence of noise and varying degrees of contrast in the darkly pigmented area of the eye. This research seeks to address these problems by developing of two methods to improve the distinctiveness of iris patterns and to localize the iris in order to minimize false rejection and acceptance rates. The first method is a combination of Multiscale Retinex and Homomorphic filtering techniques which have been used to cope with illumination variations and eradicate shadows to produce refined iris patterns. Multiscale Retinex improves the contrast of an image after the image has been cleared from any form of interference and Homomorphic filtering eliminates shadows in a motion-blurred image. To address the problem of imprecise localization of the boundary between the iris and the pupil, a second method or Unified Segmentation Approach was used to consider the iris location, outer and inner boundary localizations with contrast enhancement and noise detection. Besides that, the approach would increase the intensity level of the darkly pigmented area to enable the inner boundary of a pupil and iris to be correctly localized. To evaluate the performance of the proposed approach, UBIRIS.v2 dataset comprising motion-blurred images at a distance was used. Experimental results showed that, the approach exhibited a localization accuracy of 99.0%. Ultimately, the distinctiveness and localization problems were addressed when the use of both methods achieved a recognition accuracy of 99.89% with minimal false rejection and false acceptance rates |
|---|