A novel approach implementation of eyelid detection in biometric applications

Iris recognition systems commonly use iris images taken under the constrained conditions, where a high accuracy rate is guaranteed. However, sometimes these constrained conditions in image acquisition are inconvenient for users and restrict the scope of the applications. If the unconstrained iris images are used, the performance of iris recognition will be degraded considerably due to noise factors such as eyelids, eyelashes and reflection. Generally, eyelids are detected before normalization of image. The eyelids can be approximated with parabolic arcs; so parabolic detection is done. But parabola detection in Cartesian coordinates is time consuming as angle of rotation of eye is also to be considered. To make eyelid detection faster, the parabolic detection should be implemented on rotation invariant normalized iris template. Also, eyelashes occluding the iris region are noise factors that degrade the performance of iris recognition. If they are incorrectly classified as the iris region, the false iris pattern information will increase, decreasing the recognition rate. Thus, reliable detection of eyelashes is required to improve the performance of iris recognition. Noises due to reflection influence the features of both noise regions and their neighboring regions, which will result in poor recognition performance. Here, a novel approach for eyelid detection is proposed and eyelashes detection technique has been modified to achieve better performance. Also, conventional methods for noise removal are implemented and compared with the proposed techniques. This electronic document represents implementation of these approaches for validation of this work.