Application of Bayesian Graphical Models to Iris Recognition

Abstract Recognition of humans based on their biometric signatures is becoming of increasing importance because of its applications in access security, reliable identification for benefits distribution and homeland security among others. Popular biometric modalities include face images, fingerprints, iris images, palm prints, gait patterns, voice, etc. Iris images are of significant interest because of the excellent recognition rates they offer in controlled image acquisition conditions where the image quality is expected to be good enough to capture the details of the iris. However, in more realistic scenarios, iris images may not be of necessary quality because of image distortions and occlusions due to eyelids and eyelashes and the recognition rates provided by standard approaches may not be sufficient. In this chapter, we discuss how Bayesian graphical models can be used to achieve improved iris recognition in the presence of image impairments such as nonlinear deformations and occlusions. We illustrate the performance of these methods on sample iris image data sets.