Application of genetic algorithms for the improvement of off-axis iris recognition.

摘要

Biometrics is an emerging topic amongst both governmental and commercial applications. While fingerprinting has been around for many years, several newer biometrics have come to the forefront of current research, including iris recognition. The iris is a very stable feature, hardly changing as people age. This makes irises a prime candidate to be used as an identifying biometric.;Many recognition algorithms currently exist and are very effective at distinguishing whether or not two iris images came from the same eye. However, a continuing confounding factor for iris recognition is the requirement that the eye image be taken orthogonally; in other words, the subject must be looking directly at the camera when the image is captured. While this is very achievable in controlled environments, perfectly orthogonal images become more unrealistic in real-world situations. However, developing different kernels for encoding an orthogonal enrollment image will generate multiple off-axis templates. These off-axis templates can then be used for matching when a verification image is taken where the subject's eye is not orthogonal to the plane of the camera.;Such kernels can be found through the use of a genetic algorithm. Genetic algorithms draw on the principles of evolution and genetics to find optimal solutions to the problem the algorithm is faced with. In this case, the optimal solution will be the kernel resulting in the best (smallest) fractional hamming distance when the orthogonal image is compared to a non-orthogonal image. This problem is ideal for a genetic algorithm because, at its most basic level, it is the minimization of a function. While this function has many local minima, genetic algorithms excel at finding the global minima amongst a sea of local minima, a common struggle for other conventional methods of minimizing functions.