A multiresolution approach to iris synthesis

Abstract: Databases of human iris images are created and distributed for the purposes of testing iris identification algorithms. For logistical and privacy reasons, these databases are often too small to fulfill their potential applications. In this work we develop a novel multiresolution approach to augment iris image databases. First, using a multiresolution obtained from reverse subdivision we decompose the example iris images into a set of lower resolution components. The components are a complete representation of the original image and consist of a low resolution approximation and a set of characteristics. To generate synthetic iris images we combine a set of components chosen from the original images. To ensure a unique, yet realistic, iris image each component of a synthetic image is chosen from different iris images. We quantitatively validate our approach by employing a classical iris recognition algorithm to compare our synthetic images with those that were used to create them. The results demonstrate that our approach is effective at augmenting iris image databases with iris images that are unique, yet exhibit both visually and statistically realistic characteristics. [ABSTRACT FROM AUTHOR]