This is part of a series of blogs diving into the technical aspects of Veridium’s distributed data model, biometrics, and computer vision research and development by our chief biometric scientist Asem Othman.
The term ocular, derived from its Latin root oculus, broadly refers to the anatomical regions and structures related to the eye. Ocular biometrics refers to the recognition (identification or verification) of individuals using the information offered by ocular modalities. The public perception of ocular biometrics has been largely limited to iris recognition. However, research suggests that several other ocular modalities could serve as biometric indicators, such as retina (retinal vascular pattern inside the eye globe), conjunctival vasculature (the blood vessel patterns observed in the sclera, or white, of the eye), and the ocular region of the face (the eye and the areas around it, including the iris, conjunctiva (white of the eye), eyelashes and the eyelids (if the eye is closed), eyebrows, and moles/scars around the eye).
Of all the ocular biometric traits, however, iris is considered to be the most reliable. This is based on its uniqueness, performance, and circumvention properties.
Iris as a Biometric
Irises exhibit an extraordinary amount of textural details that are believed to be different between individuals and between different eyes of the same individual. The texture of an iris can be described as a multilayered, tangled mesh-like structure that imparts a highly complex texture to its surface. Compared with fingerprints, iris image data acquisition is usually non-invasive. Thus, the iris has become one of the most reliable biometric traits for identity verification and recognition.
However, capturing the texture of irises using a traditional camera is challenging. Iris images are typically acquired using sensors that operate in the near-infrared (NIR) spectrum. The wavelength of the illuminating sources range between 700-900nm. The usage of NIR spectrum for iris recognition provides two critical benefits. First, it is observed that the effect of melanin, a color-inducing compound, is negligible at longer wavelengths. Using NIR spectrum ensures that the acquired image reveals information related to the texture of the iris, rather than its pigmentation. Second, compared to the visible spectrum, the texture of dark-colored irises can be well observed using NIR.
Iris Representation and Matching
Patterns of the left iris of an individual are assumed to be different from those of the right iris in the context of iris recognition systems. Therefore, during the enrollment phase of an iris-based authentication system, the operator/user must indicate the iris from which eye is enrolled. Then, during recognition, the system must capture the iris from the same eye image so that it can be successfully matched with the corresponding one in the database. Most of the iris-based authentication systems analyze the iris texture for recognition.
The texture of the iris is formed by many interlacing minute characteristics, such as pigment spots, stripes, furrows, and crypts, that are embedded on a stroma. Based on these features, the recognition takes place. But, prior to recognition, the iris region must first be localized and segmented from an image of the eye. Errors in the segmentation step will lead to poor performance due to the inclusion of noise (eyelashes, sclera, pupil, eyelids, and specular reflection) in the image.
Several iris representation techniques have been proposed in the literature and the matching is mainly based on the method of representation – the method used for encoding the iris texture. Thus, most existing techniques for iris recognition can be divided into two major classes. The first class represents the iris texture using filters or transforms, the second class of methods seeks to capture local and macro iris feature, such as freckles, crypts, and furrows, in the spatial domain.
Daugman’s phase encoding technique, which falls under the first class, is the most common and promising among the different iris recognition techniques.
First, a camera acquires an image of an eye and the iris annular region is segmented. Next, the annular iris is geometrically normalized – unwrapped from raw image coordinates to polar coordinates. A texture filter is applied to the normalized iris image, and the filter responses are quantized into a binary representation (iris code). Finally, the comparison between two iris codes is done by computing the fractional HD as a dissimilarity measure.