Detecting eyes in digital images

Detecting eyes in digital images taken in a variety of natural lighting conditions is not a straightforward task. Great difficulties are experienced when one attempts to employ traditional methods such as the Radon or Hough transforms. They appear to be not robust enough to perform the detection task in images that include blur, speckles, or a reflection in the pupil or iris areas. Also, in some practical applications these techniques are not computationally efficient. We propose a simple quadruple axis spatial-domain-based symmetry indicator for detecting circularly symmetric objects in digital images. The proposed detector is an inherent part of a larger system for estimating 2D characteristics of the human eye. The efficacy of the proposed technique versus a frequency domain energy detector has been evaluated using artificial images. It has been shown that the proposed method outperforms the energy-based detector for moderate levels of noise. Also, the method appears to be more robust than its frequency domain counterpart and correctly detects the eye in a wide variety of images. An application of the detector to clinical assessment and patient diagnosis in optometry is provided.