Robust Iris Segmentation by combining adaptive fuzzy c-means, geodesic active contour and circular hough transform algorithms for detection of fixational eye movements

Purpose: Most of our visual perception occurs during fixation. Fixational eye movements (microsaccades, drift & tremor) occur at different frequencies and amplitudes. Frequencies of 0.3 to 5 Hz have been reported for microsaccades. Yarbus reported 95-97% of fixation time can be atttributed to drift and up to frequency of 100 Hz has been noted for tremor in different studies. The device and algorithm should be able to detect such movements to paint an accurate picture of the eye movements involved in visual perception. For a study like evaluation of fixational eye movements with such high frequency and low amplitude, accuracy is the most important consideration.Methods: We present an algorithm for detecting fixational eye movements. The algorithm includes initial eye clustering by modified fuzzy C-means. Then, geodesic active contour is applied to detect palpebral fissure. Finally, circular hough transform extracts the iris.Results: We show that the proposed algorithm results in higher precision than routine methods. Applicability of the algorithm for various ocular conditions is shown.Conclusion: The results are in favor of using active contour algorithms as an intermediate step rather than the last for