MEASUREMENT OF TORSION FROM MULTITEMPORAL IMAGES OF THE EYE USING DIGITAL SIGNAL PROCESSING TECHNIQUES.

Two methods of measuring ocular torsion from digital images of the eyes were developed and tested. One method measures torsion from the translation of two landmarks using a rectilinear coordinate system. The second method measures torsion from the translation of two landmarks using a polar coordinate system. The center for the polar sampling is the center of the pupil. After thresholding and filtering the images, landmark translation is measured from the interpolated peak in the normalized cross correlation of the reference landmark with the image. The standard deviation of the measurement error for the first method using artificially rotated well-framed 256 multiplied by 256 multiplied by 8 single-eye images was 0. 042 degree in the absence of noise and 0. 061 degree for a noise-to-signal ratio of 0. 1. The corresponding measurement accuracies for the radial sampling method were 0. 019 degree and 0. 031 degree . The precision of the torsion measurement for high-quality experimental images was 0. 132 degree . The landmark tracking method on the rectilinear grid can be used when the rotation is within a plus or minus 5 degree range. The measurement technique using the polar sampling can be used when there is a single point which is moderately well known. Thus, digital signal processing techniques can be used to measure ocular torsion from images of the eye with a precision similar to the precision obtained by human photographic interpretation.