Purchase this article with an account.
Nathan Luck, Scott Liddle, Stephen Grenon; A Method for Automatically Positioning a Camera for Tear Film Imaging. Invest. Ophthalmol. Vis. Sci. 2014;55(13):4854.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop software that works in conjunction with the LipiView camera and three-axis motorized stage that is capable of identifying the tear film and positioning the camera such that the pupil is in a specified position and the tear film is in focus.
First, the pupil is detected by taking an image from the camera and finding the largest region of contiguous pixels within a threshold of the minimum brightness occurring in the image that is between 3,000 and 20,000 pixels in size. The centroid of the pupil is calculated, and the motors are moved to bring the pupil closer to the center of the image. The pupil is again found and its centroid calculated. If the centroid is within 7 pixels vertically and horizontally of the center of the image, the autoposition algorithm is complete. Otherwise, iterations of moving the motors and finding the location of the pupil continue until the pupil is within 7 pixels of the center or three iterations have elapsed. Autofocus is performed using contrast detection in a 400 x 200 pixel rectangle located below the pupil. This region contains the shadow of the eyelashes, which provides a significant contrast differential in comparison to the illuminated region. The focus motor scans while images are simultaneously captured, then returns to the location at which the greatest contrast was observed. Finally, an anterior offset of 1.5 mm is applied because the image of the eyelashes reflection appears posterior to the surface of the tear film. The targeting software was used to position the camera prior to image capture for 20 eyes (10 individual participants). The final distance between the pupil centroid and the center of the image was captured. The final focus position was compared to the optimal focus position as determined by a human observer, and the error was recorded.
The mean horizontal position error was 20.0 pixels (range 1 to 40) while the vertical position error was 13.8 pixels (range 2 to 32). The mean focus error was 0.5 mm (range 0 to 1.5).
The maximum position error was 40 pixels, less than the acceptable error of 50 pixels. The depth of field of the system is 5 mm, greater than the maximum error of the focus axis, indicating that the tear film was in focus for every image. The results demonstrate that the autofocus and autoposition algorithms successfully and reliably center the pupil in the image and bring the tear film into focus.
This PDF is available to Subscribers Only