Iris image annotation tool

Background

A simple tool in Matlab for annotating iris and eyelid boundaries, eye corners and irregular occlusions. Circular approximations of pupil and iris boundaries are assumed, and parabolic approximations for eyelid shapes are used. This tool was developed to conduct research on diurnal patterns in eyelid apperture and pupil size summarized in: A. Czajka, K. W. Bowyer and E. Ortiz, "Analysis of diurnal changes in pupil dilation and eyelid aperture," in IET Biometrics, vol. 7, no. 2, pp. 136-144, 3 2018

Instructions:

1. Run runSegmentation script (no input arguments are needed). You should see the first image read from data-raw subfolder.
2. Now you can choose one of four possibilities by pressing:

• p for pupil annotation,
• i for iris annotation,
• u for upper eyelid annotation,
• l for lower eyelid annotation.

3. Mark as many points on the boundary as you want. Press RETURN when you’re ready.
4. You can repeat each of the steps shown in 2. many times. For instance, if you have the upper eyelid annotated, but you are not happy with the result, just press u again and repeat the annotation.
5. Additionally, you can correct the resulting mask by choosing m (and + or - buttons to make the brush larger or smaller).
6. When all four elements are accurately annotated press s. The results will be saved in data-results folder.
7. Press RETURN for the next image.

All four steps (pupil, iris, upper eyelid and lower eyelid) use the least squares minimization to fit actual curves. Hence the only thing that must be done is selection of points (the more the better) on iris boundaries and eyelids. Good practice is to annotate at least five points when marking the pupil, iris and eyelids, and trying to place them equidistantly on the visible boundaries.

Extras

You can stop annotations and return to it later — Matlab saves the current state in savedState.mat file. So when you run runSegmentation again, it will start from the next non-annotated image in data-raw folder. If you want to start from scratch, simply delete the savedState.mat.

To check that all is correctly saved in files, you can run visualizeSegmentation.m.