Results

The retinal photographs tested were taken using a fundal camera with a $30^\circ$ field of view (Kowa FX-50R, Kowa, Tokyo, Japan) using Ilford FP4 (125 ASA) photographic film (Ilford Imaging UK Ltd., Knutsford, England). Photographic negatives were digitised using a Nikon 35 $mm$ film scanner (LS-1000, Nikon, Tokyo, Japan). Digitised images were $2400 \times 2800$ pixels in size. For the analysis of complete fundus images, these were reduced to 600 x 700 pixels by applying a Gaussian pyramid low-pass filter [23].

Figure 10 shows the segmentation process described in the previous section, (a) negative red-free gray scale image, (b) $\gamma$, (c) $\kappa$ and (d) the segmented image. For the reduced size images, an interval of $1 \le s \le 12$ pixels was used.

Figure 10: The segmentation method. (a) Original, (b) $\gamma$, (c) $\kappa$ and (d) segmented. Image size 600 x 700 pixels.

Figure 11 shows the segmented images corresponding to the original images shown in Figure 1, red-free and fluorescein.

Figure 11: Segmented images from Figure 1, respectively, (a) red-free, (b) fluorescein. Image size 600 x 700 pixels.

It can be seen that despite the much poorer contrast, the red-free image segmentation is nearly as good as that of the fluorescein image. We also note that the segmentation is relatively insensitive to the wide variations in intensity that are inherent in these images. In the fluorescein image, for example, the detail in the segmented image in the much brighter region just beside the darker optic disk in the left centre of the image is nearly the same as in the higher contrast portions of the image.