Figure 17

Segmentation processes of a 700 × 605-pixel non-pathological retinal fundus image obtained from STARE dataset. Input image (a), manual segmentation (first observer) (b), mask image (c), image edge obtained using the proposed phase map (d), image structure obtained using the phase map (e), binary image obtained from (d) using Otsu thresholding[43](f), skeletonized version of (f) after eliminating outliers (g), dilated version of (g) (h), segmented image using the proposed method after 61 iterations (i).