72 6. DETECTION OF THE ONH USING USING PHASE PORTRAITS
distance = 3 pixels and overlap = 0.87 with the intensity condition and using the Hough transform; the method based on phase portraits also successfully detected the ONH with distance = 47.4 pixels, as shown in part (b) of the same figure. The image im0010 from the STARE dataset is shown in Figure 6.9 (c); in this case, the phase portrait method has detected the center of the ONH more accurately than the Hough transform, as shown in Figure 6.9 (d). The Hough transform was misled by white scar and vessel curvature in the image.
Some of the images in the STARE dataset are out of focus. In Figure 6.10 (a), the result of detection for image im0035 using the Hough transform is shown. Because the image is blurry, it is difficult to detect the edges in the images; hence, the Hough transform could not succeed in detecting an appropriate circle to fit the ONH. Part (b) of the same figure shows the result of detection using phase portraits, which is a successful detection with distance = 26 pixels. The advantage of the method based on phase portraits is that it can detect the vascular pattern existing in the image even when the image is blurry.
(a) |
(b) |
Figure 6.10: (a) The results of detection of the ONH for the STARE image im0035 using the Hough transform. The black dashed circle corresponds to the highest local maximum in the Hough space that also meets the condition based on 90% of the reference intensity. The cyan contour in solid line is the contour of the ONH marked by the ophthalmologist. Distance = 454 pixels, overlap = 0 with the intensity condition. (b) The result of detection using phase portraits. The black square is the first peak detected in the node map that also meets the condition based on 50% of the reference intensity. The black triangle is the center of the ONH marked by the ophthalmologist. Distance = 26 pixels.
6.5REMARKS
A procedure based on phase portraits to locate the ONH was described in detail in this chapter; no similar method has been reported in any of the published works on the detection of the ONH. The
6.5. REMARKS 73
blood vessels of the retina were detected using Gabor filters and phase portrait modeling was applied to the orientation field to detect points of convergence of the vessels. The method was evaluated by using the distance from the detected center of the ONH to that marked independently by an ophthalmologist. With the inclusion of a step for intensity-based selection of the peaks in the node map, a successful detection rate of 100% was obtained with the 40 images in the DRIVE dataset.