Figure A2.

Effect of threshold value, $\theta$, on the binary background mask when the background is noisy. (a) RGB image, (b) red channel, and (c–l) binary background masks for different $\theta$. In (c) $\theta =0$, (d) $\theta =1$, (e) $\theta =5$, (f) $\theta =10$, (g) $\theta =15$, (h) $\theta =20$, (i) $\theta =25$, (j) $\theta =35$, (k) $\theta =55$, and (l) $\theta =65$. Depending on the intensity of the background noise, we need to decide the value of $\theta$. If we assign too small value to $\theta$, then some noisy background pixels may be considered as the foreground (see (c–i)). If we set too high value to $\theta$, then some area of the foreground will be considered as the background (see (k,l)). Source of fundus photograph: STARE data set and image file: im0240.ppm.