Feature-Based Retinal Image Registration Using D-Saddle Feature

Figure 3

Multiresolution DoG pyramid for D-Saddle. The size of the image in subsequent octave is downsampled by half to detect feature points on vessels of varying sizes.