Scale-invariant vesselness filtering for simultaneous extraction of optimal medial and boundary paths in retinal images

In this paper we present a new multiscale vesselness filtering technique to simultaneously compute optimal medial axes and boundaries on fundus images of the retina. The scale-invariance of the vessel cross-sectional profile in the frequency domain is examined, and phase congruency implementing the scale-invariance is utilised for multiscale vesselness filtering. This allows the vessel ridge and boundary evidence to be preserved under single filtering settings. We have performed width measurement experiments on a dataset of fundus images, using an optimal medial axis skeletonisation scheme as a post-processing step, to compare the proposed technique with a generalised Gaussian profile modelling approach.