In this thesis, two vessel enhancement techniques are introduced in order to produce more accurate vesselness measures and vessel direction estimations that are less subject to local intensity abnormalities....[ Read more ]

In this thesis, two vessel enhancement techniques are introduced in order to produce more accurate vesselness measures and vessel direction estimations that are less subject to local intensity abnormalities.

Vessel enhancement using local line integrals (VE-LLI) is first proposed which embeds a vessel model by regarding a vessel segment as a straight line. It uses the second order information of the local line integrals. Quantities similar to eigenvalues and eigenvectors of the Hessian matrix are produced, which can be substituted in any existing Hessian-based enhancement methods.

A variational optimization (VO) is then introduced to improve qualities of vesselness measures. A more generic curve model is embedded to model vessels and a variational framework is used to locate the possible vessel segment to which a point belongs. Features on the vessel segment are then used to generate an optimized vesselness measure.

Experiments are conduced on 2D retinal images. It is experimentally shown that VE-LLI produces more accurate vessel direction estimations than conventional Hessian-based methods. Both VE-LLI and VO can give more accurate vesselness measures.