Curvilinear Structure Enhancement in Biomedical Images

Abstract

Curvilinear structures can appear in many different areas and at a variety of scales. They can be axons and dendrites in the brain, blood vessels in the fundus, streets, rivers or fractures in buildings, and others. So, it is essential to study curvilinear structures in many fields such as neuroscience, biology, and cartography regarding image processing.
Image processing is an important field for the help to aid in biomedical imaging especially the diagnosing the disease. Image enhancement is the early step of image analysis.

In this thesis, I focus on the research, development, implementation, and validation of 2D and 3D curvilinear structure enhancement methods, recently established. The proposed methods are based on phase congruency, mathematical morphology, and tensor representation concepts.

First, I have introduced a 3D contrast independent phase congruency-based enhancement approach. The obtained results demonstrate the proposed approach is robust against the contrast variations in 3D biomedical images.

Second, I have proposed a new mathematical morphology-based approach called the bowler-hat transform. In this approach, I have combined the mathematical morphology with a local tensor representation of curvilinear structures in images.
The bowler-hat transform is shown to give better results than comparison methods on challenging data such as retinal/fundus images. The bowler-hat transform is shown to give better results than comparison methods on challenging data such as retinal/fundus images. Especially the proposed method is quite successful while enhancing of curvilinear structures at junctions.

Finally, I have extended the bowler-hat approach to the 3D version to prove the applicability, reliability, and ability of it in 3D.