| Summary: | Mathematical morphology is an established field of image processing first introduced as an application of set and lattice theories. Originally used to characterise particle distributions, mathematical morphology has gone on to be a core tool required for such important analysis methods as skeletonisation and the watershed transform. In this thesis, I introduce a selection of new image analysis techniques based on mathematical morphology. Utilising assumptions of shape, I propose a new approach for the enhancement of vessel-like objects in images: the bowler-hat transform. Built upon morphological operations, this approach is successful at challenges such as junctions and robust against noise. The bowler-hat transform is shown to give better results than competitor methods on challenging data such as retinal/fundus imagery. Building further on morphological operations, I introduce two novel methods for particle and blob detection. The first of which is developed in the context of colocalisation, a standard biological assay, and the second, which is based on Hilbert-Edge Detection And Ranging (HEDAR), with regard to nuclei detection and counting in fluorescent microscopy. These methods are shown to produce accurate and informative results for sub-pixel and supra-pixel object counting in complex and noisy biological scenarios. I propose a new approach for the automated extraction and measurement of object thickness for intricate and complicated vessels, such as brain vascular in medical images. This pipeline depends on two key technologies: semi-automated segmentation by advanced level-set methods and automatic thickness calculation based on morphological operations. This approach is validated and results demonstrating the broad range of challenges posed by these images and the possible limitations of this pipeline are shown. This thesis represents a significant contribution to the field of image processing using mathematical morphology and the methods within are transferable to a range of complex challenges present across biomedical image analysis.
|
|---|
