Analysing 3D images stacks and extracting curvilinear features

This thesis is concerned with the innovation of processes for the automatic extraction and explicit representation of three dimensional curvilinear structures from three dimensional image stacks.The motivation for the work came from neurophysiologists wishing to extract 3-D generalised cylinder model representations of neuron structures using stacks of laser scanning confocal microscope images taken at different levels through brain tissue. Previous methods were either manual or, at best, semi-automatic, and the process of extraction of the features from the images and construction of the full 3-D models was particularly labour intensive.Research into more automatic image analysis procedures for the extraction and model building is presented in this thesis. The process involves a number of stages including initial processing of the image, identification of voxels with a high probability of being on the centre lines of the tree structure using a novel 3-D thinning algorithm, extraction of a continuous centre line representation of the tree using a novel combined 3-D minimum spanning tree and minimum cost path algorithm and finally construction of the generalised cylinder model around the centre line tree representation.