Registration of 3D Volumetric CT Images

Abstract: This master thesis aims to develop a system for analyzing transformation between two volumetric CT images. The volumetric image data we process is taken from a composite material. This composite material combines wood fibre and plastic and can be used to make for instance hockey sticks or furniture. Because of the wood fibre embedded in this composite material, it absorbs water and sometimes deforms. By observing volumetric images generated by micro computed tomography (micro-CT), we know that the organization of fibre embedded in this material is very complicated. This makes it difficult to predict the deformation on beforehand. In our study, we have seen rigid transformations, non-rigid transformations and even discontinuities transformations (cracks). For a pair of very small sub volumes, in dry and wet condition, we have found that the transformation can approximated by a rigid transformation combined with a scaling value. To find this transformation, our system includes two key phases. In the first phase, we extract feature points in dry and wet condition. In the second phase, we register the feature points derived from dry and wet condition. In the feature point extraction phase, we have adapted different methods, for instance the Scale- Invariant Feature Transform (SIFT) method is used to extract features. In the registration phase, we have tested three different registration algorithms. The first algorithm is based on concepts from Random Sample Consensus (RANSAC). The second algorithm is inspired from the Iterative Closest Point (ICP) method. The third method is a novel algorithm that we call Spatial Invariant Registration. In the report, we compare the different methods in the feature extraction phase and in the registration phase. Finally, we discuss how our system can be extended to give better results with better accuracy.