Generic object registration using multiple hypotheses testing in partition trees

Using image processing techniques that partition an image into a number of homogenous regions, the authors´ work regards image analysis as a process to be conducted within a binary partition tree without regard to the image processing stage used to generate it. This work recognises through subjective tests that tree-based representations are amenable to segmentation based on simple descriptions. The method considers the novel application of relative topology of constituent regions within the tree representation and applies it to form registrations in the spatial and temporal domains. To achieve this requires the use of graph-matching techniques to circumvent the non-direct mapping between tree conjugates and derived models. Search strategies are successfully applied in the tree using multiple hypotheses testing in a Bayesian formulation framework, facilitating object registration from very simple models.