The goal of this project is to develop a robust registration technique that adapts accurately and efficiently to the dynamic features captured from individual data. An elegant solution to it is to rely on statistical atlases generated from a set of previously registered dynamic data. This project is divided into the following four tasks:
Task 1: Dynamic data acquisition and extraction of dynamic features
The first part of this task deals with the choosing and acquiring the dynamic data of the objects to be studied. We plan to employ marker based methods in order to accurately locate and track material points and minimize potential inaccuracies. In the second part we will develop data analysis methods, with a specific focus on the extraction of dynamic features. Methods like strain analysis will be employed, assuming dense, approximately regular spatial sampling of the recovered 3D model is available at each time phase. Dynamic features such as principle directions and magnitude of the deformation will be identified on the surface, based on the analysis results.
Task 2: Registration using dynamic data
A reliable registration method will be developed, based on the dynamic features extracted from Task 1. For a given sets of motion vector fields that represent each individual’s principal directions and degree of deformations, the objective is now to systematically estimate the correspondence so that surface points with similar degree and orientation of internal strain can be placed in correspondence. ‘Feature space’, i.e. information that will be used for matching, as well as the similarity metric, will be constructed on the motion vector space.
Task 3: Statistical model construction
The object of this task is to build an atlas, which is a summary of a population data. Average shape and variations of the observed data will be captured by the atlas which will be used as a template for further process (atlas-based registration and parameterization in Task 4). This involves first building a mathematical model for compact representation of the population data, and second, building a joint probabilistic map of identity shapes and deformation fields.
Task 4: Revision of registration with the statistical model
This task involves the enhancement of the registration technique developed in Task 2, armed with the statistical model constructed in Task 3. The registration will be formulated as finding an optimization problem: find optimal parameters of the statistical model that best matches the given dynamic data. Partial mapping will be experimented to rigorously validate the method.