Speaker / Predavač: Dr Attila Tanacs

Title / Naslov: Estimation of affine deformations of binary objects in 2D and 3D

Abstract / Abstrakt:

Image registration is one of the main tasks of image processing, its goal is to find the geometric correspondence between images. Shape or object matching is an important task of registration. Matching in this case consists of two steps: First, an arbitrary segmentation step provides the shapes or objects and then these are registered. This solution is especially viable when the image intensities undergo strong nonlinear deformations that are hard to model, e.g., in case of X-ray imaging. Domokos et.al. proposed an extension to the parametric estimation method of Francos et.al. to deal with affine matching of crisp shapes in 2D. These parametric estimation methods have the advantage of providing an accurate and computationally simple solution, avoiding both further correspondence problems as well as the need for optimization. The solution can be determined by solving a polynomial system of equations. In my talk I will cover the extensions of this 2D binary method we proposed recently. One direction is the investigation of the case when the segmentation method is capable of producing fuzzy object descriptions instead of a crisp binary representation. We also extended this method to handle 3D binary objects, available in either crisp or fuzzy forms. Via synthetic test of thousands of image pairs, we observed the effect of the number of quantization levels of the fuzzy membership function to the precision of image registration and we compared the results with the binary case. In 3D, we also investigated cases when the image pairs are degraded. Preliminary results are available for real X-ray and CT images.

Date / Datum: 23.06.2010. Time / Vreme :15:00