Abstract
2-D/3-D image registration is a technique to estimate 3-D pose/position of 3-D model from 2-D image.
It has been used in various fields including industrial and medical fields; especially, it has been applied for
analyzing knee joint kinematics. Roughly, matching score used in 2-D/3-D image registration can be classified
into two approaches: landmark based method and intensity based method. In case of landmark based method,
the matching score is defined as a distance between landmarks in the images. Although the method is a simple
and high speed processing, the registration accuracy is depend on extraction accuracy of landmarks. In
contrast, in case of intensity based method, the matching score is defined as differences of intensity between the
images for all pixels. The method has an advantage that it is unnecessary to extract any landmarks, however, it
has a tendency toward the falling into local solutions. In this paper, we introduce attended/unattended ROI
(region of interests) into matching score. By representing the ROIs with fuzzy spatial maps, we can decrease
the dependency of registration accuracy on defining ROIs. In addition, by employing ROIs, the proposed
method improves the registration accuracy in comparison with intensity-based method. To validate the
performance of the proposed method, it has been applied to image matching between 3-D knee bone model
reconstructed from multidetector-row CT and 2-D digital radiographic image. The experimental results
showed that the proposed method estimated knee angles with a root mean squared error (RMSE) of 0.4 deg that
was superior to conventional intensity-based method.
