In this paper, we will describe a new 2-D–D registration algorithm, intended to solve a 3-D localization problem.
On the contrary of usual algorithms, classified as matching-based or similarity-based, our morphology-based algorithm is able to register images without the need for landmark extraction, contour segmentation, or quantitative grey value computation. These conditions turn out to be necessary in the field of bronchoscopic image registration.
Our algorithm allows us to register images with local minima located through the shape of their neighbourhood. This algorithm is largely initial data tolerant, because no quantitative quality function is involved and minimized. Instead, our method is based on “daemons”, that push locally one image in the registration direction.
We present various experiments in order to study algorithm convergence. For that, we introduce the use of dynamical system diagrams. The results presented here demonstrate the method's efficiency and rapidity (less than 1 second for 100 × 100 images on a conventional workstation).
We present applications to computer-assisted bronchoscopy: this registration method can help in analysing bronchoscopic camera movements inside bronchial tree, and in achieving bronchoscopic and CT scan 2-D—D data fusion.