概要

We present an image matching method between endoscopic images for reconstructing 3-D shape of an intestine. In this case, image matching is very difficult task because there are few explicit interest points in these images. In addition, by using usual interest point detector, many of the detected points are caused by lighting conditions and halations and those are not on the folds of the intestine. In this paper, we first propose a method for emphasizing intestine folds which are suited for reconstructing the intestine structure. We also propose a feature descriptor for the detected points and a matching method between the descriptors. We show the effectiveness of our method using real image examples.