3D volume localization using miniatures

The prediction of the position of a given volume sample in a full body atlas, also known as a volume localization, is a part of an initial stage of image retrieval in most of the dedicated CAD systems. In this paper we present two methods for volume localization, namely histogram matching and classifier regression. Since the histogram matching method ignores the spatial orientation, it is used when the orientation of the volume cubes are not the same. On the other hand the classifier regression is much faster and can be used as a quick estimation and as a tool to reduce the scope of the initial problem. Both presented methods were tested on a dataset with 3962 volumes of a human body atlas. The accuracy and the speed of execution was compared and is presented in this paper.