Feature extraction of volume data based on multi-scale representation

Abstract

In this paper, we present a novel algorithm that can detect and extract salient points as features in 3-D volume datasets. This algorithm extracts not only the locations of feature points, but also finds out the scales at which the features are most significant. It applies the scale-space theory by adaptively processing the input volume in a few discrete scales. The features points, as well as their scales, detected can be used for subsequent processing, such as content-based volume data authentication and content-based volume rendering.We have implemented the algorithm and tested its effectiveness on several volume datasets. Some optimization has also been done on time-consuming intermediate steps to speed up the feature detection process. Copyright © 2003 by the Association for Computing Machinery, Inc.