Shape information is utilized by numerous applications in computer
vision, scientific visualization and computer graphics. This paper
presents a novel algorithm for exploring and extracting 2D shape
information from greyscale images with no {a priori information about
the input data or the objects represented. Our method outputs a
high-level, shape-based feature breakdown of the information contained
in an image. More specifically, the algorithm gives the intensity
range spanned by each significant object as well as a versatile shape
model of the object that is directly useful for many applications.
The technique introduced is based on the computation of the shape
gradient, a numerical value for the difference in shape. In this
case, the difference in shape is caused by the change in threshold
value applied to the image. The use of this gradient allows us to
determine significant shape change events in the evolution of object
forms as the threshold varies. Our algorithm uses the Union of
Circles shape representation, which is flexible, has an effective
shape metric and allows multiscale processing. The extraction method
is stable and robust, and works in the presence of noise and other
artifacts. We show the results of applying this method to artificially
created images and to real medical images.
