Image Segmentation and Shape Representation Using Deformable Surfaces 1

We present a technique for constructing shape representation from images using free-form deformable surfaces. An object is modelled as a closed surface that is deformed subject to attractive fields generated by input data points and features. Segmentation is achieved by initializing the surface at some location in the scene and by letting it deforms itself until it fits the input data. Surface deformation is controlled by applying standard theory of mechanical systems. The algorithm is general in that it makes few assumptions on the type of features, the nature of the data and the type of objects. The algorithm does not assume that a clean segmentation of the input data is available. It correctly recovers shape even in the presence of spurious features and data points. We present results in a wide range of applications: reconstruction of smooth isolated objects such as human faces, reconstruction of structured objects such as polyhedra, and segmentation of complex scenes with mutually occluding objects. We have tested the algorithm using data from different sensors including grey-coding and laser range finders and video cameras, using one or several images. We briefly describe the application of this representation to the problem of computing stable grasp position for the manipulation of unstructured objects.