headed by Prof. Dr. Alexandru C. Telea
Skeletons, also called medial axes, are compact, powerful shape descriptors, used in computer vision, image processing, and 3D shape processing, compression, and matching. We compute robust skeletons of complex, noisy 2D and 3D shapes using a new technique based on a so-called importance measure. Different definitions of the importance measure allow us to obtain different types of shape simplifications and segmentations, as well as a multiscale shape representation framework.
Skeletons have a number of interesting properties which make them effective and efficient in a number of applications:
- "low-dimensional:" skeletons have a lower dimension than the shapes they describe. A 2D shape has a 1D (curve) skeleton.
- "hierarchic:" skeletons have a graph-like structure which corresponds to the part-whole structure of the shapes they describe
- "perceptually intuitive:" skeletons are centered in the shape they describe, which matches our intuition about the 'essence' of that shape
- "multiscale:" skeleton branches encode the size (area, volume) of the shape boundary features
- "homotopic:" skeletons encode the topology of the shapes they describe
Skeletons can be used in many exciting practical applications:
- part-type and patch-type segmentation of 2D and 3D shapes
- edge-preserving smoothing of shapes
- shape matching, tracking and compression
- shape analysis in computer vision
- path planning in robotics applications