skeletonization

Skeletonization is a process in image processing and computer vision that produces a skeletal representation of a shape. From a binary image, the object is reduced to a thin set of connected lines that preserves the topology and the general structure of the original region. In theory, this relates to the medial axis transform, the locus of points equidistant to multiple boundary points; in practice, discrete skeletons are generated by thinning or by extracting ridge lines from distance transforms.

There are several approaches to skeletonization. Thinning algorithms, such as Zhang–Suen and Guo–Hall, iteratively remove boundary pixels while maintaining connectivity and without breaking the object into separate parts. Distance-transform-based methods identify the ridges of the distance map to form a skeleton. In three dimensions, voxel-based thinning yields medial surfaces rather than lines. Other methods use morphological operations or optimization criteria to produce topology-preserving skeletons that better reflect branching structure.

Skeletons are not unique: different algorithms or preprocessing choices can produce different results for the same shape. They are also sensitive to noise and small perturbations, which can create spurious branches. Preprocessing steps such as smoothing, noise reduction, or feature pruning are common, as is post-processing to remove insignificant branches and to improve interpretability.

Applications include shape analysis, pattern recognition, handwriting and character recognition, fingerprint matching, and medical imaging for extracting vascular or airway trees. In computer graphics and robotics, skeletons support shape representation, pose estimation, and path planning by providing concise abstractions of complex shapes.