What is Crack Skeletonization?

Definition

Crack skeletonization is an image processing step that reduces a binary crack mask to a line one pixel wide, called the skeleton. A binary crack mask is an image where crack pixels have one value and all other pixels have another. A model produces it by crack segmentation .

Skeletonization is used for feature extraction and to represent the topology of an object. Topology here means how the parts of the crack connect. In crack work, the skeleton is a way to show the topology of the crack and to measure it. Fan et al. use a one-pixel skeleton for this purpose (Fan et al., 2020 ).

Methods

Thinning methods remove border pixels from the mask in repeated passes. They stop when no pixel can be removed without breaking the connectivity of the object.

• Zhang-Suen (1984) makes successive passes and removes border pixels. The scikit-image function skeletonize uses Zhang’s method for 2-D images. It is also the default of ximgproc::thinning in the OpenCV contrib modules. That function takes an 8-bit single-channel image with the crack at 255 (OpenCV contrib ).
• Guo-Hall (1989) is a parallel thinning algorithm with two sub-iterations in each pass. It preserves eight-connected components. The scikit-image function thin implements it, and the OpenCV ximgproc::thinning function offers it.
• Lee (1994) works on a 3 by 3 by 3 neighborhood. It is designed for 3-D images, and scikit-image shows it on a 2-D image only for illustration.
• The medial axis is the set of points that have more than one closest point on the object boundary. It can also return the distance to the boundary, which gives an estimate of the local width.

For a skeleton with fewer branches, scikit-image recommends its skeletonize function over the medial axis (scikit-image ). Fan et al. chose a medial-axis algorithm for their crack skeletons.

Spurs and branches

A rough mask edge gives the skeleton short side branches, called spurs. The non-smooth boundary of the crack affects the skeleton. Wang and Pang note that the skeleton then does not describe the crack precisely (Wang and Pang, 2025 ).

You can find the branches of a skeleton with a graph analysis. The skan library counts the neighbours of each skeleton pixel. A pixel with three or more branches is a junction, and a pixel where the skeleton ends is an endpoint. The library then gives one table row per branch, with its length and its type. The types are endpoint-to-endpoint, junction-to-endpoint, junction-to-junction and isolated cycle (skan documentation ). With such a table, you can find short end branches and decide whether to remove them.

Crack length and width from the skeleton

The length of a crack follows from its skeleton. Fan et al. sum the lengths of the skeleton elements, scaled by the calibrated displacement of the pixels. They assume no geometric distortion in the image. The calibration is the scale of the image in millimetres per pixel. See pixel to millimetre calibration .

The skeleton also gives a width estimate. The distance from each skeleton pixel to the background is an estimate of the local width. Other methods define the width as the shortest distance from the skeleton to the crack boundary. A rough boundary makes the skeleton imprecise, so the width estimate can also be imprecise. See crack width measurement for how pavement manuals use the width.

Crack length in PAVER

The PAVER manual for asphalt airfields asks for the length of longitudinal and transverse cracks, in linear feet or linear metres, together with their severity. If the severity changes along a crack, the manual asks you to record each portion separately. It asks for the area of alligator cracking, in square feet or square metres.

The severity of a crack depends on spalling, FOD potential and the condition of the filler. For a non-filled crack, the mean width also counts, with a limit of 6 mm. A skeleton gives a centreline length. It does not give any of these severity factors.

A skeleton is one way to get a centreline length from a crack mask. The manual does not name this or any other method.