Tihendusalgoritmeid

Tihendusalgoritmeid, or thinning algorithms, are a class of image processing techniques used to reduce binary shapes to their skeletal representations while preserving topological structure. The primary goal is to extract a one-pixel wide contour that maintains the connectivity and endpoints of the original object, facilitating pattern recognition, shape analysis, and feature extraction tasks. By iteratively removing peripheral pixels, thinning algorithms produce a medial axis that approximates the underlying geometry of the input figure.

Several families of thinning algorithms exist, notably the iterative hit‑or–miss approach, Zhang‑Suen two-iteration method, and medial axis transform. The Zhang‑Suen technique applies two sub-iterations per pass, evaluating each pixel against a set of 12 templates to decide removal, while ensuring no isolated pixels are eliminated. Alternative methods rely on morphological operations such as erosion and opening, followed by subtraction to obtain the skeletal core. Computational complexity varies: simple hit‑or–miss schemes run in linear time relative to the number of pixels, whereas more elaborate medial transform methods may involve additional distance calculations and edge tracking.

Practical applications span handwritten character recognition, fingerprint analysis, road network extraction, and biological vessel mapping. Thinning often precedes further analysis stages, such as feature extraction or classification, by simplifying the shape while retaining critical spatial relationships. Common drawbacks include sensitivity to noise; pre‑processing steps like morphological closing or Gaussian smoothing are frequently employed to mitigate spurious skeleton branches. Continuous research seeks to improve robustness, reduce artefacts, and adapt thinning operations for multi‑scale or color images, expanding their usability in diverse computer vision pipelines.