Minimal path algorithms for the robust detection of linear features in gray images

Minimal paths are proposed as a powerful way to extract faint linear structures in noisy graylevel images. A global algorithm based on gray-weighted distance transforms is rst proposed and shown to sometimes be a compelling alternative to watersheds. A more general local minimal path method is also introduced, together with an algorithm for extracting minimal path cost at each pixel location. The proposed algorithms are very e cient, and examples of applications are used to demonstrate their robustness.