Detecting curves with unknown endpoints and arbitrary topology using minimal paths.

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• State-of-the-art minimal path techniques require significant user input, such as endpoints or curve length, especially for branching curves.
• Existing methods struggle with complex topologies like multiple branches or closed cycles without extensive prior knowledge.

Purpose of the Study:

• To develop a novel minimal path-based algorithm for more general curve extraction in images.
• To reduce user input requirements compared to existing minimal path algorithms.
• To enable detection of both open and closed curves, including complex topologies.

Main Methods:

• A novel minimal path-based algorithm is presented.
• The algorithm requires only a single arbitrary input point on the desired curve.
• It handles complex topologies including multiple branch points and closed cycles without a priori specification.

Main Results:

• The algorithm successfully extracts open and closed curves, including complex topologies.
• It requires significantly less user input (a single arbitrary point) compared to existing methods.
• Quantitative evaluation on diverse images (pavement cracks, catheter, retinal) shows accurate curve extraction.

Conclusions:

• The novel algorithm accurately detects curve-like objects with reduced user interaction and prior knowledge.
• It offers a more flexible and general approach to curve extraction in image processing.
• Future work includes application to other 2D objects and extension to 3D curve detection.