In the image below one can visually recognize 7-8 slightly curved lines, which sometimes are crossing each other. The challenge thus is to implement/test multiple efficient curved line detection/enhancement algorithms on this type of image, and compare the results between the algorithms. Some current popular approaches are generalized Hough transform and windowed radon transform, but there may also be other suitable approaches. A recognized problem with some of these algorithms are though their sensitivity to noise.

NB: The number of lines in the image is unknown, and the order of magnitude for the line length should be a parameter to the process

As this image is just one 2D cross-section from a 3D image (same concept as in 3D ultrasound images), a possible extension may also be to extend these methods from "curved line detection in 2D images" to "curved plane detection in 3D images".

Supervisor: Leonardo Rocha