Efficient Image Segmentation Using Weighted Pseudo-Elastica

摘要

We introduce a new segmentation method based on second-order energies. Compared to the related works it has the significantly lower computational complexity O(N log N). The increased efficiency is achieved by integrating curvature approximation into a new bidirectional search scheme. Some heuristics are applied in the algorithm at the cost of exact energy minimisation. Our novel pseudo-elastica core algorithm is then incorporated into a user-guided segmentation scheme which represents a generalisation of classic first-order path-based schemes to second-order energies while maintaining the same low complexity. Our results suggest that, compared to first-order approaches, it scores similar or better results and usually requires considerably less user-input. As opposed to a recently introduced efficient second-order scheme, both closed contours and open contours with fixed endpoints can be computed with our technique.