Accelerating the Optimization of a Segmentation Ensemble using Image Pyramids

In this paper, we propose an image pyramid-based noisy energy function evaluation method for the local search technique simulated annealing. The method is primarily designed for the optimization of image segmentation algorithms, and it maintains solution quality with significantly reduced time requirement. The strategy to select the proper image pyramid levels during the search is theoretically determined via adapting results regarding evaluation in simulated annealing based on imprecise measurements. As a demonstrative application, we perform parameter-optimization of a segmentation ensemble dedicated to the extraction of bone structures from CT images.