Parameter-orientated segmentation algorithm evaluation

Quantitative testing of segmentation algorithms implies rigorous testing against ground-truth segmentations. Though under-reported in the literature, the performance of a segmentation algorithm depends on the choice of input parameters across core, pre- and post-processing stages. The paper highlights the importance of post-processing parameters when the figure of merit is the Berkeley F-measure. It also shows that the search of a parameter space with a genetic algorithm is not only accelerated through the inclusion of a time factor in the cost function but the relative importance of different parameters is highlighted.