Quantitative Results

Table 1 shows quantitative evaluation. We evaluate the algorithms using the evaluation protocol developed in [1]. The algorithms are evaluated both in terms of boundary and region accuracy by comparing to ground truth. For all metrics (except variation of information), a higher value indicates better fit to ground truth. ODS and OIS are the best values of results of the algorithm tuned with respect to a threshold on the entire dataset (ODS) and each image individually (OIS), and the difference applies only to gPb and CB. Our method out-performs all methods on all metrics.

Table 1: Quantitative results for Brodatz synthetic dataset

Table 2: Quantitative results for real Texture dataset