Fig. 3.
Leave-one-out performance by our method on the atlases for the first cross-validation experiment when different appearance and searching windows are used. Since the results produced using appearance window with r = 1 are significantly worse than using larger appearance windows, for better visualization, we separately plot the results using r = 1 on the left. Note that the best results produced by our method is about 1% better than those produced by LWGau in Fig. 2.