Fig. 4: Demonstration of model discrimination power.

a Cross-validation results for models fit using k = 3, 5, and 9 images of each surface. The cross-validation was done to provide guidance about the number of images and the choice of DF (ν). There were no false positives or false negatives in this analysis, so it did not provide any conclusive results. b Rates of false positive and false negative classifications (in %) using models trained on the four different sets of surfaces and tested on consecutive subsets of those images for k = 2, 3, …, 9. A full summary of the results is provided in Supplementary Table 1. c Distributions of the log-odds of a match using models trained on the four different sets of surfaces and tested on subsets of k consecutive images for k = 2, 3, …, 9, for a model with ν = 10. d Rates of false positive classifications (in %) using models trained on the four different sets of surfaces using only the images with at most 50% overlap and tested on subsets of k consecutive images for k = 2, 3, 4, 5 and using only the 3 non-overlapping images and tested on subsets of k consecutive images for k = 2, 3. A full summary of the results is in Supplementary Tables 2 and 3.