Figure 12
From: Contextual Classification of Image Patches with Latent Aspect Models
Comparison of the true positive rate versus false positive rate curves for all patch classification methods, obtained by varying the likelihood ratio threshold T : (a) performance of the baseline methods for different numbers of K-means clusters and GMM components, (b) best baseline results compared to the aspect models with a vocabulary size of Nv = 1000 patches and NA = 20aspects.