Figure 14
From: Contextual Classification of Image Patches with Latent Aspect Models
Image patch classification examples at TEER. Results provided by the following: first column, K-means empirical distribution; second column, aspect model 1; third column, aspect model 2. The first three rows illustrate the case of a correctly identified marked natural image context by aspect model 2, resulting in a more accurate patch classification as compared to aspect model 1 and empirical distribution. The fourth row shows a correctly identified marked man-made image context by aspect model 2, with an improved number of correctly classified points. The last row shows the confusion in patch classification, when the context is not correctly identified (in this case, overestimated) by aspect model 2.