Fig. 4From: Learning from label proportions for SAR image classificationClassification results for Tianjin and Rosenheim. a TerraSAR-X image of Tianjin. g TerraSAR-X image of Rosenheim. b, h Ground truth maps from pixel labeling (for accuracy calculation). c, i PL+SVM. d, j GL+SVM. e, k ∝SVM. f, l GL+LpcSVM with n k =0. Note that the result image of GL+LpcSVM with n k ∼N(0,0.052) is not listed here for it looks very similar to that of GL+LpcSVM with n k =0Back to article page