Fig. 5From: Bioimage classification with subcategory discriminant transform of high dimensional visual descriptorsClassification results comparing different descriptors. Results are obtained by classifying the descriptors using (1) SVM only without SDT and (2) our method that applies SDT before SVM classification. The features tested include IFV, L/H/C/G (denoting the combination of LBP, HOG, CENTRIST and GIST), features generated using the VGG-VD model pretrained on ImageNet, and features generated using a CNN model trained on each dataset. The mean accuracy and standard error are shownBack to article page