Fig. 2

Illustration of our method design. During the training process, the high-dimensional visual descriptors of the training images are extracted, and the SDT models are learned. The visual descriptors of the training images are then transformed using the learned SDT models, and a linear-kernel SVM is learned from the transformed descriptors. During the testing process, the high-dimensional visual descriptor of the test image is extracted, and then transformed using the learned SDT model. The transformed descriptor is then classified using the learned SVM to produce the classification output