UBHIC: Top-Down Semi-supervised Hierarchical Image Classification Algorithm

At present, most of hierarchical image classification methods are built under the premise that label of images are known. For unlabel image classification, the accuracy and efficiency can be improved. To solve this problem, this chapter proposes a top-down semi-supervised hierarchical image classification algorithm (UBHIC) with efficiency as the optimization objective. Firstly, the AP clustering is used for the batch of unlabel images, then the similarity between cluster centers and nodes is compared from top-down. Finally, the best classification path is determined by evaluating multiple classification paths. Experiments show that the UBHIC has better accuracy and efficiency than the classical methods.