An Applicable Multiple-Level Classification Based on Image Semanti Ccorrelation from User

In this paper, we propose a multiple-level image classification; the multiple-level image semantics classifier is constructed according to the hierarchical semantics tree from user. Image features are derived from the training set using prior knowledge, and the hierarchical classifier is constructed according to the class correlation measure. This measure considers the relation of the classifiers between different levels, and between the classifiers in the same level. The unlabelled pictures can be classified from the top down and assigned to corresponding class and semantic labels. In our experiment, meta-classifier is a binary SVM classifier; the hierarchical classifier is build by selecting meta-classifiers with the best combining performance. The experiment result shows that the hierarchical classifier is not effective even though every meta-classifier perform very well. Meanwhile, it proves our method is applicable