Random Sampling SVM Based Soft Query Expansion for Image Retrieval

This paper focuses on the problem that relevance feedback schemes based on support vector machines (RF-SVM) always give a poor performance when the numbers of positive/negative feedback examples are strongly asymmetric. To address this issue, we propose a random sampling SVM based query expansion for relevance feedback learning. Firstly, we adopt a random sampling method to construct multiple asymmetric bagging SVM classifiers (hard or binary SVM each) and aggregate them to form a compound SVM classifier by classifier committee voting. Subsequently, the voting results are combined with query expansion to sort the final feedback ranking results. The proposed method can effectively restrain the negative effect of the sample asymmetry. Thus it provides a good error-tolerant ability to training data. Experimental results on a subset of COREL image database demonstrate the effectiveness and robustness of the proposed approach.