Abstract
The relevance feedback process in content-based image retrieval is generally treated as a classification problem, where the small sample size learning difficulty and the fast response requirement make it difficult for most classifiers to achieve a satisfying performance. In this paper, we incorporate the stochastic classifier ensemble method as a solution to alleviate this problem. In particular, the random subspace method is adopted in relevance feedback process to both improve the retrieval accuracy and decrease the processing time. Experimental results on 5,000 images demonstrate the effectiveness of the proposed method.
