Relevance Feedback Algorithm Based on Semi-supervised Linear Neighborhood Propagation

A feedback semi-supervised linear neighborhood propagation method(FSLNP) is proposed.FSLNP method can not only preserve the positive and negative constraints but also preserve the local and global relevance structure information of the whole graph.With both labeled and unlabeled images in relevance feedbacks,a better structure for relevance representation among images is found to reveal the semantic structure.Experimental results show that FSLNP can effectively improve retrieval accuracy,and after long term learning,an optimal relevance graph space can be obtained.