A novel learning for image retrieval based on both keyword feature and instance feedback

To bridge the semantic gap between low-level visual features and high-level semantic concepts, this paper puts forward a novel feedback mechanism which is based on both instance and keyword features. In offline part, keyword space model is first constructed and updated using manifold ranking annotation; in online image retrieval and feedback part, the keywords which is return to user for labeling are obtained by Bayes algorithm; then by use of labeled keywords and images, the visual features are reweighted by mining the relationship between keyword and visual features; and finally the top n images are returned after learning image labels and then combining them by our ranking function. Our ranking function is flexible and can be adjusted easily. Experimental results on COREL 1000 images show our method improves image retrieval performance from all aspects.