Abstract: When there are no sufficient feedback samples provided by Relevance feedback, supervised learning methods may suffer from the over-fitting in image retrieval. This paper proposed a novel neighborhood preserving regression algorithm which makes efficient use of unlabeled images. The algorithm selects the function which can minimize the empirical loss on the labeled images, thus, the function can respect both semantic and geometrical structures of the image database. The experimental results show that the algorithm is effective for image retrieval.

Key words: Manifold learning,Neighborhood preserving,Relevance feedback,Image retrieval