Manifold Based Semi-Supervised Learning for Content-Based Image Retrieval

We propose a manifold regularization based semi-supervised learning for image ret-rieval (MRBS2L) in this paper. Given query image,MRBS2L at firstly calculates the similarities between the query image and images in database with L1 distance and ranks,and constructs the weighted graph. Then it selects some similar images from ranked results by the user to form relevance feedback image set,which will be used to act as labeled samples and the remainder images are used as unlabelled data,and then trains classification function with laplacian manifold regularization based semi-supervised learning. Finally,it classifies and ranks the images with trained classification function and returns retrieval results. The experimental results on Corel image database demonstrate the effectiveness of the proposed algorithm.