Image Retrieval Based on Manifold Learning and Incorporate Clustering

This paper presents a novel unsupervised learning framework named image retrieval based on manifold learning and incorporate clustering. The dimensionality of image descriptors used in image retrieval applications is quite high. Given a query image, our algorithm first makes use of manifold learning (LPP) for dimensionality reduction and manifold ranking algorithm to explore the relationship among all the data points in the feature space, and then measures relevance between the query and all the images in the database accordingly. Then we use the similarities among target images for improving the performance of the image retrieval systems by cluster-based retrieval of images by unsupervised learning. Our algorithm retrieves image clusters as retrieval results by applying K-means clustering algorithm to a collection of images collected by manifold ranking algorithm. Experimental results on a general-purpose image database show that our algorithm attains a significant improvement over existing systems.