Image Mining by Data Compactness and Manifold Learning

Abstract

One important issue in image mining is how to analyze the compactness of image data and apply it to image mining. In this paper we study the class compactness and boundary compactness of image data, which are used in image classification and data confining. The data confining results in relevance graph, which is used in calculating the distances between images. Manifold learning techniques are applied in the computation of distances between images and manifolds of images. Image retrieval is based on these distances. Experiments are reported to show the effectiveness of our approach.