Laplacian Eigenmap for Image Retrieval

Xiaofei He

2002-01-01

Abstract:Dimensionality reduction has received much attention in image retrieval. Even though the dimension of image feature vectors is normally very high, the embedded dimension is much lower. Before we utilize any image retrieval technique, it is beneficial to first perform dimensionality reduction. Principal component analysis (PCA) is one of the popular methods for dimensionality reduction, and is shown to be optimal when the underlying structure of data points is linear. However, it fails to discover the nonlinear structure if the data lie on a low-dimensional manifold embedded in a high-dimensional feature space. In this paper, we propose a geometrically motivated linear dimensionality reduction algorithm, called Locality Preserving Projection (LPP), for representing high dimensional data lying on a low dimensional manifold, which has locality preserving properties. The algorithm provides a computationally efficient approach to linear dimensionality reduction. Moreover, a novel relevance feedback scheme on image manifold is proposed. Experimental results on real-world image collections have shown the effectiveness and robustness of our proposed approaches.