Immersive Visualization of Visual Data Using Nonnegative Matrix Factorization

Abstract

Over the last two decades, dimension reduction for visualization has gained a high amount of attention in visual data mining where the data is represented by high-dimensional features. Basically, this approach leads to an unbalanced and occluded distribution of visual data in display space, giving rise to difficulties in browsing the data. In this paper we propose an approach for the visualization of image collections in such a way as (1) images are not occluded by each other, and the provided space is used as much as possible; (2) the similar images are positioned close together; (3) an overview of data is feasible. To fulfill these requirements, we propose to use regularized Nonnegative Matrix Factorization (NMF) controlled by parameters to reduce the dimensionality of data. Experiments performed on optical and radar images confirm the flexibility of proposed method in visualizing large-scale visual data. Finally, an immersive 3D virtual environment is suggested, to visualize the images, to allow the user to navigate and explore the data.