Fast and accurate near-duplicate image elimination for visual sensor networks

Abstract

Currently, a huge amount of visual data such as digital images and videos have been collected by visual sensor nodes, that is, camera nodes, and distributed on visual sensor networks. Among the visual data, there are a lot of near-duplicate images, which cause a serious waste of limited storage, computing, and transmission resources of visual sensor networks and a negative impact on users’ experience. Thus, near-duplicate image elimination is urgently demanded. This article proposes a fast and accurate near-duplicate elimination approach for visual sensor networks. First, a coarse-to-fine clustering method based on a combination of global feature and local feature is proposed to cluster near-duplicate images. Then in each near-duplicate group, we adopt PageRank algorithm to analyze the contextual relevance among images to select and reserve seed image and remove the others. The experimental results prove that the proposed approach achieves better performances in the aspects of both efficiency and accuracy compared with the state-of-the-art approaches.