PrePrint: Large-Scale Image Phylogeny: Tracing Back Image Ancestry Relationships
Similar to organisms that evolve in biology, a doc- ument can change slightly overtime while each new version is also able to generate other versions. Multimedia Phylogeny investigates the history and evolutionary process of digital objects which includes finding the causal and ancestry document relation- ships, source of modifications and the order and transformations that originally created the set of near duplicates. Multimedia Phylogeny has direct applications in security, forensics, and infor- mation retrieval. In this paper, we explore the phylogeny problem for near-duplicate images in large-scale scenarios, and present solutions that have straightforward extension to other media such as videos. Experiments with about two million test cases (with synthetic and real data) show that our methods automatically build image phylogeny trees from partial information about the near-duplicates, improving the efficiency and effectiveness of the whole process, and represent a step-forward determining causal relationships of digital images overtime.
PrePrint: Partial-Duplicate Image Retrieval via Saliency-guided Visually Matching
In this paper, we propose a novel partial-duplicate image retrieval scheme based on saliency-guided visually matching, where the localization of duplicate is done simultaneously. The image is abstracted by the Visually Salient and Rich Regions (VSRR), which are of both high visual saliency and rich visual content. To obtain the compact representation with sparsity at the region level, VSRR is represented by sparse code with group lasso. Further, we construct a relative saliency ordering constraint to refine the retrieval result, which captures the robust relative saliency layout among interest points of the VSRR. Collaborating with this constraint, we propose an efficient algorithm to embed it into the index system to speedup the retrieval. Comparison experiments with state-of-the-art methods on five image databases show the efficiency and effectiveness of our approach.
PrePrint: Video Copy Detection and Localization with Scalable Cascading of Complementary Detectors and Multi-scale Sequence Matching
For video copy detection, it has been recognized that none of any single audio-visual feature, or single detector based on several features, can work well for all transformations. In this article, we propose a novel video copy detection and localization approach with scalable cascading of complemen-tary detectors and multi-scale sequence matching. In this cascade framework, a soft threshold learning algorithm is utilized to estimate the optimal decision thresholds for de-tectors, and a multi-scale sequence matching method is em-ployed to precisely locate copies through 2D Hough transform and multi-granularities similarity evaluation. Excellent per-formance on the TRECVID-CBCD 2011 benchmark dataset shows the effectiveness and efficiency of our approach.
PrePrint: A Multimedia Semantic Retrieval Mobile System Based On Hidden Coherent Feature Groups
We propose a multimedia semantic retrieval system based on Hidden Coherent Feature Groups (HCFGs) to support multimedia semantic retrieval on mobile applications. The system is able to capture the correlation between features and partition the original feature set into HCFGs, which have strong intra-group correlation while maintaining low inter-correlation. Specifically, a feature similarity matrix is built using correlation information between feature pairs, and the Affinity Propagation algorithm is applied to identify the HCFGs, each of which is modeled by one or more classification methods. A novel, multi-model fusion scheme is presented to effectively fuse the multi-model results and generate the final ranked retrieval results. In addition, to incorporate user interaction for effective retrieval, the proposed system also features a user feedback mechanism to refine the retrieval results. Experimental results demonstrate the effectiveness of the proposed framework.
PrePrint: Querying Blobs in Vehicular Networks
In this paper we study querying binary large objects such as video and voice clips in a network of vehicles communicating wirelessly. We develop a set of query processing strategies and compare them along three dimensions, namely push versus pull, whether or not communication infrastructure is utilized, and whether metadata dissemination is separated from blob dissemination. We analyze these strategies theoretically and experimentally s in terms of answer throughput and communication overhead.