2008-IUPR-31Jan_1134.pdf

0.0 0.2 0.20 0.25 0.3 0.30 0.33 0.35 0.4 0.45 0.6 0.8 1.0 100 1002-1009 101 11.1 1107-1135 1470-1477 15.7 157-162 16.2 16.6 16.8 17.9 1816-1823 19.6 194 1978 199 1998 2000 2001 2003 2004 2005 2006 2007 208 21.7 2200 226-239 233-243 24.7 25.4 26.4 27.1 28.2 29.0 29.4 3.1 3.2 3.3 31.4 31.6 33.6 34.0 34.2 34.8 35.5 37.0 4.000 4.1 4.2 4.3 421-430 460-472 461 464 469-286 5.5 500 500-bin 588-dimensional 7.3 70.6 703 715 78-dimensional 79-88 791 8.2 86-91 911-920 961-386261 97.000 ACM ALIPR API Abstract Acknowledgements Acquiring Adrian Also Anal Ann Annotation Appl Approach Artificial Automated Automatic B.S BIC Barbara Barnard Bayesian Bernoulli Bleib Both Breuel Breuel1 Browsing Budapest CVPR Can Categories Center Challenge Chang Christian Circuits Classification Classifiers Closest Color Combining Compact Compared Comparing Computer Computerized Concept Concepts Conclusions Conf Content-based Corresponding Criterion Curtis Cybern DCT DFKI Daniel Data Database Department Descriptors Deselaers Despite Detection Diego Different Dimension Discriminative Duin During Duygulu EURASIP Each Effective Eng Engine Estimating Experimental Experiments Extraction Fair Feature Features Fei-Fei Feng Fergus Fig Figure Finally Finland First For Forsyth Frames Framework Freitas From Fusion Gaithersburg Gaussian Gemert Generalization German Geusebroek Given Google Graphics Group Guide H.-J Hammoud Hatef Hauptmann Histograms Hoad However Hsu ICCV IEEE Image ImageCLEF Img InViRe Indexing Inf Information Innovation Instead Int Intell Intelligence Intern Introduction J.-M J.-R Jordan Kaiserslautern Kennedy Key Keyframe Keysers Keysers2 Kittler Large-Scale Late Lavrenko Learn Learning Learns Left Lienhart Like Linguistic Lozano-Perez MAP MPEG-4 MPEG-7 MUSCLE Mach Man Manjunath Manmatha Matas Matching MediaMill Modalities Model Models Modern Mohr More Mori Most Motion Multimedia Multiple Multiple-Instance Naphade News Ney Note Object Obviously Ohm Online Ontology Oulu Our PF1 PF3 PFj Patches Pattern Pattern-based Perception Perona Pictures Pipeline Pipelines Portal Portals Practitioner Probabilistic Problem Proc RGB Real Real-Time Real-time Recognition References Related Relevance Reliable Res Research Retrieval Revver Rheinland-Pfalz Right S.-F S.L Sample San Santa Schulze Schulze2 Schwarz Science Search Second Section Selecting Semantic Seq Sequences Shots Signal Signatures Since Sivic Smeulders Smith Snoek Some Sports Stat Stiftung Survey Sys Syst System Systems T.C TRECVID TRECVID3 Table Tag Tagging Tamura Techn Technical Techniques Tesic Text Textural Texture The Thereby Therefore These This Thomas Though Time Training Trans Transition True Two Ulges Ulges1 Understanding University Using V.V Vasudevan Video Videos Vision Visual Vol Wang Washington Watching When While Words Work Worksh Workshop Worring XViD4 Y.-F YUV Yamada Yamawaki Yang Youtube Zhang Zisserman Zobel access accidents according achieve achieved achieves actions activities activity adapt adapted adapting adaptive adaptivity adapts additional addressed addressing adequate agrees algorithm allows analogy analysis annotated annotation annotations annual answer appear appearance application applied approach architecture area arg aspect aspects associated associates attracted automatic automatically autonomously available average avoid background bag-of-visual-words bag-of-words based baseline baselines basketball beach best better biking bins block blue boost boundary camera car case cat categories cats causes center certain characteristics characterized check chroma claim classifier clipped clips clustering clusters clutter codec coefficients collections color comb combination combinations combine combined combining come comes commercial common comparability comparable compare compared complement complete component components compressed concatenated concatenating concept concepts concerned concert confirm considerable contain containing contains content content-based contest context contribution corpus correctly cosine costly crash crashes created crucial curves dancing data database dataset datasets dealing dealt deformations delivers demo demo.iupr.org demonstrates depends described descriptor descriptors desert detect detected detection detects determined developed different difficult directly discrete discriminative diversity divide-and-conquer divided does domain download downloaded downloading duplicate duplicates e.g earlier early easily easy edited efficiency efficient eiffeltower eliminate emphasizes employed enormous entertainment entropy entry envision essential estimated estimates evaluate evaluation evaluations evidence evolved exact example excellent exist existing exists exp expanding expect experiment experimental experiments explanation explicit explicitly explosion extensive extract extracted extraction fact fair false far fashion feature feature-specific features fed final finally fit focus following follows form frame frame-level frames frequency fus fused fusing fusion future gave general generalize generalized generalizes generally generates given gives global golf good greater green ground groups helicopter helpful helps high high-level highest highlighted hiking histogram histograms hits home hosts hours hrs http i.e idea identify illustrate illustrated illustrates image images implicitly improve improved improvements improves in-depth including incomplete increases indexing indicate indicates indicating individual infer influence informatik.uni-kl.de information input instance integrated integration intensity interested interval interview introduce introduced investigate irrelevant iterative iupr.dfki.de joint k-means key keyframe keyframes keysers keyword keywords kind kinds knowledge labeled lack large large-scale late late-fusion latent layout leading leads learn learned learning learns left level like limitation linked list local locations log-linear long loss lot lots low low-frequency low-quality lowest lowlevel makes manual manually match matching material max maximum maximum-entropy mean measures method methods min minyY misleading mixture modalities model modeled modeling models motion motion-affine motorbike multiple named near near-duplicates near-perfect nearest need neglected negligible neighbor news noise novel number object objects observed obtain obtained obtaining occlusion occur occurrences occurs online optimal order organize outlined overall overview paper parallel parameter parameters partial particularly patch patch-based patches pattern peak performance performed performs pieces pipeline pipelines pixels plays plotted poorly portal portals poses positives possible posterior posteriors practical practice precisely precision presence present presented previous principle prior probability problem problems processing prod produces product production productions project projects properties propose proposes prototype provided providers providing publication publicly purpose purposes quality quantitative quantity questions race races ranging real-world reasons recall recall-precision recallprecision recognition red redundancy referred reg regions regular regularly related reliable rely remove removed replace represent representation representative represented representing represents requires research respect respected respectively restricted result resulting results retrieval return returns reveals revver revver.com right rightmost riot robust robustness role row rule sailing sample samples sampling scale scaling scene scenes schulze score scores search second secondlife seconds section segment selected semantic semantics separated series serve server set sets setup share sharing shot shots showing shows signature significantly similar simple single size skiing slightly small soccer sorted source sources space spatial specialized split standard standardization starting static statistical step steps stored strat strategies strong strongest strongly studied study styles successful successfully sum superior support supported supports suspiciously swimming systems t01v table tag tagged tagger tagging tagging-description.html tags talkshow tank targeted task techniques terms test tested testing text-based textual texture tile tiles time-consuming times tmb total trained training transform true truth try tuning two-dimensional type types ulges unclear unlabeled unmatched unreviewed upload uploaded uploading use used user-generated users uses using usually validation valuable van varies vector vectors video videoblog videos videotagging view visit visual visually vocabulary vote votes watching way weak weaker website weight weighted weights well-known word words work workshop world-wide worse wsum www-nlpir.nist.gov www.xvid.org years youtube youtube.com zigzag