Detail publikačního výsledku

Brno University of Technology at TRECVid 2010 SIN, CCD

HRADIŠ, M.; BERAN, V.; ŘEZNÍČEK, I.; HEROUT, A.; BAŘINA, D.; VLČEK, A.; ZEMČÍK, P.

Originální název

Brno University of Technology at TRECVid 2010 SIN, CCD

Anglický název

Brno University of Technology at TRECVid 2010 SIN, CCD

Druh

Stať ve sborníku v databázi WoS či Scopus

Originální abstrakt

This paper describes our approach to semantic indexing and content-based copy detection which was used for TRECVID 2010 evaluation.

Semantic indexing

1.  Theruns differ in the types of visual features used. All runs use severalbag-of-word representations fed to separate linear SVMs and the SVMs were fusedby logistic regression.

  • F_A_Brno_resource_4: Only single best visual features (on the trainingset) are used - dense image sampling with rgb-SIFT.
  • F_A_Brno_basic_3: This run uses dense sampling and Harris-Laplace detector incombination with SIFT and rgb-sift descriptors.
  • F_A_Brno_color_2: This run extends F_A_Brno_basic_3 by adding densesampling with rg-SIFT, Opponent-SIFT, Hue-SIFT, HSV-SIFT, C-SIFT and opponenthistogram descriptors.
  • F_A_Brno_spacetime_1: This run extends F_A_Brno_color_2 by adding space-timevisual features STIP and HESSTIP.

2. Combining multiple types of visualfeatures improves results significantly. F_A_Brno_color_2 achieve more thantwice better results than F_A_Brno_resource_4. The space-time visual featuresdid not improve results.

3. Combining multiple types of visualfeatures is important. Linear SVM is inferior to non-linear SVM in the contextof semantic indexing.

Content-based Copy Detection

1.    Two runs submitted, but with similar settings; the difference isonly in amount of processed test data (40% and 60%)

  • brno.m.*.l3sl2: SURF,bag-of-words (visual codebook: 2k size, 4 nearest neighbors used insoft-assignment), inverted file index, geometry (homography) based imagesimilarity metric

2.    What if any significant differences (in terms of what measures) didyou find among the runs?

  • only one setting used - nodifferences

3.    Based on the results, can you estimate the relative contribution ofeach component of your system/approach to its effectiveness?

  • slow search in referencedataset due to unsuitable configuration of used visual codebook

4.    Overall, what did you learn about runs/approaches and the researchquestion(s) that motivated them?

  • change the way of describingthe video content - frame based (or key-frame based) approach is not sufficient

Anglický abstrakt

This paper describes our approach to semantic indexing and content-based copy detection which was used for TRECVID 2010 evaluation.

Semantic indexing

1.  Theruns differ in the types of visual features used. All runs use severalbag-of-word representations fed to separate linear SVMs and the SVMs were fusedby logistic regression.

  • F_A_Brno_resource_4: Only single best visual features (on the trainingset) are used - dense image sampling with rgb-SIFT.
  • F_A_Brno_basic_3: This run uses dense sampling and Harris-Laplace detector incombination with SIFT and rgb-sift descriptors.
  • F_A_Brno_color_2: This run extends F_A_Brno_basic_3 by adding densesampling with rg-SIFT, Opponent-SIFT, Hue-SIFT, HSV-SIFT, C-SIFT and opponenthistogram descriptors.
  • F_A_Brno_spacetime_1: This run extends F_A_Brno_color_2 by adding space-timevisual features STIP and HESSTIP.

2. Combining multiple types of visualfeatures improves results significantly. F_A_Brno_color_2 achieve more thantwice better results than F_A_Brno_resource_4. The space-time visual featuresdid not improve results.

3. Combining multiple types of visualfeatures is important. Linear SVM is inferior to non-linear SVM in the contextof semantic indexing.

Content-based Copy Detection

1.    Two runs submitted, but with similar settings; the difference isonly in amount of processed test data (40% and 60%)

  • brno.m.*.l3sl2: SURF,bag-of-words (visual codebook: 2k size, 4 nearest neighbors used insoft-assignment), inverted file index, geometry (homography) based imagesimilarity metric

2.    What if any significant differences (in terms of what measures) didyou find among the runs?

  • only one setting used - nodifferences

3.    Based on the results, can you estimate the relative contribution ofeach component of your system/approach to its effectiveness?

  • slow search in referencedataset due to unsuitable configuration of used visual codebook

4.    Overall, what did you learn about runs/approaches and the researchquestion(s) that motivated them?

  • change the way of describingthe video content - frame based (or key-frame based) approach is not sufficient

Klíčová slova

TRECVID, semantic indexing, Content-based Copy Detection, image classification

Klíčová slova v angličtině

TRECVID, semantic indexing, Content-based Copy Detection, image classification

Autoři

HRADIŠ, M.; BERAN, V.; ŘEZNÍČEK, I.; HEROUT, A.; BAŘINA, D.; VLČEK, A.; ZEMČÍK, P.

Vydáno

28.12.2010

Nakladatel

National Institute of Standards and Technology

Místo

Gaithersburg, MD

Kniha

2010 TREC Video Retrieval Evaluation Notebook Papers

Strany od

1

Strany do

10

Strany počet

11

URL

http://www-nlpir.nist.gov/projects/tvpubs/tv10.papers/brno.pdf

BibTex

@inproceedings{BUT34908,
  author="Michal {Hradiš} and Vítězslav {Beran} and Ivo {Řezníček} and Adam {Herout} and David {Bařina} and Adam {Vlček} and Pavel {Zemčík}",
  title="Brno University of Technology at TRECVid 2010 SIN, CCD",
  booktitle="2010 TREC Video Retrieval Evaluation Notebook Papers",
  year="2010",
  pages="1--10",
  publisher="National Institute of Standards and Technology",
  address="Gaithersburg, MD",
  url="http://www-nlpir.nist.gov/projects/tvpubs/tv10.papers/brno.pdf"
}