Unsupervised Dynamic Texture Segmentation Using Local Spatiotemporal Descriptors

  • Upload
    pepin

  • View
    24

  • Download
    3

Embed Size (px)

DESCRIPTION

Unsupervised Dynamic Texture Segmentation Using Local Spatiotemporal Descriptors. Jie Chen University of Oulu, Finland. A demo show. Segmentation of a dynamic texture. Input. Output. Outline. Motivation Related work Our methods Experimental results Conclusion. Dynamic texture. - PowerPoint PPT Presentation

Citation preview

  • Unsupervised Dynamic Texture Segmentation Using Local Spatiotemporal Descriptors

    Jie ChenUniversity of Oulu, Finland

  • A demo showSegmentation of a dynamic texture InputOutput

  • OutlineMotivation

    Related work

    Our methods

    Experimental results

    Conclusion

  • Dynamic textureMotivationDynamic textures or temporal textures are textures with motion.There are lots of DTs in real world, including sea-waves, smoke, foliage, fire, shower and whirlwind, etc.Click the figure

  • Dynamic texturePotential applications: Remote monitoring and various type of surveillance in challenging environments:monitoring forest fires to prevent natural disasterstraffic monitoringhomeland security applicationsanimal behavior for scientific studies.

  • Related workMixtures of dynamic texture modelA.B. Chan and N. Vasconcelos, PAMI2008 Mixture of linear modelsL. Cooper, J. Liu and K. Huang, Workshop in ICCV2005 Multi-phase level setsD. Cremers and S. Soatto, IJCV2004 Gauss-Markov models and level setsG. Doretto, A. Chiuso, Y. N. Wu and S. Soatto, ICCV2003 Ising descriptorsA. Ghoreyshi and R. Vidal, ECCV2006 Optical flowR. Vidal and A. Ravichandran, CVPR2005

  • Related workOur method is based on the following work: LBP-TOP: local binary patterns in three orthogonal planes,Zhao and Pietikinen, PAMI 2007

    local binary pattern and contrastOjala, and Pietikinen, PR 1999

  • Our methodsFeature: (LBP/C)TOPLocal binary patterns and contrast in three orthogonal planes

  • MeasureSimilarity measurement

    Distance between two sub-blocks

    d={LBP, XY, LBP, XT, LBP, YT, C, XY, C, XT, C, YT }T.

    (a)

    x

    y

    t

    XY

    XT

    YT

  • DT segmentationThree phases: Splitting, Merging, Pixelwise classification. SplittingMergingPixelwise classificationInput

  • SplittingRecursively split each input frame into square blocks of varying size.

    criterion of splitting: one of the features in the three planes (i.e., LBP and C, =XY, XT, YT) votes for splitting of current block

    (a)

    x

    y

    t

    XY

    XT

    YT

  • MergingMerge those similar adjacent regions with smallest merger importance (MI) value

    MI : MI=f(p)(1-) is the distance between two regions f(p)= sigmoid(p). (=1, 2, 3, ) p=Nb/NfNb is the number of pixels in current blockNf is the number of pixels in current frame

  • Pixelwise classificationCompute (LBP/C)TOP histograms over its circular neighbor for each boundary pixel. Compute the similarity between neighbors and connected models.

    Re-label the pixel if the label of the nearest model votes a different label.

  • Experimental results Some results on types of sequences and compared with existing methods. [6] G. Doretto, A. Chiuso, Y. N. Wu and S. Soatto, Dynamic Texture Segmentation, ICCV, 2003[7] A. Ghoreyshi and R. Vidal, Segmenting Dynamic Textures with Ising Descriptors, ARX Models and Level Sets, ECCV, 2006

  • Experimental resultsResults on sequences ocean-fire-small (a) Frame 8 (b) Frame 21(c) Frame 40(d) Frame 60(e) Frame 80(f) Frame 100

  • Experimental resultsResults on a real challenging sequence

    (a) Frame 5 (b) Frame 10

  • ConclusionProblem: segmenting DT into disjoint regions in an unsupervised way.

    Methods: Each region is characterized by histograms of local binary patterns and contrast in a spatiotemporal mode.

    Results: It is effective for DT segmentation, and is also computationally simple.

  • Thanks!