Skin and Bones: Multi-Layer, Locally Affine, Optical Flow and Regularization with Transparency

Abstract

This paper describes a new methodfor estimuting op-tical jlow that strikes a balance between the jlexibil-ity of local dense compututions and the robustness and accuracy of global parameterized jlow models. An affine model of image motion is used within locul im-age patches while a spatial smoothness constraint on the afine$ow purameters of neighboring patches en-forces continuity of the motion. We refer to this as a “Skin and Bones ” model in which the affine patches cun he thought ofus rigid “bones ” connected by aJEex-ible “skin ”. Since local image putches may coiztuin multiple motions we use a layered representation fir the affine bones. To regularize this luyered motion rep-resentation we develop a new framework for regular-ization with transparency. 1

Cite

Text

Ju et al. "Skin and Bones: Multi-Layer, Locally Affine, Optical Flow and Regularization with Transparency." IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1996. doi:10.1109/CVPR.1996.517090

Markdown

[Ju et al. "Skin and Bones: Multi-Layer, Locally Affine, Optical Flow and Regularization with Transparency." IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1996.](https://mlanthology.org/cvpr/1996/ju1996cvpr-skin/) doi:10.1109/CVPR.1996.517090

BibTeX

@inproceedings{ju1996cvpr-skin,
  title     = {{Skin and Bones: Multi-Layer, Locally Affine, Optical Flow and Regularization with Transparency}},
  author    = {Ju, Shanon X. and Black, Michael J. and Jepson, Allan D.},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year      = {1996},
  pages     = {307-314},
  doi       = {10.1109/CVPR.1996.517090},
  url       = {https://mlanthology.org/cvpr/1996/ju1996cvpr-skin/}
}