Low-level Vision


Overview

Gradient descent optimization is often used in low-level vision problems, such as optical flow, tracking, or image alignment. The basic assumption in gradient descent is that the beginning of the search lies within the basin of attraction of the right answer. This assumption is often incorrect, especially in the case of large displacements. Blurring the image alleviates the problem, but destroys some of the pixel information. We propose a method for widening the basin of attraction, while avoiding loss of pixel value information. We show improved results on multiple computer vision problems, like object tracking, image alignment, and optical flow.

At UMass, we are actively doing research in several key areas of low-level vision research, including:



hyper face recog

Faculty


Graduate Students


Collaborators


References

  • Laura Sevilla-Lara, Deqing Sun, Erik G. Learned-Miller, Michael J. Black.
    Optical Flow Estimation with Channel Constancy
    European Conference on Computer Vision (ECCV), 2014. (To appear)
    [pdf]
  • Manjunath Narayana, Allen Hanson, Erik Learned-Miller.
    Background Subtraction - Separating the Modeling and the Inference
    Machine Vision and Applications (MVA) journal, 2013.
    [pdf]
  • Manjunath Narayana, Allen Hanson, Erik Learned-Miller.
    Coherent Motion Segmentation in Moving Camera Videos using Optical Flow Orientations
    International Conference on Computer Vision (ICCV), 2013.
    [pdf] [supplementary material] [project website]
  • Manjunath Narayana, Allen Hanson, and Erik Learned-Miller.
    Improvements in Joint Domain-Range Modeling for Background Subtraction
    Proceedings of the British Machine Vision Conference (BMVC), 2012.
    [pdf]
  • Laura Sevilla-Lara and Erik Learned-Miller.
    Distribution Fields for Tracking
    IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012.
    [pdf] [code] [project website]
  • Manjunath Narayana, Allen Hanson, and Erik Learned-Miller.
    Background Modeling using Adaptive Pixelwise Kernel Variances in a Hybrid Feature Space
    IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012.
    [pdf]
  • Laura Sevilla-Lara and Erik Learned-Miller
    Distribution Fields
    Technical Report UM-CS-2011-027, Dept. of Computer Science, University of Massachusetts Amherst, 2011.
    [pdf]