Methods and Theory : Congealing Complex Real-World Object Images

Overview

Congealing is a method of jointly aligning a set of images. While this method, which transforms images to minimize the entropies of pixels at corresponding locations in the image set, is relatively straightforward for binary images, it is more complicated to make it work for real world images. In particular, real photographs (of people, for example), contain complex lighting effects, varied backgrounds, and great variability in the subject itself. Applying congealing in a straightforward manner simply does not work. In this work, Gary B. Huang developed methods to extend congealing to complex, real-world images such as collections of unconstrained face photographs, or photographs of automobiles taken from a particular point of view.

You can read more about this technique here, which addresses joint alignment of faces. The method has also been applied to automobiles, which is described in the ICCV paper referenced below.

Faculty


Graduate Students


Publications

  • Gary B. Huang, Vidit Jain, and Erik Learned-Miller.
    Unsupervised joint alignment of complex images.
    International Conference on Computer Vision (ICCV), 2007.
    [pdf]