The Congealing Page

What is congealing?

Congealing is an algorithm for the joint alignment of a set of images. This notion of "alignment" can be construed very broadly. It includes various types of spatial alignment, aligning images under, say, rotation and translation. The binary digits above are aligned by using the set of affine transformations to transform each zero until it matches the others best. The set of faces above are aligned by using similarity transformations. (Mouse over the images to see the alignment happen.) But congealing can also refer to the removal of other forms of non-spatial variability, like brightness transformations, that make a set of images different. The magnetic resonance brain images above are being "aligned" with respect to a set of smooth brightness transformations. This can help eliminate artifacts that may cause the erroneous automatic interpretation of the images.

Congealing demo code.

You can download code for congealing. This code runs congealing on a set of handwritten zeros. I hope to put more comprehensive code on the web at some point, but this will let you see how it works.

Congealing download 1.0: Tar ball

Congealing Papers

The following publications are related to congealing.

My own papers:

Erik Learned-Miller, (2005) Data driven image models through continuous joint alignment. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 28:2, pp. 236-250, 2006.

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

(Best Paper Award) Lilla Zollei, Erik Learned-Miller, Eric Grimson and William Wells, (2005) Efficient population registration of 3D data. Workshop on Computer Vision for Biomedical Image Applications: Current Techniques and Future Trends, at the International Conference of Computer Vision (ICCV).

Erik Miller, Nick Matsakis and Paul Viola, (2000) Learning from one example through shared densities on transforms. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Vol. 1, pp. 464-471.

Erik Learned-Miller and Vidit Jain, (2005) Many heads are better than one: Jointly removing bias from multiple MRs using nonparametric maximum likelihood. Proceedings of Information Processing in Medical Imaging, pp. 615-626, 2005.

Erik Learned-Miller and Parvez Ahammad, (2004) Joint MRI bias removal using entropy minimization across images. NIPS, 2004.

Marwan Mattar, Michael G. Ross, and Erik Learned-Miller (2009) Non-parametric curve alignment. International Conference on Acoustics, Speech, and Signal Processing (ICASSP). Erik G. Miller, (Feb., 2002) Ph.D. Thesis: Learning from One Example in Machine Vision by Sharing Probability Densities. Massachusetts Institute of Technology.

Additional papers:

A. Vedaldi and S. Soatto. A rate-distortion approach to joint pattern alignment. Neural Information Processing Systems 19, 2006.

Parvez Ahammad, Cyrus Harmon, Ann Hammonds, Shankar Sastry and Gerald Rubin, (2005) Joint nonparametric alignment for analyzing spatial gene expression patterns of drosophila imaginal discs. Computer Vision and Pattern Recognition (CVPR), 2005.

Congealing Movies

This movie shows a set of threes from the NIST database being congealed. (12 Megabytes).

This movie shows a set of sevens being congealed. (12 Megabytes)

Congealing Images

Move your mouse over the images below to see the results of congealing on the images.

This set of images is a set of infant MR images before bias correction.
MR Bias Correction

This set of images is the same set of infant MR images after our bias correction method.
MR Bias Correction

Finally, the set below is a set that have been corrected using a single image method. Notice the repression of the developing white matter in the mid-brain (butterfly-like shapes).
MR Bias Correction

What if I have other questions about congealing?

If you have questions about congelaing, please email me at
elm"at"cs"dot"umass"dot"edu