Computational Information Geometry Wonderland

Segmentation consists in partitioning the image into homogeneous regions intended to represent objects. Human excels in segmenting but computers have the difficult task to solve this problem is a bottom-top approach. In fact, each individual may bring its own segmentation result but computers solve some optimization problem.

• Graph (MST, normalized cuts)
• Clustering (k-means, GMMs)
• Region growing (SRM)
  Nock R., Nielsen, F., (2004). Statistical region merging. IEEE Transactions on Pattern Analysis and Machine Intelligence, 26(11):1452-1458.
  see some ecological applications
• Watershed
• etc

An interesting result by Kleinberg is to define a set of three essential properties one good clustering should have, and then prove that there does not exist an objective function to optimize yielding those properties...
J. Kleinberg. An Impossibility Theorem for Clustering. Advances in Neural Information Processing Systems (NIPS) 15, 2002.

Watershed is a segmentation technique that proceeds by filling/detecting basins by dropping water. It takes a greyscale image and manipulate it as a heightmap. To see that it is not trivial task, look at the right column and try to guess the corresponding image on the left column... !!!
Segmentation is a never-ending problem, so what are the next big milestones to focus on?

@FrnkNlsn

The list of accepted work at International Conference on Computer Vision(ICCV) is available for a couple of days. As usual, there are many exciting titles. I made a rough selection considering my interests.

Here it is:

• A Nonparametric Riemannian Framework on Tensor Field with Apllication to Foreground Segmentation
• A New Distance for Scale-Invariant 3D Shape Recognition and Registration
• Learning Nonlinear Distance Functions using Neural Network for Regression with Application to Robust Human Age Estimation
• Fisher Discrimination Dictionary Learning for Sparse Representation
• Means in spaces of tree-like shapes
• Learning Mixtures of Sparse Distance Metrics for Classification and Dimensionality Reduction
• Positive Definite Dictionary Learning for Region Covariances
• StereoCut: Consistent Interactive Object Selection in Stereo Image Pairs
• Panoramic Stereo Video Textures
• Fisher vectors to model spatial layout for image categorization
• Unsupervised Metric Learning for Face Identification in TV Video
• Complementary Hashing for Approximate Nearest Neighbor Search
• A Dimensionality Result for Multiple Homography Matrices
• Efficient Similarity Search for Covariance Matrices via the Jensen-Bregman LogDet Divergence

Frank.

Motion estimation is a fundamental problem in computer vision. A recent paper at SIGGRAPH 2009 dealt with efficient batch nearest neighbors in images using the patchmatch technique. Inspired by this technique, we design Graphmatch : randomized motion estimation on a segmentation tree.

Visualization of the obtained motion estimation : First row, image 73 and image 74 from sequence Football, second row the motion computed using patchmatch algorithm, the motion computed using graphmatch algorithm (finest scale is shown)

This work will be presented at ICIP 2010.

Here is the

• poster
• paper