Computational Information Geometry Wonderland

What is a texture in an image? This is one of the long outstanding problems, we face in computer vision. We have came up a long road already by classifying them as stochastic, semi-geometric, etc, building atlases, mixing them, synthesizing them, etc. But the question remains: what is a texture ???

Can information theory helps in grabbing those concepts? Clearly it should be related to a notion of information that is homogeneously conserved... Our recent work takes this approach by performing multiscale image analysis, and focuses on texture/structure transitions; It deals with alternate min-max entropies. Egg and chicken approach...

Here is a visual

The paper is presented by Prof. Soatto at ECCV

Wednesday, September 08, 2010
09:00 - 10:40
Oral W1: Image Features, Matching, and Motion

The abstract reads as follows:
We present an approach to multiscale image analysis. It hinges on an operative denition of texture that involves a small region", where some (unknown) statistic is aggregated, and a large region" within which it is stationary. At each point, multiple small and large regions co-exist at multiple scales, as image structures are pooled by the scaling and quantization process to form textures and then transitions between textures dene again structures. We present a technique to learn and agglomerate sparse bases at multiple scales. To do so efficiently we propose an analysis of cluster statistics after a clustering step is performed and a new clustering method with linear time performance. In both cases, we can infer all the small and large regions at multiple scale in one shot.
The pdf paper
Frank.