Patch based spatial redundancy analysis using a contrario methodsPP

We use a contrario methods to identify spatial redundancy in natural images. Several patch similarity functions are investigated and compared (noise robustness, innovation). We focus on textured images and set the a contrario model to be an Asymptotic Discrete Spot Noise, ADSN, or more generally a Gaussian field. Theoretical properties are derived for such models and experimental results are provided. This is a joint work with Agnès Desolneux, Bruno Galerne and Arthur Leclaire.

This is poster number 62 in Poster Session

Valentin De Bortoli (ENS Paris Saclay)
computer vision, image representation, stochastic processes