CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0587
Title: Hyperspectral image segmentation based on functional kernel density estimation Authors:  Laurent Delsol - University of Orleans (France) [presenting]
Cecile Louchet - University of Orleans (France)
Abstract: Splitting a picture into a set of homogenous regions is a common problem, called segmentation, in image analysis. The detection of such regions is usually a relevant way to identify specific parts of the scene. Various methods have been proposed to segment gray-level or multispectral images. The maximum a posteriori approach, based on Potts random field as prior and density estimation on each region, is an interesting use of Bayesian statistics in that domain. On the other hand, a great variety of functional statistical methods are nowadays available to deal with data sets of curves. The kernel density estimator has been adapted to such data. We focus on hyperspectral images for which each pixel is described through a curve (discretized on a thin grid) and discuss the way functional kernel density estimation and maximum a posteriori approach may be combined. The choice of granularity and smoothing parameters will be discussed. And we will present some simulations and applications on real world images.