Inference Strategies for Texture Parameters

JORGE MARTINEZ; SILVINA PISTONESI; ANA GEORGINA FLESIA

Proceedings CIARP 2015

Springer International Publishing Switzerland 2015

Año: 2015; p. 460 - 467

The Autobinomial model is a commonplace in Bayesianimage analysis since its introduction as a convenient image model. Suchmodel depends on a set of parameters; their value characterize textureallowing to perform classification of the whole image into regions withuniform properties of the model.This work propose a new estimator of the parameter vector of theAutobinomial model based on Conditional Least Square minimizationvia Real Coded Genetic Modeling and analyze its performance comparedto the classical linear approximation, which exchanges the CLS equationwith a reduced Taylor equation prior to minimization. Our simulationstudy shows that the genetic modeling approach gives more accurateestimations when true data is provided. We also discuss its influence in aset of classification experiments with multispectral optical imagery, estimatingthe scalar vector parameter with our estimator and the classicallinear one. Our experiments show promising results, since our approachis able to distinguish image features that the classical approach does not.