Segmentación de coberturas de la tierra espectralmente similares empleando campos aleatorios de Markov y características de textura estructural y estocástica

In this thesis,Markovian modeling is applied to perform segmentation of land cover from remote sensing and digital images. The segmentation problem is approached as a classification problem, where the goal is to decompose an image in a set of homogeneous regions using a similarity characteristics se...

Full description

Bibliographic Details
Author: ERIKA DANAE LOPEZ ESPINOZA
Format: doctoral thesis
Status:Versión aceptada para publicación
Publication Date:2009
Country:México
Institution:Instituto Nacional de Astrofísica, Óptica y Electrónica
Repository:Repositorio Institucional del INAOE
Language:Spanish
OAI Identifier:oai:inaoe.repositorioinstitucional.mx:1009/399
Online Access:http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/399
Access Level:Open access
Keyword:info:eu-repo/classification/Segmentación de imagen/Image segmentation
info:eu-repo/classification/Textura de imagen/Image texture
info:eu-repo/classification/Procesos de Markov/Markov processes
info:eu-repo/classification/Remote sensing/Remote sensing
info:eu-repo/classification/Análisis espectral/Spectral analysis
info:eu-repo/classification/Procesos estocásticos/Stochastic processes
info:eu-repo/classification/cti/1
info:eu-repo/classification/cti/12
info:eu-repo/classification/cti/1203
Description
Summary:In this thesis,Markovian modeling is applied to perform segmentation of land cover from remote sensing and digital images. The segmentation problem is approached as a classification problem, where the goal is to decompose an image in a set of homogeneous regions using a similarity characteristics set. In the Bayesian framework using Markov Random Fields (MRF) the image texture is introduced as clique potentials of a second-order posterior energy function. These clique potentials or texture fields are obtained by means of the 2-DWold decomposition and the obtained final function is called texture energy function (TEF). Texture fields are obtained from the frequency domain, therefore, a model is defined through both the spatial (contextual constraint) and frequency (reference fields) domain. This model allows us to define better the segmented image borders. Experiments were carried out on a variety of synthetic and real images. From the segmentation results, it is observed that by incorporating texture fields to the posterior energy function, the segmentation quality is improved. In this thesis, the main result is the TEF function which is possible to introduce within MRF and tree-structured Markov random fields (TS-MRF) models. In this way, a new model for segmentation of classes with similar spectral response based on TSMRF and the TEF function is proposed. In addition, a methodology that involves the TEF function and a stochastic geometry model to improve image segmentation is proposed. The segmentation preliminary results on synthetic images are encouraging, but there is still work to be done in this direction.