Thesis
/ | ROMDOC-THESIS-2010-032 |
Metode avansate de modelare statistică pentru segmentarea imaginilor satelitare în zone urbane, rurale şi mixte
Oprisescu, Serban (UPB)
2007-01-01
Abstract: Motivation: Corrected satellite images are still affected by random noise from a variety of sources and, therefore, their processing is compulsory in order to reach a “correct” reading of the covered land. Addressed problem: Let us consider a true satellite image of a mixed area, rural and urban. We aim at classifying this image (segmentation and regularization) by means of an optimal classifier. We propose a complete methodology for the analysis and processing of satellite images for mixed areas, rural and urban. The following stages of analysis are addressed: statistical modeling of the satellite image, training the classifier, optimal segmentation of the image, optimal regularization of the segmented image. -The statistical model involves Gaussian distributions for the “rural” areas and multivariate Gaussian mixtures for the “urban” areas. The model is supported by a thorough texture analysis. Segmentation is achieved by means of an adequate Maximum a Posteriori (MAP) classifier. Optimal training of the classifier is obtained by means of the EM algorithm (for the iterative construction of the maximum likelihood estimator) in association with the Akaike criterion (for optimal selection of a statistical model). The MAP-Markov regularization method is proved to be the optimal regularization technique. All studied methods are validated through an extended experimental study accompanied by computational algorithms and semi-automate methods.
Keyword(s): Modelare statistică -- Teze ; Segmentarea imaginii -- Teză de doctorat ; Prelucrarea imaginii -- Teză de doctorat ; Satelit artificial Prelucrarea imaginii -- Teză de doctorat
Note: Buzuloiu, Vasile
OPAC: See record in BC-UPB Web OPAC
Full Text: see files
Notice créée le 2010-12-07, modifiée le 2011-11-11
Notices similaires
Les utilisateurs qui ont vu cette page ont aussi vu: |
|
|
Soyez le premier à noter ce document
Commencer la discussion sur n'importe quel aspect de ce document.