Centre de documentation
scientifique

Segmentierung und Interpretation digitaler Bilder mit Markoff-Zufallsfeldern / J. Klonowski (1999)

Public Titre : Segmentierung und Interpretation digitaler Bilder mit Markoff-Zufallsfeldern Titre original : [Segmentation and interpretation of digital images using Markov random fields] Type de document : Thèse/HDR Auteurs : J. Klonowski, Auteur Editeur : Munich : Bayerische Akademie der Wissenschaften Année de publication : 1999 Collection : DGK - C Sous-collection : Dissertationen num. 492 Importance : 91 p. Format : 21 x 30 cm ISBN/ISSN/EAN : 978-3-7696-9532-8 Note générale : Bibliographie Langues : Allemand (ger) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] analyse de groupement [Termes IGN] analyse texturale [Termes IGN] champ aléatoire de Markov [Termes IGN] distribution de Gibbs [Termes IGN] image multibande [Termes IGN] milieu urbain [Termes IGN] segmentation d'image [Termes IGN] théorème de Bayes [Termes IGN] variance Index. décimale : 35.20 Traitement d'image Résumé : (Auteur)This thesis defines and solves the problem of the interpretation of digital images by labeling processes on two levels based on Markov random fields. Under the assumption that the objects to be identified in the image differ by their textures, pixels are labeled at the low level according to their affiliation to classes of textures. Clusters of pixels with identical labels are forming regions. They are labeled at the high level of the image analysis to obtain the meaning of the objects. Uncertainties are considered by variances for the description of textures and objects as well as by the probabilistic approach for the labeling. The textures at the low level are represented by the Gibbs distribution of the Markov random field for the gray values. Prior information on the labels concerning the textures is introduced by the Gibbs distribution of the Markov random field of the labels. Application of Bayes' theorem joins the two densities to the posterior distri-bution. Its maximization at every pixel yields the labels for the textures. Also on the basis of Markov random fields a description for the existing and the expected objects is obtained at the high level of image analysis. Prior information on the unknown object labels consists of the frequency of the occurence of objects and their neighborhood relations. Maximization of the posterior density leads to the labels. Uncertain interpretations of regions are found by posterior odds for hypotheses. The label "unknown" is attributed to them. The quality of the interpretation is mainly influenced by the segmentation at the low level. Therefore an in-teraction between the two levels on the basis of posterior odds has been realized. The percentage of area of the regions labeled "unknown" is used as an indicator for improving the segmentation, which leads to a better result for the interpretation. The texture classification is first investigated by generated data for a/ better judgement of the influences of the choice of the parameters for the distributions. The experience gained is used in the segmentation of real multispectral aerial photographs. Tests of a color transformation and an image pyramid of the image data with respect to the quality of the symbolic description of the image are following. Finally the interaction between the two interpretation levels is tested on aerial multispectral photographs of urban areas. Numéro de notice : 28003 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=63350

Exemplaires(2)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
28003-01	35.20	Livre	Centre de documentation	Télédétection	Disponible
28003-02	35.20	Livre	Centre de documentation	Télédétection	Disponible