Contribution des pyramides irrégulières en segmentation d'images multirésolution
Pascal Bertolino · HAL (Le Centre pour la Communication Scientifique Directe) · 1995
This thesis shows the contributions of the irregular pyramidal structure within the field of grey level image multiresolution segmentation. First, multiresolution techniques are reviewed. Then, several image segmentation techniques are explored, involving either region or contour approaches. Thirdly, irregular pyramid construction and its use are thoroughly described, including its flaws. In order to improve the relevance of this approach, a cooperative process combining the pyramidal structure and either a classical edge operator or a stochastic edge detector (specially developed) is proposed. Finally, two new methods that provide multiresolution segmented image pyramids are proposed: The first method provides, by using the different levels of an irregular pyramid, a large number of derived segmentations, each one at a particular resolution. The second method, based on the wide decimation, directly generates a real image pyramid in which each level contains details at a particular resolution. Experimental results are given all along the document. A comparison among the proposed methods is performed on test images.