Segmentation non-supervisée d'images couleur par sur-segmentation Markovienne en régions et procédure de regroupement de régions par graphes pondérés

Content Provider	Semantic Scholar
Author	Hedjam, Rachid
Copyright Year	2009
Abstract	This thesis presents a new approach to hierarchical unsupervised color image segmentation into regions. The problem is how can we extract the objects that the human perceives on the image rather than to partitioning it into several regions not significant (over-segmentation). This pro blem of over-segmentation can not be resol ved by a classical segmentation algorithm, so we thought to introduce a new hierarchical method based on 1) Bayesian-Markovian grayscale segmentation, in this step an oversegmentation map is obtained with homogeneous regions in gray-Ievel sens, followed by 2) a region segmentation based on graph partitioning (texturaI segmentation). Then we have to repeat step 2 (re-segmentation) (with automatic adjustment of a regularization parameter) as many times as necessary until a stop criterion is reached, so that from an iteration to another, the regions more similar are merged to form larger and homogeneous objects that representdiscernible objects in the picture . The proposed approach was validated on the Berkeley images database.
File Format	PDF HTM / HTML
Alternate Webpage(s)	https://papyrus.bib.umontreal.ca/xmlui/bitstream/handle/1866/7221/Hedjam_Rachid_2009_memoire.pdf;jsessionid=9A887A1424AEC12909A857D2B7998B63?sequence=1
Language	English
Access Restriction	Open
Content Type	Text
Resource Type	Article