A Fuzzy Possibilistic Scheme of Study for Objects With Indeterminate Boundaries. Application to French Polynesian Reefscapes

submitted (under 2^nde review) to: IEEE Transactions Geoscience and Remote Sensing

Serge Andréfouët (*), Ludovic Roux (**), Yannick Chancerelle (***), Alain Bonneville (*)

(*) Laboratoire de Géosciences Marines et Télédétection - Université Française du Pacifique -
BP 6570 Faaa Aéroport - Tahiti - French Polynesia
E-mail: andrefou@ufp.pf, bonnevil@ufp.pf
Fax: (689) 803.804 Tel: (689) 803.830

(**) Institut de Recherche en Informatique de Toulouse - Université Paul Sabatier -
118, Route de Narbonne, 31062 Toulouse Cedex - France
(Present address: National Space Development Agency of Japan - Earth Observation Research Center 1-9-9 Roppongi, Minato-ku, Tokyo 106, Japan)
E-mail: roux@eorc.nasda.go.jp

(***) Centre de Recherches Insulaires et Observatoire de l'Environnement- URA EPHE CNRS 1453-
BP 1013 Papetoai Moorea- French Polynesia
E-mail: criobe@orstom.rio.net
Fax: (689) 56.13.45 Tel: (689) 56.28.15

Abstract: Boundaries are zones of main interest for ecologists. When an ecosystem is made of object with indeterminate boundaries, they may be considered as zones which have a membership degree in all the class prototypes. Fuzzy multi-sources analysis give the opportunity to build such membership functions in the framework of theory of possibility. One of the advantage of fuzzy techniques over traditional "hard" algorithms is that a pixel is not assigned to one of the class prototype until the final decision stage. Thus, the thematician may interpret the set of data at an intermediate level taking into account the complete hierarchy of memberships degrees. Boundaries and transition zones are recognized with adequate definitions and properties of these hierarchies. In addition with the ability to model spatial indeterminacy, we also assess that a possibilistic representation of objects of the ecological system gives interesting results for image classification. A comparison is given between the multi-source fusion algorithm, a supervised "hard" parametric algorithm and an unsupervised "hard" algorithm. A scheme of analysis over underwater shallow reefscapes in the lagoon of Moorea Island, French Polynesia, is proposed.

Keywords-Theory of possibility, multi-source fusion, spatial analysis, indeterminate boundaries, reefscape.