Approchecomparèe de mèthodes de classification d’images aèriennes: Une ètude de cas

Abstract

Comparative Approach to the Classification Methods of Aerial Images: a case study The current expanding availability of high resolution satellite and aerial images in digital form (< 5 m) has forced image processing analysts and developers to reconsider Classification strategy and procedures. Clas-sical procedures, although improved and validated for low resolution images, are no longer able to make full use of the richer and more complex high resolution image content. Using a case study on the evaluation of Lotharhur-ricane forest damages, the authors intend to compare the process and performances of two different Classification approaches: a classical unsupervised method - clustering - and a regional growth Classification method. Results are then compared with thoseobtai-ned by photo-interpretation. Although results from a visual interpretation and from numerical processing cannot be compared on a quantitative basis, due to the use of different rules applied for contour generalisa-tion of spatial units, it is meaningful to consider the results of photo-interpretation process as a reference for this comparison. This paper includes a brief presentation of the regional growth Classification procedure, emphasising basic principles and procedure Steps. As the aerial image used in this experimentation contains a strong hotspot effect, it demonstrates the strength of thematic signature definition proposed by the regional growth method. All damaged areas were successfully depic-ted throughout the heterogeneously illuminated image with this latter method, as unsupervised Classification produced less satisfying results in weakly reflective zones. © Author(s) 2003. This work is distributed.