The aim of this study is to classify structural cartographic objects in high-resolution satellite images. The target classes have an important intra-class variability because the class definitions belong to high-level concepts. Structural attributes seem to be the most plausible cues for the classification task. We propose an Adaboost learning method using edge-based features as weak learners. Multi-scale sub-pixel edges are converted to geometrical primitives as potential evidences of the target object. A feature vector is calculated from the primitives and their perceptual groupings, by the accumulation of combinations of their geometrical and spatial attributes. A classifier is constructed using the feature vector. The main contribution of this paper is the usage of structural shape attributes in a statistical learning method framework.We tested our method on CNES dataset prepared for the ROBIN Competition and we obtained promising results.
展开▼
