Multivariate watershed segmentation of compositional data
DGCI'09 Proceedings of the 15th IAPR international conference on Discrete geometry for computer imagery
Hi-index | 0.00 |
In this paper, a new method for segmenting multispectral remote sensing image is proposed, which combines spectral properties of the pixels and their spatial properties. Spectral properties are studied by analyzing spectral angle of pixels while spatial properties are studied by morphological method. The spectral angle of each pixel is first computed by taken them as a vector with n-dimension. After an automatic selection of significant minima, an initial segmentation is achieved by applying watershed transformation to spectral angle map. To overcome the over-segmentation problem of watershed transformation, a region merging process using region adjacency graph (RAG) is employed to get the final segmentation result.