"Selective" region growing - an approach based on object-oriented classification routines
暂无分享,去创建一个
Modern satellite sensors of the IKONOS generation offer very high resolution data. As a consequence, thematic classes are represented with high spectral variance. Hence, traditional pixel-based classification often falls short and delivers incomplete and inhomogeneous results. To avoid or at least to reduce these effects a "selective" region growing algorithm was developed, which combines the evaluation of class specific spectral information and immediate vicinity relations of pixels. In a first pass the data are classified by using exclusively spectral information. Based on the assumption that neighbouring pixels are more likely to belong to same class, a second classification cycle is applied with a stringent "vicinity" condition combined with a wider definition of the spectral characteristics. Different from conventional region growing methods, these wider spectral characteristics can be defined separately for each class and are independent from spectral similarity to the original class. By re-iteration of the algorithm on the data, neighbouring pixels, belonging to the respective class, grow selectively, closing successive classification gaps. The algorithm reduces unclassified pixels remarkably, increasing the overall classification accuracy and can be easily adopted in other tasks.
[1] R. Kettig,et al. Classification of Multispectral Image Data by Extraction and Classification of Homogeneous Objects , 1976, IEEE Transactions on Geoscience Electronics.
[2] Arno Schäpe,et al. Multiresolution Segmentation : an optimization approach for high quality multi-scale image segmentation , 2000 .
[3] T. M. Lillesand,et al. Remote Sensing and Image Interpretation , 1980 .