Automated techniques for satellite image segmentation

In this paper a combined method between “classical” and automatic approach for remote sensing image analysis is presented. Tipically, satellite images are used in order to detect the distribution of vegetation, soil classes, built-up areas, roads, and water body as rivers, brooks, lakes, ecc. Referring for example to Landsat-TM images, the identification of such aspects is performed through the “classical” approach of image classification. Basically, it deals with the use of pseudocolors and/or combinations of various spectral bands to acquire different thematic layers from the images. In this work a further processing step is introduced, namely a segmentation algorithm is applied to color images in order to improve the image analysis both from qualitative and quantitative point of view. This algorithm belongs to the class of operations performed in the automatized unsupervised analysis of color images. Last recent advances in the field of computer science and CPU performance have lead to a great reduction of data processing times, allowing therefore to apply also in the field of Remote Sensing more complex algorithms. The proposed segmentation algorithm is based on a feature-space approach and implements two processing techniques: the “histogram thresholding” [1] and the “clustering” [2]. Some interesting results applied to Remote Sensing images will be provided.