Integrated methodology for segmentation of large optical satellite images in land applications of remote sensing
Per-pixel classification plays an important role within remote sensing. Pixels are divided into classes with each class having its own label. Segmentation is a method which uses the characteristics from both the spectral and the spatial domain in the labelling process. Therefore it is expected that segmentation can improve the classification process. To detect (strong) edges an edge detection (Frei-Chen method) is applied. After some pixel thick edge fragments have been detected, these are linked by an edge linking method which adds missing edge pixels. Not linked edge fragments are removed. The linked edge fragments are closed polygons which divide the image into assumed independent parts. Within each polygon a region growing method is then applied. After the merging, edges are added to regions obtained whenever possible or are maintained as a separate region. An illustration of the developed hybrid method is given for three different land applications.
Bibliographic Reference: EUR 16292 EN (1995) 176 pp., FS, free of charge
Availability: Available from Joint Research Centre, Public Relations and Publications Unit, I-21020 Ispra (Va.) (IT)
Record Number: 199511264 / Last updated on: 1995-11-23
Original language: en
Available languages: en