| Title | Classification or Enhancement: A New Method for Digital Analysis of Multichannel Raster Data |
| |
| Author | Cihlar, J; Xiao, Q; Beaubien, J; Chen, J M |
| Source | International Symposium, Geomatics in the Era of RADARSAT (GER'97), Ottawa, Canada, May 25-30; 1997 p. 7 |
| Year | 1997 |
| Alt Series | Earth Sciences Sector, Contribution Series 20041999 |
| Document | book |
| Lang. | English |
| Media | paper |
| Abstract | A new procedure for digital image classification is described. The procedure, labeled Classification by Progressive Generalization (CPG), was developed after examining the drawbacks associated with most
supervised and unsupervised classifications. Instead of using training samples the CPG aims to identify all significant spectral clusters within the image. This is achieved by treating all the initial combinations of spectral digital values as
potentially meaningful clusters, and using a series of quantization, spatial, classification, ordering and chaining operations to combine the initial large number of clusters into a reduced number. The last classification step is labeling, i.e.
assignment of the clusters to the classes in the classification legend. The labeling is assisted by displaying the clusters in colours that are meaningful to a human image interpreter. The CPG procedure was tested with high (Landsat TM) and medium
(AVHRR 1km composites) resolution data. It was found that this approach yields classifications that retain most of the information content of the initial image data, as evidenced by the fact that the images representing the original data and the
classifications are not easily distinguished. Yet, the number of spectral clusters is reduced dramatically, typically from many thousands to less than 100 in a land image of northern landscapes. Tests have shown that classification accuracies
comparable to, or better than, those for current methods. |
| GEOSCAN ID | 218801 |
|
|