Classification using the watershed method

A. I. WATSON, R. A. Vaughan, Malea Powell

International Journal of Remote Sensing · 1992 · 14 citations · 5 references

Concepts

Abstract

Abstract Several methods of producing a thematic map, suitable for forestry inventories, are evaluated as to their relative accuracy and efficiency of production. It is argued that all probability-based methods are founded on assumptions that are always false, and therefore necessarily lead to higher error rates. An alternative non-probabilistic method, the watershed, is put forward as a better solution to the classification problem. In order to fully establish the superiority of the watershed method, a complex mountainous area was deliberately chosen to provide difficult and testing conditions. It is demonstrated that the watershed method is far superior to the traditional probability-based methods, both in respect of the efficiency with which a thematic map can be produced, and its accuracy of classification. With the same data, the accuracy of classification were: hybrid method—77 percent supervised maximum likelihood method—82 percent watershed method—96 percent.

References

5