Improving the quality of remotely sensed derived land cover maps by incorporating mixed pixels in various stages of a supervised classification process

Conventional per-pixel classification methods may be inappropriate to classify images dominated by mixed pixels, as these are based on pure pixel assumption. The aim of this paper is to demonstrate the improvement in the quality of land cover classification by accounting for mixed pixels in all the stages of supervised image classification process. Three markedly different methods - a maximum likelihood classifier, a fuzzy c-means algorithm and a linear mixture model have been used.