Specific class extraction from remote sensing imagery based on nearest neighbor classification

Specific class extraction is an important part of information extraction from remotely sensed imagery. Based on the nearest neighbor classification rule, this paper studies the specific class extraction from remote sensing imagery. With the nearest neighbor classifier, the specific class extraction is considered as a two-class case, the interested and uninterested class. Firstly the mean shift based clustering technique was used to guarantee a good sample selection for the uninterested class. Then the nearest neighbor classification was performed to extract the interested class. To evaluate the quality of the interested class extraction, classification error probability was computed in the experiment.