Spatial contextual noise removal for post classification smoothing of remotely sensed images

Extracting accurate land use and land cover information from remote sensing data is a challenging problem due to the gap between theoretically available information in remote sensing imagery and the limited classification ability based on spectral analysis. Traditional classification techniques based on spectral analysis of single pixel usually produce "noisy" results that contain many wrongly classified pixels. This paper presents a novel post classification method to detect the pixels that are wrongly classified and reassign them to correct fields in spatial context. The strategy is demonstrated through the classification of a benchmark digital aerial photograph. The experimental results show that the proposed approach can produce a more accurate classification than previous approaches.