PIXEL-BASED RECLASSIFICATION OF CLASSIFIED IMAGES AT DIFFERENT SPATIAL RESOLUTIONS.

The use of Contextual information from within classified image data at different spatial resolutions for improving classification accuracy was investigated using a simple pixel-based reclassification technique. Classification accuracy improvement resulting from the application of this technique was evaluated by comparing classification error rates before and after reclassification. Whilst reclassification tended to improve classification accuracy at finer resolutions this effect was countermanded by the initial error rate associated with each individual class present in the scene. Classification accuracy improvement following reclassification was found to be strongly dependent on initial classification error rate.