Detailed accuracy maps communicate effectively limitations of RS classification products. Using interpolation methods in the spectral space we demonstrate that per-pixel accuracy predictions are feasible and they offer a better representation of accuracy variability in a given scene.
Further information: R. Khatami, G. Mountrakis, S. Stehman (2017). Mapping per-pixel predicted accuracy of classified remote sensing images. Remote Sensing of Environment, 191, 156–167. [3.4MB pdf]