From fuzzy and object based classification to fuzzy and object based uncertainty evaluation

Regarding thematic processing of remote sensing data new problems have arisen with the rapid increase of geometric and spectral resolution. These have been partly solved through the application of object oriented methods and alternative (e.g. fuzzy logic) approaches for the actual allocation of a feature to a topographical object whereas these methods do not apply comprehensively to the quality assessment of the processed data. We present an integrated approach for the assessment of classified high-resolution remote sensing scenes which considers uncertainties - not only in the classified data but in the reference ("ground truth") data as well. Instead of discrete object boundaries we define transition zones between adjacent objects; a fuzzy function describes the distribution of class membership values within these zones. Thus we can compute an evaluation measure on the basis of the uncertainty model - the CFCM (Class-specific Fuzzy Certainty Measure) provides a quality assessment for classified remote sensing data considering uncertainties in geometry and semantics. The work is part of the project "CLassification Assessment using an Integrated Method (CLAIM)".