An "optimal" filter for maps showing nominal data

An "optimal" filtering technique for use with nominal data, such as land use and land cover categories, has been developed. This method is based on the conditional probability joins of neighboring data elements. In addition to its use in performing filtering, the method can be used to calculate the likelihood of each data element being properly classified. The technique was tested on a land use data set for an area in Walnut Valley, Calif. The computer program performing the filtering process proved to be computationally efficient and produced satisfactory results. Useful statistics of the error estimation process were also generated. Future applications of the method to spectrally classified Landsat data are being explored.