Direct sub-pixel mapping exploiting spatial dependence

Remotely sensed images usually contain both pure and mixed pixels. Crisp classification techniques assign mixed pixels to the class with the highest proportion of coverage or probability. Unfortunately, during this process information is lost. Soft or fuzzy classification techniques were introduced to make up for this loss by assigning fractions to the land cover classes in correspondence with the area represented inside a pixel. A fuzzy classification yields a number of fraction images equal to the number of land cover classes considered in the classification. However, the assignment to these classes renders no information about the location of these fractions inside the pixel. P.M. Atkinson (1997) stated that it is possible to assign the fractions spatially to so called ´sub-pixels´. Sub-pixels are a finer representation derived from a parent pixel. This work introduces a sub-pixel mapping algorithm exploiting spatial dependence. The spatial arrangement of the different class fractions in surrounding pixels is used to find the location of the sub-pixels inside the central pixel. The sub-pixel mapping algorithm is intended to he applied to fraction images of a high spatial resolution. Errors due to coregistration and poor classification are excluded by using synthetic imagery that was created by degrading hard classifications to coarser spatial resolutions. Resulting images are interpreted as fraction images. The algorithm is then tested for its ability to reconstruct the original hard classification. The accuracy of the algorithm is evaluated using standard classification accuracy measures. The algorithm proposed incorporates spatial dependence in a simple manner, reaching accurate results in a limited computation time