Super-resolution mapping of urban buildings with remotely sensed imagery based on prior shape information

Urban building extraction is an important task of remotely sensed imagery analysis. Many automatic or semi-automatic algorithms have been proposed to address this problem using various remote sensing sources. Although very high spatial resolution remotely sensed imagery has been used in this field, the large number of mixed pixels that currently still exist in these images makes the extracted urban buildings inaccurate, especially with respect to determining building boundaries. Super-resolution mapping is a promising technology that will improve the spatial resolution of land cover mapping with remotely sensed imagery. This technology uses the fraction maps derived with soft classifications as input and converts them into high resolution land cover maps based on the land cover spatial pattern, which is often described with the maximum spatial dependence principle. Although previous research proved the effectiveness of this principle, it is still not suitable for some special land classes, especially those of man-made objects. In this study, we revised the normal spatial dependence principle to incorporate the prior shape information of urban buildings to make the super-resolution mapping technology more suitable for urban building extraction. The proposed algorithm was evaluated with several simulated images. Our results show that the proposed method can obtain more accurate maps than those produced by the standard super-resolution mapping method. Incorporating more specific prior information improves the performance of super-resolution mapping with remotely sensed imagery.