A framework for multi-date multi-sensor image interpretation

This paper describes a methodological framework for multi-date multi-sensor remotely-sensed image interpretation. The framework was built based on an integrated multi-sensor classification scheme. The classification scheme considers a remotely-sensed input image to consist of homogeneous and textured regions. This approach allows flexibility in performing digital image mosaicking and data fusion for the purpose of cloud cover removal and a more consistent scene interpretation. The authors have determined that both in optical and radar images the land cover types can be grouped into homogeneous and textured regions. Using their classification scheme, the discrimination between land cover types in each group can be enhanced. It was found that better classification accuracy can be obtained by not assuming that the input image is fully homogeneous where a classifier based on image tone or spectral features is applied or fully textured where a classifier based on texture features is applied. Several results on digital image mosaicking and data fusion using the proposed classification scheme are presented