On segment based image fusion

The new generation of satellite and aircraft sensors provides image data of very and ultra high resolution which challenge conventional aerial photography. The high-resolution information, however, is acquired only in a panchromatic mode whereas the multispectral images are of lower spatial resolution. The ratios between high resolution panchromatic and low resolution multispectral images vary between 1:2 and 1:8 (or even higher if different sensors are involved). Consequently, appropriate techniques have been developed to merge the high resolution panchromatic information into the multispectral datasets. These techniques are usually referred to as pansharpening or data fusion. The methods can be classified into three levels: pixel level (iconic) fusion, feature level (symbolic) fusion and decision level fusion. Much research has concentrated on the iconic fusion because there exists a wealth of theory behind it. With the advent of object or segment oriented image processing techniques, however, feature based and decision based fusion techniques are becoming more important despite the fact that these approaches are more application oriented and heuristic. Within this context, the integration of GIS based information can easily be accomplished. The features can come from a specific segmentation algorithm or from an existing GIS database. Within the context of feature and decision based fusion, we present two exemplary case studies to prove the potential of decision and feature based fusion. The examples include