Abstract :
Image segmentation is often regarded as the first and most important step for other higher level image interpretation, e.g. information extraction and image mining. Although a lot of researches have dedicated to this field, due to its intrinsic dilemma, there exist a wide range of shortcomings of current segmentation methods. When applied to remotely sensed imagery which are commonly with tremendous data volume and very complex ground feature distributions, it will encounter much more difficulties in extracting meaningful and valuable patterns. In this research, we classify remotely sensed imagery into two types: the gray value and texture imagery, and then search their respective suitable segmentation methods. More than 12 segmentation algorithms are implemented and integrated into a multi-scale segmentation framework, which is illustrated and validated with two typical applications on segmenting and extracting manmade objects from high spatial resolution remotely sensed imagery.
