Change identification of remote sensing images based on textural and spectral features

In this study, a framework is proposed for the automation of change identification of remotely sensed images. Automation has long been hindered by the difficulty caused by illuminant inconsistency in directly applying the statistics of an earlier image to a later image. The difficulty is due mainly to inconsistent solar illumination and atmospheric conditions in a multitemporal sequence of images. A change detection process is performed in order to detect low-change areas. Whether or not a pixel is changed is determined based on textural and spectral features. Those low-change samples are used for radiometric calibration between different dates. Furthermore, the low-change samples may provide class statistics for subsequent images. Suppose that class information in the first image is sufficiently representative. By means of post classification, a series of multitemporal images can thus automatically produce the change types with time.