Skip to Main Content
As a greater amount of remotely sensed data becomes available, many of the applications will require the interpretation of images collected over a regular period (perhaps every day or every week) in order to identify changes associated with applications such as land-cover mapping, and environmental modelling and monitoring. Since these applications will require the transmission of significant numbers of remotely sensed images, efficient data transmission and analysis is required. Rather than transmit the complete image each time (even with compression), in this paper we propose to separate the unchanged portions (the majority) of the image from the changed portions between successive images of the same area. We have iteratively fitted a linear predictive model between the previous image and the unchanged data in the current image. The model is iterated-through a process of outlier removal-until the two successive models do not change significantly. As a result, the residuals are small for the stable pixels that have little need for updating, whereas the changed pixels are identifiable due to their relatively large residuals. Unlike traditional compression algorithms, our technique reduces the amount of data that is required to be transmitted whilst highlighting the areas in the image that have changed.