Gradient Descent Techniques for Multitemporal and Multi- Sensor Image Registration of Remotely Sensed Imagery

Gradient-descent algorithms have been successfully applied to many applications of computer vision, such as stereo matching, object recognition and medical image registration. Our work focuses on applying these techniques to remote sensing image data with a particular emphasis on data acquired under different conditions, such as multi-temporal or multi-sensor data. While most previous work has focused on the geometric component of image registration, our research also deals with the radiometric component associated with different viewing conditions; e.g., different seasons or different atmospheric conditions for multi-temporal data, or different wavelengths for multi-sensor data. In this paper, we will show how pre-processing can detect these variations, and we will highlight the algorithms’ implications when taking into account this radiometric component. Test data include images from the AVHRR, GOES, SeaWIFS, Landsat-7 ETM and IKONOS sensors.