Applying pattern recognition and high-to-low resolution image matching techniques for automatic rectification of satellite images

Two major objectives were achieved in this research. A technique for rectifying satellite imagery based on matching small-format aerial images, whose coordinates were achieved through navigation sensors, was introduced. A new technique for matching images with large scale differences was also developed.; Small-format digital images were used as a source for ground control points for rectifying satellite imagery. The process involved the determination of the position of the center point of the small-format images using onboard global positioning system (GPS) receiver and aircraft attitude measuring sensors and matching these images with a satellite image. Using this technique, planned flight lines over a satellite scene area provided sufficient control points for rectifying the satellite scene in approximately 5 hours of data acquisition time and 10 hours of processing time.; One of the problems in applying such a technique, which has not received sufficient interest in the image matching literature, was the need to match images that are significantly different in resolution. The ground instantaneous field of view (GIFOV) of the small-format aerial images was approximately 25 centimeters while the satellite image had a GIFOV of five meters. The developed matching technique modified traditional least squares matching by adding filter parameters, used to filter the high-resolution image to match the frequency content of the low-resolution image, as unknowns in the least squares matching technique.; The results of both traditional and modified matching techniques were evaluated using 30 control points distributed over the satellite scene. The root mean square errors obtained using the results of the modified matching technique were less than those obtained using traditional matching technique results. This indicated the successful implementation of the new modified matching technique. Less than two pixels root mean square error was achieved in both the x and y directions when object space coordinates of the aerial images center points were derived from the navigation sensors. This accuracy was acceptable considering the accuracy of the navigation sensors.; The correlation coefficient resulting from initial correlation matching provided a good measure for predicting the success of the least squares matching process. The satellite image rectification accuracy was tested using different numbers of matched aerial images. Different correlation coefficient thresholds were used to select these images. The results indicated that increasing the number of images led to an increase in the satellite image rectification accuracy. On the other hand, the results showed that adding more images increased the risk of adding images with false matching.