Improving Image Alignment in Aerial Image Mosaics via Error Estimation of Flight Attitude Parameters

A means of flight attitude parameter error estimation preprocessing is presented for the assembly of overlapping aerial image mosaics, via image registration using a pattern search method, mapped onto a GIS grid. The method presented first predicts which images will not align well from a data set, then it uses a correlation function as an optimiser within a modified Hooke and Jeeves algorithm to find a more optimal transformation function input for each image to the Mosaic program. Using this improved input, the Mosaic program will generate mosaics whose constituent images are better aligned. We demonstrate that creating more area based regions for alignment within the images, filtering them for disqualifying parameters, and using the good ones to optimise the above transformation input will improve the quality of mosaics produced by improving the alignment of these error estimate selected, difficult images. The process has been shown to group the misaligned images into the worst roughly 12% of the data. The worst case image is followed as the process takes it from an RMS error of approximately 12 pixels to within one or two pixels of perfect alignment.