This work is part of the GEOSAFRAS program - estimates and harvests in Brazil - which is coordinated by the National Company of Supply and the United Nations Development Programme. A regular use in the management of harvests is the monitoring of the cultivation fields changes by using satellite images obtained at different dates. For this purpose, the spatial compatibility between these images is a necessary condition, even more when it comes to images from different sensors. This article describes a methodology for the semi-automation of the geometric rectification process aimed at facilitating the geometric adjustment between images from different sensors. First, a base image is manually rectified from data extracted from a digital topographic map. After such rectification, the correct image is used as a reference element to rectify other images (in this work, called adjustment images), which are acquired at other dates and/or from other sensors. Once the base image is rectified, all images (base and adjustment ones) are segmented and classified. The segments classified as forest vegetation are selected to compose the relational mesh. The adjustment images are registered by the image-image process by means of the forest vegetation segment centroids. The forest vegetation segments in the base image are compared to the corresponding segments in the adjustment images, to find pattern matching. The matching process involves the application of genetic algorithms. After obtaining a positive result regarding the pattern matching, the centroids corresponding to the detected segments are calculated; These centroids are used as control points for the image registry process. The results show that genetic algorithms have found the optimal solution in most experiments. However, regarding the LANDSAT 2002 resampled image, the solution found was sub-optimal because one of the segments showed large variations in relation to the same segment of the base image.
Image Registration; Region Matching; Genetic Algorithm
