Epipolar resampling of linear array scanner scenes

Normalized image generation (epipolar resampling) is an important task for automatic image matching. Normalized images facilitate the detection of feature correspondences in the images and therefore provide the advantages of reducing the search space as well as the matching ambiguities. Normalized image generation is a well-established procedure for images captured by frame cameras. Digital frame cameras that produce resolution and ground coverage comparable to those of analog aerial photographs are not yet available. Instead, linear array scanners can be used on aerial or space platforms in order to obtain such characteristics. The resulting scenes are formed by stitching the captured one-dimensional images that are produced as the sensor moves. Rigorous modeling necessitates accessing or estimating a large number of exterior orientation parameters of the images. The resulting epipolar lines are non-straight lines, which causes difficulties in epipolar resampling using the rigorous model. By comparison, the parallel projection model requires a smaller number of parameters, and it results in straight epipolar lines. In addition, as the flying height increases and the angular field of view decreases, similar to the case of space-borne scanners, the true perspective geometry can be approximated by parallel geometry. The mathematical models and the transformations related to the parallel projection model and its relation to the rigorous perspective projection model are developed. An approach for epipolar resampling of linear array scanner scenes based on the parallel projection model is established. Experimental results using synthetic as well as real data prove the feasibility of the developed approach.