Digital image matching is a crucial process in photogrammetry that aims at identifying corresponding features from stereo pairs of digital images. This research presents an investigation to the precision reachable by applying three image matching techniques to automatically locate conjugate points in digital aerial imagery. One technique uses region-based moments to characterize the shape of an object with a set of parameters that are invariant to geometric changes. The other ones are the well-known correlation and least-squares matching techniques. A test pair of digital aerial images covering an urban area and having a nearly 42-μm pixel size is used for the research experimentation. A set of distinct points are picked in the left image and considered matching techniques are applied to determine their conjugates in the right image. Least-squares matching is implemented using each of the nearest-neighbor, bilinear and cubic resampling techniques to interpolate pixel values of computed conjugate locations. The matched pairs of points in the test pair are employed in a relative orientation process to assess the global matching precision. According to the attained results, a global matching precision of a fraction of a pixel were yielded by the three matching techniques. However, the least-squares matching technique using bicubic interpolation provided the best precision figures. |
