This paper presents predicted performance for area-based image matching where two images of a non-planar object model differ by a general perspective geometric transformation. The study shows there exists a window size that will maximize or minimize certain performance parameters for a given perspective distortion and object planarity variance. The analysis also indicates that for a given perspective distortion where pitch angle is the only parameter, many performance criterion have an optimum window size if the object model is allowed to vary. The performance measures examined are expected peak value, peak-to-sidelobe ratio (PSR), probability of acquisition (P<SUB>CA</SUB>), and image registration error covariance. Window adaptation based on precomputed metrics is applied to extend distortion tolerance. Statistically consistent image sets are geometrically transformed by a general perspective spatial mapping using statistically consistent independent non- planar object models with arbitrary generalized autocorrelation functions. The two images are then registered through an image matching technique, the defining functions analyzed and limitations on the amount of perspective viewpoint change of an imaging system in an aerial tactical arena are given while still allowing proper image correspondence. Monte Carlo simulation verification of theoretical predictions and results are extended to a variety of common area-based image matching techniques.
This paper presents an investigation of area-based image correlation under full perspective geometric distortion. Image matching techniques in the presence of geometric distortion modeled by affine transformations have been examined thoroughly in earlier work while this research investigates the effects of full-perspective distortion on spatial domain area-based matching techniques. With area- based techniques, there are two factors that heavily affect correlation accuracy; first, the amount of signal variation within the target window must be sufficient to provide detectable similarity of imagery and second, the amount of geometric distortion within the window must be small enough not to inhibit matching. Signal variation is increased by increasing the target window size while the effects of perspective geometric distortion are minimized through smaller target windows. Window dimensionality is appropriately adapted to accommodate these two conflicting effects. Window adaptation based on precomputed metrics is applied to extend distortion toleration. The image sets are derived from a tactical scenario and are geometrically transformed through planar assumptions by a nonaffine spatial mapping. The two images are then registered through different correlation techniques, the defining functions analyzed and limitations on the amount of perspective viewpoint change of an imaging system in an aerial tactical arena are given while still allowing proper image correspondence.