Abstract
Projective distortions due to terrain reliefs considerably deteriorate image correlation in stereo matching. This paper covers the derivation of the relationship between the ensemble mean of sample crosscovariances, usually used as a matching criterion, and projective distortions, in terms of correlation window size and a power spectral distribution of image densities, and further discusses their effects on correlation mathematically. In addition a simple correlation test is appended to verify the validity of the discussion. The main resulth sbtained through the discussion are as follows:
•A correlation window works as a high-pass filter cutting off the lower frequency components of spectrum, wavelength of which is shorter than a half of window size.
•Projective distortions distort the higher frequency components and deteriorate correlation, Hence, for obtaining precise matching performance, recursive correlation procedure accompanied with step-by-step corrections of distortions will be essensial in ony practical matching strategy.
•At the corners of the film exposed by a wide-angle camera, it may occur that a projective distortion caused by more than 45° slopes exceeds an allowable limit.
