THE PREDICTION OF CROSS-CORRELATION ACCURACY IN DIGITAL STEREO-PAIR IMAGES

Abstract

The behavior of digital cross-correlation algorithms as applied to image matching problems is examined in terms of the relationship between measureable image properties and algorithm characteristics. Statistical image quality measures are developed which could be employed in a preprocessor to predict the performance of automatic stereo-compilation equipment. The measures include a quantity derived from the Cramer-Rao lower bound on the variance of any unbiased parameter estimator, various contrast measures such as variance, contrast modulation, and median absolute deviation, and a stationarity detector related to the variance gradient. These measures are based on image and correlator models which describe the behavior of correlation processors under conditions of low image contrast or signal-to-noise ratio, geometric distortion, and image non-stationarity. Computer simulations using synthetic imagery were performed to verify the various models, and indicate the potential for the use of image quality measures in the predicting of correlation behavior. Implications of the models in terms of correlation processor design and implementation are discussed.