GEOMETRIC PRECISION IN NOISE-FREE DIGITAL IMAGES

A basis for determining the best possible precision for the position of a given object in a digital image is developed. The approach is general, being independent of object-form and providing a perfomance bound on all position estimation schemes. The central concepts in the development are the locale and total variation of the image function. An easily computed bound for the smallest attainable RMS position estimation error is developed. The best possible position estimate is given by the centroids of the locales. The analysis of geometric precision using locales shows how object position-uncertainty depends on the object's position. Visual inspection of the easily generated locale pattern gives a qualitative impression of the spatial distribution of geometric precision. The analytic techniques developed provide avenues for both qualitative and quantitative evaluation of target forms and position estimation schemes in terms of their precision mensuration capabilities.