Abstract
The local technique to correct for geometric image distortion in the x‐ray imaging chain has been implemented and compared with the global correction technique reported previously. Three local correction models were considered, all applied locally within each quadrilateral cell bounded by four adjacent points on a distorted 1.0 cm spaced rectilinear calibration grid. Model 1 uses bilinear interpolation between the correction vectors derived at the four vertices. In Models 2 and 3 a polynomial mapping from distorted to undistorted positions is applied, with Model 2 representing a linear six‐parameter correction and Model 3 a nonlinear eight‐parameter correction. Both techniques have been analyzed in depth concerning their accuracy. To this end, experimental (35 mm cine film and online video) and computer simulated distorted images of the calibration grid were used. Both radial and sigmoidal distortions were considered in the simulations. The main results are summarized as follows. The accuracy of the local technique depends upon the position within each quadrilateral cell and upon the local direction of the distortion gradient, contrary to the global technique. The accuracy of the local technique decreases with increasing radial as well as sigmoidal distortion. For the global technique only the latter dependency applies. The relationship of the accuracy with the pixel size is different for all models. Finally, the local technique is more sensitive to noise than the global technique. The reproducibility of the correction methods has been evaluated by means of imaging the calibration grid in various orientations with respect to the input screen of the image intensifier. The local technique turns out to reproduce a factor of 4–10 worse than the global technique. In summary, Model 1 is the worst method in all respects, whereas Model 3 is slightly better than Model 2. However, the global technique is to be preferred because it outperforms the local technique on all relevant issues.

This publication has 21 references indexed in Scilit: