Retrospective heel effect correction in conventional radiography

Three methods for retrospective correction of intensity inhomogeneities in diagnostic digital X-ray images are presented. The first method fits a theoretical heel effect model to the background intensities. The second method optimizes the parameters of a chosen image formation model to maximize the likelihood of the background pixels of the acquired image, under the assumption that their intensity values are Gaussian distributed. Additive and multiplicative linear image formation models with different degrees of freedom are fitted onto a subset of the background pixels discarding the pixels with the largest residual errors to eliminate outliers. The third method minimizes the entropy of the background and diagnostic region of the image, excluding the collimation areas. Six image formation models, using a combination of smoothly varying basis functions, are quantitatively evaluated and compared with the first method on a number of reference and real diagnostic X-ray images.