A global method based on thin-plate splines for correction of geometric distortion: an application to fluoroscopic images.

摘要

Quantitative analysis of biomedical images needs a careful correction of geometric distortion. To avoid the discontinuities of the local correction techniques and achieve good accuracy in the presence of global and local distortion, a novel global correction technique based on thin-plate splines is proposed. The technique approximates the grid points by a thin plate minimizing the weighted sum of the bending energy and the mean squared residual errors. The method proposed is compared with three traditional correction techniques: two local and one global. One local technique is linear and takes into account translation, rotation, and scaling, the other is nonlinear and includes skewing. The global technique is based on a two-dimensional polynomial model. Computer-based simulations and experimental tests on fluoroscopic images were carried out. The local techniques were sensitive to both sigmoidal and radial distortion. The polynomial and thin-plate splines global techniques were found sensitive only to sigmoidal distortion and to radial distortion, respectively. The two global techniques showed better performances with respect to any local on synthetic and real images. Where the distortion is predominantly radial or high computational efficiency is required, the polynomial global correction technique should be preferred. Where the distortion has a local nature or is predominantly sigmoidal, the thin-plate splines global correction technique should be chosen.