Since the spatial field non-linearity of gradient coils translates into image geometric distortion in MRI, in many applications, such as cardiac function analysis and interventional MR-based device tracking/guidance, where the precise geometric information is needed, the presence of geometric image distortion can not be simply ignored. To address the concern for geometric image distortion, we have developed and validated a general and efficient numerical technique for parameterizing the global image distortion for a bi-planar gradient coil as well as accomplishing image restoration as a post-imaging processing. This image correction methodology is based on a global distortion coordinate mapping function which can be systematically defined directly from the gradient field non-linearity in 3-dimension (3D) of a given gradient coil. The image correction was carried out in two steps: (1) map each pixel of the corrected image representation onto its distorted image according to the distortion mapping; (2) interpolate the pixel intensity in the distorted image using its neighboring points via a bi-linear interpolation procedure. The results showed clearly that the distortion correction method was robust in term of the capability of reducing image geometric distortion dramatically. Also it is shown that the magnetic field non-linearity or the image distortion of a typical bi-planar gradient coil can be adequately parameterized using a finite Taylor series expansion based on its design parameters. Furthermore, this image distortion correction method is very efficient in practice for performing 3D correction for any image orientation since a compact parameterized field expression contains non-zero terms.