We present an automatic subpixel registration algorithm that minimizes
the mean square intensity difference between a reference and a test d
ata set, which can be either images (two-dimensional) or volumes (thre
e-dimensional), It uses an explicit spline representation of the image
s in conjunction with spline processing, and is based on a coarse-to-f
ine iterative strategy (pyramid approach). The minimization is perform
ed according to a new variation (ML) of the Marquardt-Levenberg algor
ithm for nonlinear least-square optimization, The geometric deformatio
n model is a global three-dimensional (3-D) affine transformation that
can be optionally restricted to rigid-body motion (rotation and trans
lation), combined with isometric scaling, It also includes an optional
adjustment of image contrast differences, We obtain excellent results
for the registration of intramodality positron emission tomography (P
ET) and functional magnetic resonance imaging (fMRI) data, We conclude
that the multiresolution refinement strategy is more robust than a co
mparable single-stage method, being less likely to be trapped into a f
alse local optimum, In addition, our improved version of the Marquardt
-Levenberg algorithm is faster.