A comparison of similarity measures for digital subtraction radiography.

Subtraction is useful in detecting small changes in sequentially acquired radiographs. Even if the imaging geometry is constant, radiographs must be registered after their digitization. To compare different algorithms for image registration and to register digital X-rays themselves, various similarity measures have been proposed. This study compares eight mathematical similarity standards using 172 radiographs acquired in different, but exactly known projection. Whenever the computation time is a critical factor, e.g. registering images using methods similar to correlation techniques, the entropy of the subtraction image's histogram function (EHDI) is found to be the best similarity standard. If not, e.g. comparative assessing different image registration techniques, the cross covariance coefficient (CCC) is appropriate.