Registration revisited

J Neurosci Methods. 1993 Jun;48(1-2):1-13. doi: 10.1016/s0165-0270(05)80002-0.

Abstract

Image registration is important for numerous imaging applications such as three-dimensional reconstruction, multimodality correlations, image averaging and subtraction. Methods used for image registration are based upon either the shape and form of the image pairs or their densitometric relationships. This paper describes the algorithms used for five different registration methods; frequency domain cross-correlation, spatial domain cross-correlation, principal axes/center of mass, fiducials and manual. These methods were compared in terms of their accuracy, efficiency and application with several different data types including different species and modalities. The underlying mathematical bases for each also are presented and compared. The results of the comparisons showed that image quality influenced the behavior of all methods. Images of the blockface provide an excellent reference for subsequent registration. These results also suggest that the statistical performance of various methods is not a reliable metric when distant and different images are registered. Visual comparisons by image overlap and pixel differencing illustrated that some methods are more prone to rotational error than others, especially when repeated pairwise registrations were computed along the rostral/caudal axis.

Publication types

  • Research Support, U.S. Gov't, Non-P.H.S.
  • Research Support, U.S. Gov't, P.H.S.

MeSH terms

  • Algorithms
  • Animals
  • Autoradiography
  • Brain / anatomy & histology*
  • Cats
  • Fourier Analysis
  • Humans
  • Image Processing, Computer-Assisted*
  • Mathematics
  • Rats
  • Software
  • Tissue Fixation