Description of Registration Methods for Carlos R. Castro-Pareja, Raj Shekhar

We implemented five different histogram binning algorithms and tested their influence on the accuracy of Mutual Information-based image registration. The five algorithms are:

1) Linear histogram binning (Methods 1-7)

2) Histogram preservation (Methods 8-14)

3) Maximal entropy histogram equalization (Methods 15-21)

4) Quantization error minimization (Methods 22-28)

5) Quantization error square minimization (Methods 29-35)

Each algorithm was tested for output histogram sizes of 256, 128, 64, 32, 16, 8 and 4. Algorithms (1) and (2) have been previously described in the literature. Algorithm (3) is an adaptive version of the standard histogram equalization algorithm. The adaptive version tries to obtain the flattest histogram with the given number of bins. Algorithms (4) and (5) are focused in minimizing the increase in quantization error due to the reduction of histogram bins. The binning methods were tested by maximizing Normalized Mutual Information using the Simplex algorithm. As a preprocessing step, slices were added to make the data sets isotropic, and the background was segmented and set to zero.


1) C. Castro-Pareja and R. Shekhar, "Adaptive reduction of intensity levels in 3D images for mutual information-based registration." to be presented at SPIE's Annual Symposium on Medical Imaging, Feb. 2005.

2) J. Camp and R. Robb, "A novel binning method for improved accuracy and speed of volume image coregistration using normalized mutual information." Proceedings of SPIE, vol. 3661, pp. 24-31, 1999.

3) J. Lin et al, "The affection of grey levels on mutual information based medical image registration." Proceedings of the 26th Annual International Conference of the IEEE EMBS, pp. 1747-1750, 2004.

4) Y. Zhu and S. Cochoff, "Influence of implementation parameters on registration of MR and SPECT brain images by maximization of mutual information." The Journal of Nuclear Medicine, vol. 43, no. 2, pp. 160-166, Feb. 2002.

Return to previous page