LMS registration of rigid transformations

Methods are investigated to improve the registration of images corrupted by rigid displacements using the Least Mean Square (LMS) algorithm. Results show that LMS adaptive registration (LMSAR) is effective for small translational displacements, but fails for large translational displacements where the correlation between the rotation data sets is too weak. In an attempt to improve the robustness of LMSAR, various methods are investigated and a modified LMSAR technique is introduced. The modified LMSAR is compared with the Fourier Shift Theorem (FST) for clean and noisy images where the LMSAR accuracy is similar to the FST for clean image. As expected the LMSAR appears more susceptible to noise, but the LMSAR offers reduced computation over the FST for circumstances involving searches over a large angular range.