Alignment of adjacent picture frames captured by a CLSM

Area of Science:

• Microscopy
• Image Processing
• Computational Biology

Background:

• Image mosaicking is crucial for imaging objects exceeding sensor field of view, common in microscopy and remote sensing.
• Aligning image tiles requires registering overlapped regions, facing challenges like limited sample size, window shape anisotropy, and intensity scale variations.

Purpose of the Study:

• To evaluate the robustness of similarity-based registration methods for image mosaicking.
• To compare sum of absolute valued differences (SAVD), normalized correlation coefficient (NCC), and mutual information function (MIF) under intensity scale distortions.

Main Methods:

• Registration of adjacent picture tiles using overlapped border areas.
• Comparative analysis of SAVD, NCC, and MIF for image tile alignment.
• Extension of pilot experiments to three-dimensional (3-D) image stacks.

Main Results:

• Mutual Information Function (MIF) demonstrated the highest robustness against intensity scale distortions in most cases.
• Sum of Absolute Valued Differences (SAVD) and Normalized Correlation Coefficient (NCC) showed varying degrees of susceptibility to intensity variations.
• MIF required the most computational power compared to SAVD and NCC.

Conclusions:

• MIF is the most robust method for image mosaicking, particularly when dealing with significant intensity scale variations.
• The choice of registration method should balance robustness with computational cost based on tile content (texture, contrast, intensity distortions).
• This research provides insights for selecting cost-effective image registration strategies in microscopy and 3-D visualization.