Alignment of adjacent picture frames captured by a CLSM

Mosaicking a picture from its adjacent parts (called picture frames or tiles) is encountered in different fields of research and technology, e.g., photogrammetry, remote sensing, microscopy, etc. It is applied whenever the object of investigation is too large for the field of view of the sensor, e.g. a microscope. We describe mosaicking with a confocal laser-scanning microscope (CLSM) Bio-Rad, MRC 600 (U.K.). Aligning neighboring picture tiles was accomplished by registering the overlapped border areas of these tiles. Such registration procedures are constrained by: 1) the limited size of the registered samples (windows); 2) anisotropy of the form of the windows (usually narrow rectangles); and 3) the content of the windows, including changes of their intensity scale. Focusing on the latter problem, methods of registration were discussed and the robustness of the following three similarity based methods was studied with regard to the distortions of the intensity scales of the tiles to be registered: 1) the sum of absolute valued differences (SAVD); 2) normalized correlation coefficient (NCC); and 3) the mutual information function (MIF). Pilot experiments were extended to three-dimensional (3-D) stacks of pictures encountered in the framework of 3-D object rendering and visualization. MIF was found in most cases to be the most robust; however, it also demanded the most computational power. It is discussed how to choose a cost-effective method of the registration with regard to the content (texture, contrast, intensity scale distortion) of the tiles.