Biomedical image mosaicing: An optimized multiscale approach

In typical mosaicing or blending algorithms, it is usually assumed that the to-be-combined images are equally informative, and each component image would be processed in a similar manner. However, because of the photobleaching effect, the fluorescence intensity of confocal microscope image may degenerate; therefore, the overlapping regions will become asymmetrically informative. We formulate the problem of mosaicing such images as a multiscale optimization approach. The optimal blending coefficients can thus be obtained by a quadratic programming with constraints enforced by the biological requirements. We also demonstrate the efficacy on several confocal microscope fluorescence images as well as the EM images and compare the mosaicing results with those derived by other methods.