REAL-SPACE BASED MULTI-VIEW IMAGE FUSION

Some of the novel fluorescence microscopy methods (SPIM- Single Plane Illumination Microscope [1]) permit rotation of the sample around an axis normal to the optical axis during the imaging. Such multi-view set of three-dimensional images acquired along different imaging directions relative to the sample provide more information about the sample than any single one of them. The main reasons are i) anisotropic optical resolution of single lens systems and ii) sample opacity (i.e. single view exposes and records different parts of an opaque sample unevenly) The most convenient way to take advantage of the information in such multi-view sets is to fuse the whole dataset into a single three dimensional image. Recently, a method of fusing SPIM multi-view sets was proposed and it provides an outstanding resolution, especially with large specimens [2]. However, multiple steps of the algorithm were done in Fourier domain, which makes it time consuming and computationally demanding. Multi-view fusion algorithms consist of i) extensive preprocessing (scaling, rotating, filtering), ii) image alignment and iii) final fusion, i.e. combination of the information from different views. We achieved a vastly improved performance by replacing all Fourier-spac