Robust Automatic Montaging of Adaptive Optics Flood Illumination Retinal Images

Adaptive optics (AO) flood illumination camera acquires retinal images with a limited field of view, which can be extended by image alignment into one wide field of view montage image. The image alignment into a montage requires efficient and accurate image registration. Since manual registration is demanding and disadvantageous, automatic registration is a beneficial improvement. We propose the first fully automated AO retinal image montage procedure. Here, we present three novel fully automated registration methods, which are based on two established image processing approaches. The first method utilizes scale invariant feature transform (SIFT) in combination with specific image preprocessing. The second method uses the phase correlation (PC) approach and the last method is a connection of PC and SIFT (PC-SIFT) algorithm. In total, 200 images acquired from the left and right eyes of 10 subjects were used for creating the wide field-of-view montage images and compared with manual montaging. The automated image montage was successfully achieved. Alignment accuracy evaluated by normalized mutual information metric showed that the PC-SIFT approach established the most accurate results, these are higher than manual montaging. Therefore, the AO montaging registration methods are able to achieve promising results in accuracy and time demand in comparison with manual montaging. Hence, the latter can be replaced by those fully automated procedures.