Image stitching — the process of amalgamation of separate image fragments to form a complete representation of the entire scene — might become quite a challenging problem in the presence of non-additive noises and/or brightness variability artifacts. An
additional degree of complication may further be inflicted in situations when one is dealing with large size data, as it is usually the case in tiling microscopy. To overcome such difficulties, a novel approach to the problem of image stitching is proposed here. In the heart of the proposed
solution is Wallis filtering, which is a standard tool of image processing used for adaptive contrast adjustment and local image normalization. More importantly, Wallis filtering allows representing a given image in terms of its normalized version and associated local statistics. Subsequently,
we show that stitching the output of the Wallis filter is a much simpler and much more stable task as compared to stitching the images in their original domain. The proposed method has an additional advantage of being computational efficient, which is particularly important in tiling microscopy,
where a typical height/width of data images is on the order of tens of thousands of pixels.