Robust microscope image stitching using multiple zooming levels

Microscopes usually have limited field of view so it is necessary to stitch the raw microscopy images into a virtual micorscopy image which combines the advantages of high details resolution brought by high magnification and a large field of view which gives the experience of the full image of the entire region of interests. Compared with regular panorama stitching or scanning stitching, two significant differences of microscope stitching is the amount of images need to be stitched is much larger and the features of microscopy images are quite different from landscape or texts so it is hard to make use of existing distortion correction methods. Advanced microscopes use motor platform to manipulate the subject to capture a grid structure layouts image set with their positions, and then use these data in the stitching progress to get a seamless stitched image of the region of interests. Compared with existing microscopy image stitching methods relying on motor platform, we introduce a multi-zooming-level microscope stitching method to stitch images captured through portable microscopes which does not require the position information of each images. Instead of using high zooming rate images only, we capture enough low zooming rate images of the same area and stitch them into a texture skeleton. For each high zooming rate image, we match it with all the low zooming rate image to find its rough position to use in the later stitching progress. In this way, the computational complexity is reduced significantly and the texture skeleton also helps deal with the distortion caused by accumulating errors.