Automatic microscopy image stitching based on geometry features

In this paper, we present an automatic image stitching method for Virtual Microscopy (VM). The method aims to integrate a number of fields of view into a large image which captures the whole view of a biomedical specimen. The method aligns the images based on matching of local features. To assure matching accuracy, feature points are efficiently detected using integral images. In addition, the local features are matched by incorporating geometry features. The mismatched feature points are filtered out by RANSAC algorithm, which provides robust estimation of the transformation between images. The experiments on stitching the microscopy images demonstrate that our method is accurate and efficient.