Geometrical Patterns Based Cross-scale Image Registration for AFM and Optical Microscopy

Image registration is usually used to transform different images from different sensors, times, depths or viewpoints into one coordinate. This paper proposes a novel template matching algorithm based on geometrical patterns, which is proved to be effective in matching AFM and optical images. As for the procedures of the proposed algorithm, firstly the resolution of the AFM image is calibrated as a priori knowledge for image processing in the next few steps. Then, traditional image processing methods, including filtering, binarization and contour-searching, are applied to the raw images sequentially. Centroids of every pattern made up of edges are connected to extract the geometrical feature information of the image. In the end, a designed assessment function is applied to the whole image to calculate each point's matching possibility. Experiments show that, compared with traditional methods, the proposed algorithm provides much more reliable matching results in AFM-optical template matching, offering an effective way for cross-scale images registration.