Automatic registration of large set of microscopic images using high-level features

In this paper, we present a novel method for automatic registration of large set of microscopic images by automatically match high-level region features via finding cyclic structures in a matching graph. The use of high-level features (e.g., regions, landmarks, objects) significantly reduced the computation and provides accurate initialization, which further allows fast convergence of the maximum mutual information algorithm. The scheme is a universal one as it works for other types of high-level features and the matching process is very computationally efficient. We have applied our method in 3-D reconstruction of a unique human cochlear sample and are also applying it to two other set of large microscopic images