We build a deep neural network based on quasi-conformal theories, called QC-net, to obtain diffeomorphic registration maps between corresponding data. QC-net take the landmarks in the to-be-registered images as input and output the registration mapping between them.