We present here a novel method for whole brain magnetic resonance (MR) image registration that explicitly penalizes the mismatch of cortical and subcortical regions by simultaneously utilizing anatomic segmentation information from multiple cortical and subcortical structures, represented as volumetric images, with given T1-weighted MR image for registration. The registration is computed via variational optimization in the space of smooth velocity fields in the large deformation diffeomorphic metric matching (LDDMM) framework. We tested our method using a set of 10 manually labeled brains, and found quantitatively that subcortical and cortical alignment is improved over traditional single-channel MRI registration. We use this new method to generate a volumetric and cortical surface-based population average. The average grayscale image is found to be crisp, and allows the reconstruction and labeling of the cortical surface.