Three-dimensional image registration of CT and MRI of human knees facilitates the construction of high-quality models providing the features of both modalities.This multi-modal image registration problem is non-trivial since the great difference between CT and MRI images makes standard voxel-based registration inefficient.Research advances in shape correspondence algorithms motivate us to convert the problem of image registration to a shape correspondence one.Since the correct alignment of bone structures are of most concern in our study,we present a solution for converting tibia and femur,the two largest bone structures in these images,to meshes.Our solution consists of four steps: 1) Background intensities are masked out; 2) The contrast of bone tissues is enhanced adopting different approaches for CT and MRI images; 3) Tibia and femur are extracted as features; 4) Surfaces are reconstructed from tibia and femur point sets.At the end of the paper,we present the resulting meshes derived from CT and MRI images and demonstrate the efficiency of our solution.