This paper presents a novel compression scheme for 3D dynamic meshes. The proposed method mainly uses a segmentation approach which exploits both of temporal and spatial dependencies. The temporal coherence of the geometry component is determined using the heat diffusion properties. The motion of the resulting regions is accurately described by 3D affine transforms. These transforms are computed at the first frame to match the subsequent ones. The temporal prediction errors are then encoded. Comparative coding test, for 3D dynamic meshes, were conducted to evaluate the coding efficiency of the proposed compression scheme. Simulations demonstrate that the rate/distortion results are competitive when compared to the state of the art.