We propose a view-dependent compression algorithm for 3D binary voxel models, based on the skeleton representation. An input model is first partitioned into several segments. The segmentation increases the number of skeleton points, requiring more bits than the original model. Therefore, we develop a remodeling algorithm, which reduces the number of skeleton points in each segment effectively without any loss of information. Each segment is then compressed with a context-based arithmetic coder. Before the transmission, the encoder is informed of the viewing parameters via feedback channel. Then, the encoder transmits visible segments in detail, while cutting off invisible segments. Simulation results show that the proposed algorithm requires much less bitrate than the conventional skeleton-based algorithm, while exhibiting faithful image quality.