Contour-Based Chain-Code Serialization for Lossless Compression of Voxelized 3D Objects

Voxel representations provide a simple way to represent three-dimensional objects as binary occupancy signals, but dense voxel grids and direct sparse encodings remain costly at medium and high resolutions. This paper addresses the gap between conventional dense-grid, octree, and point-cloud-codec representations and deterministic contour-first source serialization for exact binary voxel occupancy. We propose a contour-based chain-code serialization that decomposes a voxel grid into two-dimensional slices, extracts foreground components and holes, encodes their contours using F4, 3OT, and F8 variants, and separates contour symbols from positional metadata before applying general-purpose lossless compression. The method is evaluated on 3983 ModelNet40-derived voxelized objects across 40 classes and resolutions N = 8, 16, 32, 64, 128, 256, and 512, using the X-axis for the main evaluation. It is compared against OCC1, BINVOX, breadth-first octree masks, and geometry-only G-PCC. The proposed streams are not competitive at N = 8, where zstd-compressed octree masks achieve the best mean bpv. From N = 16 onward, however, the best proposed stream outperforms the strongest evaluated baseline, with gains increasing from 20.93% at N = 16 to 84.37% at N = 512. The best proposed configuration is zstd + 3OT at N = 8 and N = 16, while zstd + F8 dominates from N = 32 through N = 512. Entropy, ablation, timing, memory, and validation analyses further show that the advantage comes from the interaction between contour-aware source serialization and backend compression, rather than from the backend compressor alone.