Unified Video-Based Spatially Scalable Coding for Dynamic 3-D Meshes

This study proposes a spatially scalable coding structure for dynamic 3D meshes, enabling adaptive services across diverse network conditions and device capabilities. The proposed method jointly encodes the geometry and texture of dynamic meshes using a multilayer unified video representation, providing spatial scalability via 2D scalable codecs. This representation is constructed by jointly packing displacement videos and texture maps across multiple layers. Specifically, a multilayer displacement video is generated by decomposing the displacements into hierarchical layers using a level-of-detail structure derived from a lifting transform, where each layer captures progressively finer geometric details. Multiresolution texture maps are generated using the proposed Smoothed Push–Pull-based method, which adaptively applies downsampling filters by distinguishing between occupied pixels with valid texture and non-occupied pixels without it. The resulting multilayer displacement video and texture maps are spatially stacked to form a single multilayer video, compressed using a conventional 2D scalable codec. This enables simultaneous spatial scalability of both geometry and color employing a single video encoder–decoder pair. Experimental results indicate that the proposed method achieves Bjøntegaard Delta (BD)-rate improvements of 37.0%, 37.0%, and 35.5% in point-cloud-based D1, D2, and luma Peak Signal-to-Noise Ratio (PSNR) under the All-Intra condition, and improvements of 35.1%, 35.0%, and 35.6% under the Low-Delay condition in the three-layer configuration, compared with those of the simulcast approach. This advance enables adaptive 3D streaming by allowing a single scalable bitstream to support devices with varying capabilities, highlighting its practical relevance for immersive media delivery.