JPEG Pleno vs 2D JPEG Point Cloud Compression

Plenoptic point clouds represent complex three-dimensional scenes by combining 3D spatial coordinates with directional, view-dependent light and color properties. While legacy 2D JPEG was built to compress planar grids of pixels using spatial frequency transforms, the JPEG Pleno framework is purpose-built to handle high-dimensional visual data natively. This article explains the fundamental differences between legacy 2D JPEG and JPEG Pleno, detailing how both standards handle data representations, spatial structures, multidimensional redundancies, and compression efficiency.

Structural Data Representation

Legacy 2D JPEG operates exclusively on two-dimensional pixel arrays structured on a fixed Euclidean grid. When applied to 3D point cloud data, 2D JPEG requires the point cloud to be projected, unwrapped, or sliced into 2D surfaces (such as depth maps or projection textures). This projection breaks spatial contiguity, creates empty non-sampled regions, and causes severe seam distortions along geometric boundaries.

In contrast, JPEG Pleno treats point cloud data as a native 3D structure. It directly processes sparse, unstructured sets of points defined by coordinates \((X, Y, Z)\) alongside their associated plenoptic attributes (such as color, reflectance, normals, and angular radiance). By avoiding arbitrary 2D flattening, JPEG Pleno preserves natural 3D geometric relationships and avoids projection-induced artifacts.

Geometry and Attribute Separation

2D JPEG relies on color-space transformations (such as RGB to \(\text{YCbCr}\)) followed by 2D Discrete Cosine Transform (DCT) across uniform \(8 \times 8\) pixel blocks. It cannot distinguish between structural geometry and surface appearance.

JPEG Pleno decouples the compression pipeline into two distinct but interconnected components:

  1. Geometry Compression: The spatial coordinates are typically partitioned using hierarchical data structures such as octrees, voxel grids, or predictive trees. This efficiently encodes where points exist in 3D space by eliminating volumetric emptiness.
  2. Attribute and Directional Compression: Once geometry is established, plenoptic attributes are encoded conditionally based on spatial location. JPEG Pleno utilizes tools such as the Graph Fourier Transform (GFT), Region-Adaptive Hierarchical Transforms (RAHT), or localized directional wavelets that adapt to the irregular surfaces of the point cloud rather than rigid square blocks.

Exploiting Plenoptic and Angular Redundancies

Plenoptic point clouds capture view-dependent effects, meaning a single point may exhibit different colors or light intensities depending on the viewing angle.

If legacy 2D JPEG is used across multiple perspectives, it must treat each viewpoint as a separate static frame. This approach redundantly stores identical underlying geometry and ignores the smooth angular variations of reflected light across perspectives.

JPEG Pleno explicitly models the higher-dimensional plenoptic function. It exploits both spatial correlation (points near each other often share properties) and angular correlation (the light reflected from a point changes continuously across viewing angles). By using spherical harmonics, radiance functions, or multi-view prediction mechanisms, JPEG Pleno encodes these continuous lighting variations with minimal bit overhead.

Coding Efficiency and Artifact Profiles

When compressing dense volumetric data to low bitrates, legacy 2D JPEG introduces classic 2D block boundary artifacts and ringing around high-contrast edges. When projected back into 3D space, these 2D artifacts manifest as severe geometric displacement, floating "phantom" points, and torn surfaces.

JPEG Pleno provides scalable, progressive decoding tailored for 3D interactions. As the bitrate decreases, JPEG Pleno degrades point resolution or attribute fidelity gracefully throughout the volume, preventing the sharp geometric tearing typical of 2D methods. It achieves dramatically higher rate-distortion performance by ensuring that bits are allocated directly to perceptually relevant geometric features and angular light fields.