WEBSL: A Support-Grain Representation for Web Images

WEBSL is a simple web image representation pipeline based on a regular support-grain structure rather than a final pixel grid taken as the primary object.

In WEBSL, an image is modeled as a regular support of grains with an implicit geometry, a canonical row-by-row reading order, and a state attached to each grain. The final displayed image is then obtained by projection to the chosen output size.

The current implementation focuses on 2D web images and evaluates this representation against direct web image exports. The tested core includes:

• a fixed support-grain structure,
• implicit geometry through row-major traversal,
• simple local prediction on the support,
• final projection to the target output size,
• separate handling of opaque images and RGBA cutout images.

Across the current tests, the most effective support coding is generally based on simple vertical prediction (pred_up), and in some cases on mixed local prediction (pred_avg_lu). The system is intentionally lightweight and does not rely on a heavy specialized codec architecture.

The main interest of WEBSL is that competitive results already appear with a very simple representation model. In several tested cases, especially on reduced output sizes and RGBA cutout images, WEBSL produces lighter served files than direct WebP exports while preserving high structural similarity.

Current observations from the tested prototype are:

• strong behavior on structured web images,
• strong behavior on RGBA cutout images,
• competitive behavior on reduced photo outputs,
• remaining sensitivity of chroma on some full-size opaque natural images.

This deposit includes the main Python scripts used to reproduce the current tests, along with a technical memo describing the representation, the benchmark logic, and the main experimental results.