Visual model

The image is a map you can compute inside.

Each cell below stands in for an addressable region. The teal rectangle represents a requested viewport. In the real codec, the decoder resolves only blocks intersecting that viewport and reconstructs the exact crop.

Logical address

0 → 3 → 1 → 2 describes ancestry: quadrant within quadrant, not a verbose coordinate stored beside every pixel.

Physical representation

The addressed block may use raw, XOR, Sub, Paeth, color-decorrelated, or planar reversible transforms before compression.

Visibility

A crop asks the address map which regions intersect. Only their physical payloads and necessary index material need to be touched.

Live proof

Open the codec lab to race an actual crop against the conventional tile-pyramid baseline.