The first phase of development was characterized by developing the initial core product, a completely new algorithm for fully analysing any bitmap image.

This required two important problems to be solved: first mapping out the core features, then mapping out the color transitions and assembling this information into a vector-based graph.

The first product built on the core algorithm: exporting the analysed structure of a photograph as vector geometry, one vector definition per color channel. This works on photographic input, not just logos and line art, and exposes the underlying structure of the image rather than an approximated path soup.

Because the renderer can re-project those vectors at any resolution, faithful upscaling comes with the same milestone: edges stay edges at any scale, and nothing is invented.

Noise removal and sharpening on top of the faithful re-projection, developed together, completing the image enhancement product. The process is deterministic, and the structure extracted from the output can be audited against the structure of the input.

This targets organisations that need to demonstrate that enhancement did not alter the original image: legal and forensic, archival, and medical imaging, where generative enhancers are structurally disqualified.

A multi-resolution representation in which each coarser layer is a defined structural refinement of the layer below, so the representation is invertible and verifiable. This is the substrate for advanced image editing and, potentially, for AI applications.

This phase contains open research on pattern taxonomy and the layered data model, and is presented as such.

Editing a photograph by its structure: move an edge and the fill follows, select a region by what it is rather than by its pixels. This is enabled by the hierarchical representation.

Rather than building an editor, the hierarchical representation plus an SDK is licensed to the vendors who already own the editing surface.

Using the vector data as direct input for neural networks, both for efficiency (denser, higher signal-to-noise input than pixels) and for control (constraining structure directly in classification and generation).

This is a research direction rather than a product. Exploration can start once direct vectorization is available, by feeding per-channel vectors to conventional networks and measuring the effect.