local classical computer vision

Wholeness Lens

A deployable Google App Engine prototype for testing whether Alexander-like grown form, positive space, centers, and levels of scale can be seen by explicit geometry rather than by a rented remote general-purpose model.

Read the first progress report

Analyze an image

This first version does not claim to “recognize beauty.” It creates a visible, debuggable diagram of the image: regions, edges, candidate centers, scale levels, and positive-space proxies. It is meant to be wrong in interesting ways, so the next phase can calibrate it against human judgment.

What is actually stored?

The app stores a downscaled JPEG, the computed metrics, SVG diagnostic diagrams, and the generated progress-report HTML in Datastore. That keeps the first deployment inside the three-tier architecture you asked for.

What is actually computed?

The server computes Otsu thresholding, connected components, edge density, quasi-convexity of foreground and background regions, a component-size scale ladder, local symmetry proxies, and center candidates.

Why this phase?

Alexander says positive space is not leftover background: both figure and ground should be well-shaped. So the system must look at both object regions and the spaces between them.

The first deployable architecture: browser, Flask analyzer, Datastore.