Mechanism

The structure signature is a gradient-only fingerprint derived from the orientation distribution and edge density of a content artifact, computed without reference to mean luminance or background fill values. It is an optional component of the full unique identifier (UID) record, sitting alongside the global nine-dimensional variance vector and the quadrant fingerprints, and it serves a specialized matching function: recognizing logos, icons, and graphically sparse artifacts across background color changes and flat-fill variations. Because the signature reads only from spatial gradient information, it stays stable across the kinds of transformations that alter pixel values without altering edge structure, while diverging when the underlying edge structure itself changes.

The gradient-only structure vector module produces a 21-dimensional structure vector. A 16-bin gradient histogram is computed over the full canonical image, with each bin accumulating gradient magnitude weighted by edge strength across all interior pixels. The histogram is then canonicalized by rotating it to place the maximum-weight bin at index zero, yielding a representation that does not depend on the artifact's absolute orientation. A multi-threshold edge density stage computes edge density at three gradient magnitude thresholds, 0.05, 0.10, and 0.15, producing three scalar values that characterize the structural content of the image across fine, medium, and coarse edge-strength levels. A gradient histogram moment computation derives the variance and the peak of the normalized 16-bin histogram, giving compact statistics of how gradient energy is distributed.

The structure hash output concatenates the 16-bin histogram values, the three edge density scalars, the histogram variance, and the histogram peak into the 21-dimensional structure vector. That vector is hashed using the STRUCTURE_COARSE_SCALES scheme at quantization scales of 24, 32, and 40, producing a 256-bit structure hash that is then extended to a 320-bit structure identifier. The use of coarse quantization scales is what lets the signature absorb format conversion and flat-fill variation: small numerical perturbations in the structure vector quantize to the same value, so they do not move the hash.

Suppressing Luminance and Background Fill

The defining property of the structure signature is what it deliberately does not encode. The global nine-dimensional variance vector and the quadrant fingerprints both read from intensity statistics, so they respond to mean luminance and to the fill behind a graphic. The structure signature is derived exclusively from spatial gradient information, which suppresses sensitivity to mean luminance and background fill. Recoloring the canvas behind a logo, or swapping a flat-color background, changes intensity values but leaves the edge structure intact, and so it leaves the structure signature unchanged.

This is why the disclosure positions the structure signature as the matcher for graphically sparse content. A photograph carries rich variance everywhere, and the global variance vector characterizes it well. A logo or icon is mostly flat fill punctuated by a small number of strong edges, so most of its intensity signal is background that should be ignored. By reading only the orientation distribution and the multi-threshold edge density, the structure signature concentrates on the part of a sparse artifact that actually carries its identity.

Rotation Canonicalization

The 16-bin gradient orientation histogram is canonicalized by rotating it so that the maximum-weight bin sits at index zero. This is a cyclic shift of the histogram rather than a rotation of the image, and it makes the structure vector independent of the artifact's dominant orientation. Two copies of the same graphic that differ in orientation produce histograms that are cyclic shifts of one another, and after canonicalization they align to the same starting bin, so they hash to the same structure identifier.

This canonicalization mirrors the orientation handling used elsewhere in the identity pipeline, where the Z-axis gradient histogram is rotated to place its dominant angular bin at index zero, and where the canonical normalization stage rotates the image when its dominant gradient orientation exceeds approximately 0.1 radians before quadrant extraction. The structure signature applies the same principle directly to its own 16-bin histogram.

From Structure Identifier to Anchor ID

The 320-bit structure identifier does not stand alone. An anchor ID module derives a non-authoritative 128-bit anchor identifier from the 320-bit structure identifier together with the 320-bit constellation identifier, by applying the multi-segment combiner to their concatenation and retaining the first two 64-bit segments. The result is a compact routing handle for content objects whose structure and geometric composition are meant to match across background substitution, format conversion, and partial cropping.

The anchor ID is described as non-authoritative: it is a routing convenience, not the canonical identity of the artifact. The canonical identity remains the variance-derived UID, while the anchor ID gives downstream matching workflows a short, structure-aware key to group artifacts that are likely to be the same graphic seen against different backgrounds or in different formats. This is the form of routing the disclosure associates with logo recognition, reverse image search, and derivative content attribution.

The Constellation Signature Companion

The structure signature is one of a pair of optional, specialized signatures. Its companion, the constellation signature, supports matching of artifacts that share distinctive spatial compositions across cropping, scale change, and partial occlusion. Where the structure signature reads edge orientation and density over the whole canonical image, the constellation signature is built from a small set of high-salience anchor points and the geometric relationships among them.

Saliency hotspot detection identifies up to five spatially distributed high-salience anchor points within the canonical image. Saliency is computed over a 12x12 coarse grid by scoring each cell as the sum of twice its intra-cell intensity variance and its mean local gradient density, with cells sorted by descending score under a minimum inter-cell separation of one grid unit so that the hotspots spread out rather than cluster on a single prominent feature. For each hotspot serving as a focal anchor, its nearest neighbors are identified by Euclidean distance, the inter-anchor distances are normalized by the maximum pairwise inter-hotspot distance and quantized into eight bins, and the inter-anchor angles are measured in the range zero to pi radians and quantized into twelve bins of fifteen degrees each. A descriptor string encoding the hotspot grid coordinates, quantized distances, and quantized angles is hashed into a per-anchor micro-constellation hash, and the per-anchor hashes are sorted lexicographically before being concatenated and hashed, eliminating any dependence on the order in which hotspots were detected. Normalizing distances by the maximum pairwise distance makes the constellation scale-invariant, and measuring angles in the range zero to pi makes it rotation-invariant. At least three hotspots are required; if fewer than three are detected, the constellation signature is not emitted and its field in the UID record is null. The structure signature and the constellation signature address complementary failure modes: background and luminance change for the former, geometric framing change for the latter.

Role Within the UID Record

The structure and constellation signatures are optional components of the full UID record, supplementing the global variance vector and the four quadrant fingerprints rather than replacing them. The base UID is a 320-bit identifier constructed from the global variance vector and the rotation-invariantly sorted quadrant hashes. The structure signature adds a parallel 320-bit identifier specialized for sparse, background-variable content, and the constellation signature adds another for composition-driven matching.

Because these signatures are computed deterministically from the artifact's own structure, any conforming node can recompute them and compare. The structure signature, in particular, gives a matcher that survives the transformations that defeat intensity-based fingerprints on sparse graphics, while still being a fixed-length hash that participates in the same anchor registration and resolution machinery as the rest of the UID record.

Prior Art Contrast

Existing perceptual hashing systems, including difference hash, average hash, and perceptual hash algorithms, produce low-dimensional binary signatures from downsampled image representations. Per the disclosure's background, they lack multi-scale structural analysis, directional orientation decomposition, and spatial sub-region identity, and their fixed-width binary outputs cannot encode the gradient structure required for distributed anchor assignment. They also bind to luminance distributions, which makes them sensitive to exactly the background and flat-fill changes the structure signature is built to absorb.

Watermarking and metadata tagging approaches embed identity signals in the content stream or a sidecar record. Watermarks are removable through transcoding, cropping, or generative reconstruction, and metadata records are decoupled from content structure and require persistent external storage. The structure signature, by contrast, is computed from the artifact's intrinsic edge structure: there is no embedded signal to strip, and an alteration large enough to change the signature is, by construction, an alteration of the artifact's edge structure itself.

Disclosure Scope

The structure signature, comprising the gradient-only 21-dimensional structure vector built from a rotation-canonicalized 16-bin gradient orientation histogram, edge density values at the three gradient magnitude thresholds of 0.05, 0.10, and 0.15, and the histogram variance and peak statistics, hashed under the STRUCTURE_COARSE_SCALES scheme at quantization scales of 24, 32, and 40 to a 256-bit structure hash extended to a 320-bit structure identifier, and its derivation of a non-authoritative 128-bit anchor ID from the structure and constellation identifiers, is disclosed in PCT International Application No. PCT/US26/28630. This article describes that disclosed mechanism.

The scope extends to the use of the structure signature as an optional component of the UID record for recognizing logos, icons, and graphically sparse artifacts across background color changes, flat-fill variations, and format conversions that alter pixel values without altering edge structure, and to its pairing with the constellation signature for composition-driven matching across cropping, scale change, and partial occlusion. It does not extend to fingerprints that bind to mean luminance or background fill rather than to gradient structure, nor to embedded watermark or sidecar metadata signals that can be stripped without altering the artifact's edge structure.