Mechanism

The multi-axis variance vector extraction pipeline accepts a digital content artifact and produces a nine-dimensional variance vector organized into three structural axes, designated X, Y, and Z. Each axis encodes a distinct and complementary aspect of the artifact's internal composition, and the three axes together represent a mutation-sensitive characterization of the artifact suitable for unique identifier derivation, slope-band assignment, and similarity comparison. The vector is derived from the internal structure of the content rather than from a static identifier such as a uniform resource locator, a cryptographic hash pointer, or a file-system path.

For raster image artifacts, the pipeline first converts the input to a normalized grayscale floating-point representation through a grayscale conversion module that applies a perceptual luminance weighting of approximately 0.299 for the red channel, 0.587 for the green channel, and 0.114 for the blue channel, yielding intensity values normalized to the range zero to one. For non-image artifacts such as audio waveforms, textual documents, and binary objects, modality-specific normalization produces a comparable bounded two-dimensional scalar field from which the same multi-scale variance, gradient, and density analyses are computed.

Multi-Scale Variance Flow

The multi-scale variance flow analyzer subdivides the normalized scalar field into three nested grid resolutions: a coarse grid of 8x8 cells covering the full artifact extent, a medium grid of 16x16 cells, and a fine grid of 32x32 cells. For each cell at each grid resolution the analyzer computes a variance-based proxy defined as the variance of pixel intensity values within the cell. The per-cell variance values for each grid resolution are aggregated to produce a mean variance and a standard deviation of variance for each scale level, yielding six scalar values that collectively characterize variance distribution across the spatial hierarchy of the artifact.

These six scale-level statistics, together with a gradient histogram and edge-density and global-variance measures, supply the raw quantities from which the three axis vectors are constructed. The nested-grid design is what makes the representation multi-scale: it records not only how much variance an artifact carries but how that variance redistributes itself as spatial resolution increases.

X-Axis: Energy Behavior

The X-axis energy behavior vector encodes the artifact's energy distribution trend across scale. Three components are computed: the slope of mean variance from the coarse to the fine grid resolution, representing the rate at which variance concentrates or disperses as spatial resolution increases; the curvature of mean variance at the medium scale, representing deviation from a linear energy trend; and the asymptotic fine-scale energy value, representing the limiting energy concentration of the artifact at maximum examined resolution. These three components together capture whether an artifact is characterized by broad, diffuse energy distribution or by fine, spatially concentrated energy structures.

Y-Axis: Frequency Compaction

The Y-axis frequency compaction vector encodes the behavior of variance dispersion across scales. Three components are computed: the rate of change of variance standard deviation from coarse to fine resolution, indicating whether the artifact's variance distribution becomes more or less spatially uniform at finer scales; the spread factor between the maximum and minimum per-scale variance standard deviation, characterizing the range of distributional variability; and the variance floor convergence value, measuring the proximity of scale-level mean variance to the global variance of the full artifact, indicating how closely multi-scale behavior approximates the global statistical profile. These components are sensitive to frequency content, texture regularity, and compositional heterogeneity.

Z-Axis: Structural Phase Persistence

The Z-axis structural phase persistence vector encodes orientation and structural stability. The gradient histogram module computes a histogram of gradient magnitudes across eight angular bins spanning zero to pi radians, aggregated over all interior pixels of the normalized scalar field. An orientation canonicalization stage rotates the histogram so that the dominant angular bin is positioned at index zero, producing a rotation-invariant representation of the artifact's edge orientation distribution.

Two components of the Z-axis vector are derived from this histogram: the horizontal-vertical orientation bias, computed as the mean weight of horizontal bins minus the mean weight of vertical bins; and the diagonal-axial bias, computed as the mean weight of diagonal bins minus the mean weight of axial bins. The third Z-axis component is a stability coefficient computed as one minus the absolute difference between the artifact's edge density and its normalized global variance, measuring the structural coherence between edge density and statistical variance. The edge density module computes the fraction of interior pixels whose gradient magnitude exceeds a threshold of 0.1, and the global variance module computes the full-image variance as a proxy for information density; these two scalar values supplement the multi-scale variance flow and gradient histogram in constructing the Z-axis vector.

From Nine Dimensions to a Unique Identifier

The resulting nine-dimensional vector, comprising three components each along the X, Y, and Z axes, constitutes the primary variance representation of the artifact. The representation is designed to be stable under format conversion, resolution rescaling within a defined canonical size, and lossy compression at moderate quality levels, while varying predictably with semantic-content-altering transformations such as object insertion, removal, significant cropping, style transfer, or compositional remixing.

The nine-dimensional vector is combined with quadrant sub-region vectors and optionally supplemented by structure and constellation signatures to produce the final unique identifier. The combined twenty-seven-dimensional vector used for hashing is formed as a weighted linear combination of the X, Y, and Z axes organized into three dominant triads: an X-dominant triad weighting X at 0.6 and Y and Z each at 0.2; a Y-dominant triad weighting Y at 0.5, X at 0.3, and Z at 0.2; and a Z-dominant triad weighting Z at 0.5, X at 0.3, and Y at 0.2. This weighting scheme ensures that no single axis completely dominates the unique identifier and that the fingerprint is sensitive to structural perturbations along each behavioral dimension. The unique identifier encodes a position within a continuous variance space such that cosine similarity between two unique identifiers is directly computable without decoding a fixed binary digest.

Multi-Modal Operation

Because the extraction pipeline operates on any normalized scalar field, it extends to content modalities beyond raster images. For audio waveforms the normalization computes a time-frequency representation using a short-time Fourier transform, maps it to the mel frequency scale, and produces a normalized mel-spectrogram that serves as the scalar field input; the multi-scale variance flow, gradient histogram, and edge density computations then operate over the spectrogram matrix in the same manner as image processing. For textual documents the document is mapped to a two-dimensional token frequency scalar field, with byte-level variance over fixed-width content windows blended in as a supplementary signal. For video the pipeline operates at two levels: each frame is processed as a raster image, and a clip-level temporal delta vector is derived from the cosine similarity between consecutive frame variance vectors and itself processed through the extraction pipeline. For binary objects the byte sequence is reshaped into a two-dimensional scalar field and processed identically. All modality paths converge on the shared multi-axis variance vector extraction stage, producing a vector directly comparable to image-domain vectors through the same cosine similarity operator.

Prior-Art Posture

Existing perceptual hashing systems, including difference hash, average hash, and perceptual hash algorithms, produce low-dimensional binary signatures from downsampled image representations. They lack multi-scale structural analysis, directional orientation decomposition, spatial sub-region identity, and a continuously scaled similarity score suitable for slope-based banding or lineage tracing, and their fixed-width binary outputs cannot encode the gradient structure, variance flow, or compositional geometry required for distributed anchor assignment or derivative attribution. The disclosed variance vector instead emits a continuous, multi-axis representation over which cosine similarity is directly computable, enabling graded similarity and slope-band routing rather than binary near-duplicate detection.

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 variance vector is derived from the content's own internal structure rather than injected into it, so two artifacts whose bytes differ but whose underlying structure is preserved produce nearby vectors under cosine similarity. Ordinary content-handling operations such as format conversion, resolution rescaling within the canonical size, and moderate lossy compression therefore do not sever the identity established at registration time, while variance-shifting mutations diverge predictably.

Disclosure Scope

The mechanism described here, the extraction of a nine-dimensional multi-axis variance vector organized into an X-axis energy behavior vector, a Y-axis frequency compaction vector, and a Z-axis structural phase persistence vector, computed over nested coarse, medium, and fine grid resolutions of a normalized scalar field, together with the gradient histogram, edge density, and global variance measures, the weighted axis-dominant triad combination used for unique identifier derivation, the cosine similarity comparison over the resulting vectors, and the modality-specific normalization paths for image, audio, text, video, and binary artifacts, is disclosed in PCT International Application No. PCT/US26/28630. This article describes that disclosed mechanism using the specification's own terminology. The scope extends to additional normalization procedures that produce a bounded two-dimensional scalar field from a digital artifact of any type, provided the field serves as input to the shared multi-axis variance vector extraction pipeline.