M7Image forensicsMicroscopyLayer 1 (Deterministic)

Contrast Consistency

Checks whether local contrast varies smoothly across a micrograph. A real acquisition changes contrast gradually with position, so a map of local contrast is spatially smooth; a figure pasted from different sources shows abrupt contrast jumps, loses spatial coherence, and splits into two contrast populations.

Technical description

Tiles the grayscale image into 16 by 16 blocks and takes each block's standard deviation as its local contrast, forming a contrast grid. It reads three signals: the lag-1 spatial autocorrelation of the grid (each block vs its right and bottom neighbour), scored max(0, (0.5 - r)/0.5) * 2.0; the bimodality coefficient BC = (g1^2 + 1)/(g2 + 3(n-1)^2/((n-2)(n-3))), adding 1.5 when above 5/9 to flag two contrast populations far more selectively than raw kurtosis; and abrupt jumps, neighbour pairs differing by more than three times the contrast spread, scored min(2.0, 20 * jump_ratio). The contributions sum to a maximum of 5.0.

How it works

Layer 1 (deterministic): builds a grid of per-block local contrast and scores three departures from a smooth single-source field: a low lag-1 spatial autocorrelation (below 0.5), a bimodality coefficient above 5/9, and abrupt neighbour-to-neighbour contrast jumps. The worst jump locations are reported with bounding boxes. The contributions are summed and capped at 5.0.

Why this matters

A spliced figure is hard to see because the eye forgives a clean cut, but image formation leaves a measurable trace at the seam. A foreign region carries the contrast and lighting of its own origin, so detecting where the local response stops being consistent exposes the composite without a reference. A single acquisition produces a smooth, spatially correlated, single-population contrast field; a composite produces jumps, low spatial coherence, and two populations. Adjusting contrast to obscure or join regions is recognised misconduct, which makes the cue directly relevant to research integrity.

Score thresholds

0-1: Local contrast varies smoothly and forms one population, consistent with a single acquisition
2-3: Reduced spatial coherence, a bimodal distribution, or some abrupt jumps
4-5: Strong contrast discontinuities, low spatial coherence, and two contrast populations, consistent with a composite of different sources

Limitations

Local contrast is specimen-driven, so a genuine image with sharply bounded structures can show real contrast steps and a bimodal distribution without manipulation. The jump threshold is a multiple of the contrast spread, so when that spread is large the threshold rises and individual seams may go uncounted, leaving the autocorrelation and bimodality signals to carry the detection. The 16-pixel block bounds resolution, and uneven illumination lowers the autocorrelation legitimately. Whether the tonal response and local histogram stay consistent is indicator I9, and the spatial uniformity of the sensor noise level is indicator M6; M7 reads the spatial structure of the local-contrast field and corroborates those.