Background Inconsistency
Looks at the background immediately around each band for signs that the band was inserted or removed. A band pasted from another image brings its own local background, leaving a seam or halo where it meets the host, and a ring whose intensity differs from the rings around the other bands. The indicator measures the background ring of every band and flags rings that are intensity outliers or that show a sharp step across any of the band's four edges. It works on the pixels alone, with no model.
Technical description
W4 is a deterministic, generator-agnostic screen for inserted or deleted bands, detected from their surroundings rather than their content. When a band is pasted in, two traces appear: the strip of background just around it, its ring, takes on the intensity of the source image rather than the host, so it stands out from the rings of the genuine bands, and the boundary where the paste meets the host shows a sharp intensity step, a seam or halo, on one or more edges. W4 detects the bands, extracts the ring of background pixels around each, and computes the ring's mean intensity and the sharpest intensity step across each of the band's four edges. A band whose ring mean is an outlier relative to the others, or that has a sharp edge step on any side, is flagged. The image must be at least 32 pixels on a side and contain at least two bands with measurable rings.
How it works
Each band's ring is the band of background pixels within a fixed width around it, with the band's own pixels excluded. For every ring the mean intensity is computed, and across all rings the global mean and standard deviation give each ring a z-score; a ring whose mean lies more than two standard deviations from the global mean is an intensity outlier, the signature of a band whose background came from a different image.
The edge step is computed on all four sides of the band, not only the top. For each edge the row or column just inside the band is compared to the one just outside, and the largest absolute difference across that edge is taken; the maximum over the four edges is the band's edge gradient. Checking every edge matters because a pasted band can be haloed on its sides or bottom as easily as its top, and a top-only check would miss those. A band whose edge gradient exceeds a threshold has a sharp seam.
A band is flagged when it is a ring-mean outlier, when it has a sharp edge step, or both. The total number of outliers and sharp-edge bands sets the score: none scores 0, one scores 1.5, two scores 3.0, and more scale up to 5.0. Findings name the band and the reason, marked critical when a band is both a ring outlier and has a sharp edge. The metadata records the band count and the ring-outlier count.
Score thresholds
| Score | Meaning |
|---|---|
| 0 to 1 | The rings around the bands are consistent and their edges transition naturally. |
| 2 to 3 | One or two bands have an anomalous background ring or a sharp edge seam. |
| 4 to 5 | Several bands show inconsistent rings or seams. Consistent with inserted or removed bands. |
Why this matters
Inserting or deleting a band is among the most direct ways to falsify a western blot, and the give-away is usually not the band itself but its surroundings. A pasted band carries the local background of its source, so the strip around it has different statistics from the host image, which is the inserted-region inconsistency that blind forensics exploits: a region whose local intensity or noise does not match the rest of the image is exposed as foreign even when the splice is invisible to the eye [1]. The seam at the paste boundary is the complementary cue, a sharp halo or step that smooth optics and even illumination do not produce. Surveys of biomedical figures document how common band insertion and deletion are, often with exactly these background traces [2], and the ethical guidance is explicit that adding or erasing bands is misconduct and that the background around bands must be examined, not just the bands [3]. By reading the ring intensity and the edge steps on all four sides, W4 catches the surroundings that betray an inserted or removed band.
Limitations
The screen depends on detecting the bands and on having at least two with measurable rings; with only two bands the ring-mean outlier test is weak, since two values cannot establish an outlier, and detection then rests on the edge-step cue. The ring-mean outlier uses the mean and standard deviation, which a single strong outlier inflates, so among three or four bands one anomalous ring can mask another. The edge step is large for any band, because a band is dark against a light background, so the threshold is set above the normal band-edge contrast; a genuinely high-contrast band can approach it, and a softly pasted band can fall below it. The ring width is fixed, so a halo wider or narrower than the ring is partly missed. The thresholds are directional rather than exact. Duplicated bands are indicator W1 and a wholly constructed or tiled background is indicator W3, so W4 stays on the local background around each individual band.
Theoretical background
W4 rests on the locality of background statistics in a real exposure. In a genuine blot the background is one continuous field, so the strip around every band shares the same intensity level, noise, and gradual variation, and the transition from background to band, while sharp, is the natural edge of a diffuse mark. Pasting a band from elsewhere violates both properties at once: the imported background has the level and texture of its source, making the ring an outlier among the host's rings, and the join between imported and host background is a discontinuity that no single exposure would contain, appearing as a step on one or more edges. Deleting a band and filling the gap leaves the same kind of seam. The ring-mean comparison measures the first violation across bands, and the four-edge step measures the second locally, so a band that fails either is surrounded by background that did not come from the same image as the rest. Reading the surroundings turns the continuity of a real background into a test of whether each band belongs.
References
- Mahdian B, Saic S. Using noise inconsistencies for blind image forensics. Image and Vision Computing. 2009;27(10):1497-1503. DOI: 10.1016/j.imavis.2009.02.001
- Bik EM, Casadevall A, Fang FC. The Prevalence of Inappropriate Image Duplication in Biomedical Research Publications. mBio. 2016;7(3):e00809-16. DOI: 10.1128/mBio.00809-16
- Cromey DW. Avoiding twisted pixels: ethical guidelines for the appropriate use and manipulation of scientific digital images. Science and Engineering Ethics. 2010;16(4):639-667. DOI: 10.1007/s11948-010-9201-y