Background Tiles
Detects backgrounds that were built rather than captured. A real western blot background carries film or sensor noise and varies smoothly, so it is never perfectly flat and never repeats. A background that is artificially flat, lacking the noise a real exposure leaves, or that repeats a tile pattern from copy-paste editing, is fabricated. The indicator masks out the bands, measures the local noise of the remaining background, and looks for periodic repetition in its autocorrelation. It works on the pixels alone, with no model.
Technical description
W3 is a deterministic, generator-agnostic screen for constructed backgrounds. A genuine blot background is the unexposed or lightly exposed region of film or a digital sensor, so it carries a characteristic noise floor and gentle large-scale variation, and it does not repeat. Two fabrication modes break this: filling the background with a flat colour, which removes the noise, and pasting a patch repeatedly to cover an area, which introduces periodic structure. W3 detects the bands and masks them out, then reads two properties of the remaining background: its local noise, measured as the median within-block standard deviation, and its periodicity, measured from the autocorrelation of a central crop. An artificially flat background, with near-zero local noise, and a periodic background, with repeating autocorrelation peaks, each raise the score. The image must be at least 64 pixels on a side and have enough background to analyse.
How it works
The bands are detected and their bounding boxes are removed from a background mask. The background is tiled into a sixteen-by-sixteen grid, and for each block with enough valid pixels two quantities are formed: the block mean, whose coefficient of variation across blocks is reported, and the within-block standard deviation. The median of the block standard deviations is the local-noise measure. A real background, even one that looks uniform, has a local noise well above zero from film grain or sensor noise; a flat synthetic fill has almost none. When the median local noise falls below 1.5 on the zero-to-255 scale, the background is flagged as artificially flat, which adds 2.5 to the score. This local-noise test is used in place of the coefficient of variation of the block means, because that coefficient is also small for a genuinely noisy background whose mean is roughly constant and would misflag it.
For periodicity, a central crop is mean-subtracted and its two-dimensional autocorrelation is computed with a wrap boundary and normalized. The central region is zeroed, and the number of points exceeding half the peak is counted; a count at or above three indicates a repeating tile pattern and adds up to 2.5 to the score, scaled by the count. The contributions are summed and capped at 5.0. Findings describe an artificially flat background and a periodic pattern. The metadata records the background coefficient of variation, the local noise, and the periodic-peak count.
Score thresholds
| Score | Meaning |
|---|---|
| 0 to 1 | The background carries natural noise and shows no repetition. |
| 2 to 3 | The background is artificially flat, or shows some periodic structure. |
| 4 to 5 | The background is both flat and periodic, or strongly tiled. Consistent with a constructed background. |
Why this matters
The background of a blot is as informative as the bands, because it records the physics of exposure. A captured background carries a noise floor from the film or sensor, and the absence of that noise is one of the clearest signs that a region was painted in rather than imaged, the same noise-inconsistency principle that underlies blind image forensics, where a region lacking the host image's noise statistics is exposed as foreign [1]. Repetition is the other tell: copy-pasting a background patch to cover a deleted band or to extend a field leaves a periodic signature that the eye misses but autocorrelation reveals, and surveys of biomedical figures document how often backgrounds are manipulated in exactly this way [2]. The ethical guidance on scientific images is explicit that erasing or constructing background content is misconduct, and that detection must look beyond the bands to the spaces between them [3]. By reading both the local noise and the periodicity of the masked background, W3 distinguishes a real exposure from a fabricated backdrop.
Limitations
The screen depends on detecting and masking the bands, so missed bands leave band pixels in the background and detected-but-oversized masks remove real background. The local-noise test assumes a real background is noisy; an image that was heavily denoised or strongly compressed loses its noise floor and can read as artificially flat without being painted. The periodicity measure counts points above a threshold in the autocorrelation rather than isolating discrete peaks, so a smoothly varying background, such as one with strong uneven illumination, has a broad central correlation lobe that can inflate the count and suggest periodicity that is not there. The wrap boundary in the autocorrelation can also create edge artefacts. The thresholds are directional rather than exact. Duplicated bands are the separate indicator W1, and sharp intensity transitions around individual bands are indicator W4, so W3 stays on the flatness and repetition of the background as a whole.
Theoretical background
W3 rests on what a real exposure records in its empty regions. Film and digital sensors register a stochastic noise floor everywhere, independent of whether a band is present, so a genuine background is a textured field with a nonzero local variance and no exact repetition, because noise is by definition non-repeating. A flat fill replaces that field with a constant, driving the local variance to zero, and a copied patch replaces it with a periodic tiling, introducing exact repetition at the tile spacing. The median within-block standard deviation reads the first violation, distinguishing a textured background from a flat one even when both have a constant mean, which is why it succeeds where the variation of the block means fails. The autocorrelation reads the second, because a periodic field correlates strongly with shifted copies of itself at the period, producing secondary peaks that a non-repeating field lacks. Reading noise and periodicity together tests whether the background was captured by a detector or assembled by an editor.
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