ResAIKit
Research Integrity Toolkit
Back to the encyclopedia
W4Image forensicsWestern BlotLayer 1 (Deterministic)

Background Inconsistency

Looks at the background immediately around each band for signs the band was inserted or removed. A pasted band brings its own local background, leaving a seam or halo where it meets the host and a ring whose intensity differs from the rings of the other bands. The indicator measures each band's ring and flags intensity outliers or a sharp step across any of the band's four edges.

Technical description

Detects the bands and extracts the ring of background pixels around each. It computes the ring mean and, across all rings, a z-score; a ring more than two standard deviations from the global mean is an intensity outlier. It also computes the sharpest intensity step across each of the band's FOUR edges (row or column just inside vs just outside), taking the maximum, so a halo on the sides or bottom is caught, not only the top. A band that is a ring outlier, has a sharp edge step, or both, is flagged; the outlier and sharp-edge counts set the score (0, 1.5, 3.0, up to 5.0).

How it works

Layer 1 (deterministic): for each band, extracts the surrounding background ring, flags a ring whose mean is more than two standard deviations from the other rings, and flags a sharp intensity step across any of the band's four edges. The total outlier and sharp-edge count sets the score: none scores 0, one 1.5, two 3.0, more scale up to 5.0; a band that is both a ring outlier and has a sharp edge is marked critical.

Why this matters

Inserting or deleting a band is a direct way to falsify a blot, and the give-away is usually the surroundings, not the band. A pasted band carries the local background of its source, so the strip around it differs in statistics from the host image, the inserted-region inconsistency that blind forensics exploits. The seam at the paste boundary is the complementary cue, a sharp halo or step that even illumination does not produce. Band insertion and deletion are common and often leave exactly these background traces, which is why the background around bands must be examined, not just the bands.

Score thresholds

0-1
The rings around the bands are consistent and their edges transition naturally
2-3
One or two bands have an anomalous background ring or a sharp edge seam
4-5
Several bands show inconsistent rings or seams, consistent with inserted or removed bands

Limitations

The screen depends on detecting the bands and on having at least two with measurable rings; with only two, the ring-mean outlier test is weak, since two values cannot establish an outlier, and detection rests on the edge step. The ring-mean outlier uses the mean and standard deviation, which a single outlier inflates, so among a few bands one anomalous ring can mask another. The edge step is large for any band (dark band on light background), so the threshold sits above the normal band-edge contrast; a 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 is partly missed. Duplicated bands are indicator W1 and a wholly constructed or tiled background is indicator W3; W4 stays on the local background around each band.

References

  1. Mahdian B, Saic S. (2009). Using noise inconsistencies for blind image forensics. Image and Vision Computing 27(10):1497-1503
  2. Bik EM, Casadevall A, Fang FC. (2016). The Prevalence of Inappropriate Image Duplication in Biomedical Research Publications. mBio 7(3):e00809-16
  3. Cromey DW. (2010). Avoiding twisted pixels: ethical guidelines for the appropriate use and manipulation of scientific digital images. Science and Engineering Ethics 16(4):639-667