Limitations of Image Detection: Where ELA Can Mislead
Understand when ELA results are unreliable, common false positives, and how to reduce misinterpretation in real‑world checks.
Limitations of Image Detection: Where ELA Can Mislead
ELA (Error Level Analysis) is a powerful forensic tool, but it isn’t a universal lie detector. Knowing its limits helps you avoid false conclusions and interpret results more responsibly.
Common Sources of False Positives
1) Social‑platform recompression
Symptom: Images saved from WeChat, Weibo, Instagram, etc. often light up everywhere in ELA.
Why: Platforms recompress uploads to save bandwidth, which resets compression patterns across the entire image.
Tip: Use the original file when possible. If not, lower sensitivity and look for local anomalies rather than global brightness.
2) Screenshots and screen recordings
Symptom: ELA appears uniform, even when the content is visibly manipulated.
Why: A screenshot is effectively a full re‑capture; all pixels share the same compression history.
Tip: ELA is not reliable for screenshots. Use metadata or reverse image search instead.
3) Multiple saves or format conversions
Symptom: Messy, noisy ELA results that are hard to interpret.
Why: Repeated saves add layers of compression artifacts, which can drown out the original edit signal.
Tip: Ask for the earliest version of the file. Heavily re‑saved files have low forensic value.
4) Heavy filters or global adjustments
Symptom: The whole image appears bright or uneven.
Why: Global filters change the entire pixel field, which behaves like a full‑image edit.
Tip: Separate "global edits" from "local manipulations" when interpreting results.
Cases Where ELA Is Not Reliable
AI‑generated images
AI outputs are created from scratch, not edited from an original. The compression pattern is uniform, so ELA often looks clean even if the image is synthetic.
Better approach: Use dedicated AI‑detection tools and visual realism checks (hands, reflections, geometry).
Proving authenticity
ELA can raise suspicion, but it cannot prove an image is fake.
Compression artifacts might be caused by:
- camera processing
- legitimate edits (exposure, color correction)
- format conversions
Treat ELA as evidence, not a verdict.
Recovering the original
ELA can indicate where changes happened, but it can’t reconstruct what the original looked like.
How to Reduce Misinterpretation
- Build a baseline with known‑clean images
- Look for local anomalies (isolated bright patches in smooth areas)
- Cross‑validate with metadata, source tracing, and reverse search
ELA is a sharp tool, but only when used in the right scenario. Use it as a fast first pass, then validate with other methods.