How to Detect If an Image Was Cropped or Resized

Someone submitted a photo that’s supposed to be “the original, untouched.” Or you’re checking whether an image posted elsewhere was cropped from a source you own. Either way, the question is the same: was this image altered from a reference, and if so, how?

Why file properties alone don’t answer this

Checking the image’s dimensions and file size feels like it should settle the question, but it doesn’t:

Dimensions changing proves resizing — but matching dimensions doesn’t prove nothing happened. An image can be cropped and then upscaled back to the original width/height, which restores the numbers while destroying real content at the edges.
File size is dominated by compression, not content. Re-saving an untouched image at a lower JPEG quality can shrink the file by 80% with zero cropping or resizing involved.
EXIF metadata is trivially stripped or fabricated. A “camera model” or “date taken” field proves nothing about whether pixels were altered — see spot edited or tampered images for why metadata-only checks fail.

The only reliable method is comparing pixel content directly against a known original.

Detecting a crop

A cropped image is a subset of the original’s content — same subject matter, smaller field of view. The tells:

Composition looks “tighter” than expected — a portrait with less headroom, a product shot missing its usual background margin.
The aspect ratio doesn’t match common camera/export ratios (e.g., a suspicious 1:1 crop from what should be a 4:3 or 16:9 original).
Edges cut through content that would normally have breathing room — part of a logo, a hand, or a caption clipped exactly at the frame boundary.

If you have the suspected original, DiffALL’s image comparison tool will show a strong mismatch when comparing full images (different dimensions, different framing) — but the real test is cropping the original to match the candidate’s apparent framing and then comparing that region. A high similarity score on the matched region, combined with content missing outside it, confirms a crop.

Detecting a resize (without a crop)

A pure resize keeps the same framing but changes resolution — usually to strip identifying detail, reduce file size, or fit a platform’s constraints. Signs:

Resolution is suspiciously round (exactly 1080px wide) rather than a native camera/export resolution.
Fine detail looks slightly soft — resizing (especially downscaling then someone else re-upscaling) loses high-frequency detail that never comes back.
Compare against the source at the same displayed size: resize the reference down to the candidate’s resolution before running the comparison, since comparing a 4000px original against a 1080px copy directly will show differences from the resolution mismatch itself, not from any real alteration.

The comparison workflow

Get both images to the same dimensions first. Resize the higher-resolution one down to match (never upscale the smaller one — that manufactures detail that isn’t really there and produces misleading similarity scores).
Run them through image comparison. A near-100% similarity score with a clean, quiet diff heatmap means the content is genuinely unaltered aside from resolution.
Read the heatmap for localized differences. A crop shows up as content present in one image and missing in the other at the boundary; a targeted edit (logo removed, object cloned out) shows up as a bright, localized cluster in an otherwise quiet map.
Check compression artifacts separately — see detecting JPEG compression artifacts if the heatmap shows a faint, even pattern across the whole image rather than a localized change; that’s usually re-compression, not editing.

What a clean vs. altered result looks like

Result	Interpretation
~99–100% similarity, uniformly quiet heatmap	Same image, at most re-compressed — no crop or content edit
High similarity but content missing at one or more edges	Cropped
Moderate similarity with fine detail softer throughout	Resized (likely downscaled at some point)
High overall similarity with one bright, localized spot	A targeted edit, not a global crop/resize — investigate that region specifically
Low similarity even after matching dimensions	Not the same source image, or heavily altered

Checklist

[ ] Match both images to the same resolution before comparing (downscale the larger, never upscale the smaller).
[ ] Check dimensions and aspect ratio for signs of cropping before running a pixel comparison.
[ ] Read the diff heatmap: localized bright spots mean editing, uniform faint noise means re-compression.
[ ] Don’t trust EXIF metadata as evidence either way — it’s easily stripped or forged.
[ ] If claiming theft/unauthorized use, keep the comparison result and both source files as your evidence trail.

Dimensions and file size describe the container. Only a pixel-level comparison against the real original tells you what actually happened to the content — run the comparison instead of guessing from metadata.