What Is Blur Detection?

How software measures sharpness

A computer cannot squint at a photo, so it measures proxies. Sharp images contain dense, high-contrast edges; blurred images contain fewer and softer ones. By computing edge density and contrast across a frame, and especially across regions that matter, like detected faces, software can score sharpness numerically and rank an entire card in one pass.

Not all blur is the same

• Missed focus: the plane of focus landed behind or in front of the subject. Usually a reject.
• Motion blur: subject or camera moved during exposure. Often a reject, sometimes the whole point of the shot.
• Shallow depth of field: intentionally blurred backgrounds around a sharp subject. Not a flaw at all.

Good blur detection has to separate these, which is why face-aware analysis matters: a dreamy f/1.4 portrait is mostly blur by area, but the eyes are sharp, and that is what counts.

An example

On a dark reception dance floor you shoot 300 frames at slow-ish shutter speeds. Twenty are unusable smears, forty are borderline, and the rest are fine. Blur detection can clear the obvious smears and the obviously sharp frames automatically, leaving you a short borderline stack instead of a 300-frame slog. That is the difference between soft frames slipping into a gallery and catching them before delivery.

Limitations

Sharpness is measurable; intent is not. A deliberate slow-shutter dance-floor pan scores like a mistake. Heavy film-style grain, fog, and backlit haze can also depress sharpness scores on frames you love. Any tool that hard-rejects on a blur score alone will eventually discard art, so borderline scores belong in a review queue, not a trash can.

How Cull AI Studio handles this

Cull AI Studio analyzes sharpness, edge density, brightness and contrast, and blown or black frames locally on your computer, alongside face and eye detection, so blur is judged where it matters most, on people. Every decision carries a confidence score, and frames below the uncertainty bar are routed to REVIEW instead of being silently rejected. If you routinely keep intentional motion blur, your corrections are recorded and, over time, preference learning adapts the model to your taste rather than a generic definition of sharp.