Blur detection

What it means

Blur detection measures sharpness. Classic methods look at how strong the edges are in a frame: a crisp image has many sharp transitions between light and dark, while a blurry one has soft gradients. A common quick metric is the variance of the Laplacian, an edge filter whose output drops when an image goes soft.

Modern tools also use trained models that tell different kinds of blur apart: motion blur from fast movement, misfocus on cameras with autofocus, and the hazy smear of a fogged or wet lens. Each has a different cause and a different fix.

How it works in practice

In automated editing, blur detection is a filter. Before scoring moments, the system discards frames and clips that are too soft to use, so a beautiful jump is not represented by the one frame where the camera was shaking hardest.

Action cameras use fixed focus, so most blur in their footage comes from movement and low light. Slow shutter speeds in dusk or indoor light smear every frame of a bumpy ride. A brighter scene or a faster shutter reduces it at the source.

Droplets and fog are the other big source. Check the lens between runs; a quick wipe saves a whole session from being flagged as unusable.

What to watch out for

Some blur is intentional. Speed and the classic cinematic look both rely on motion blur, and an edge-based metric may mark beautiful footage as defective. Thresholds have to be tuned for action footage, not product photos.

Low-detail scenes fool simple metrics. A frame of open sky, fresh snow or calm water has few edges, so it reads as blurry even when perfectly sharp.

Heavy compression also hides blur and adds fake detail. Analysis on blocky, low-bitrate files is less reliable, which is another reason to keep original files rather than re-encoded copies.