Line Scan vs Area Scan for Print Inspection: Where the Architecture Changes

If the inspected surface is a continuous web with no discrete part boundary — film, paper, label stock, foil, nonwoven — line scan is usually the architecture, because there is no natural trigger event and you want seamless coverage rather than overlapping snapshots.

If the surface arrives as discrete, indexed or intermittently moving parts — sheets, cartons, printed panels, bottles — area scan is usually the architecture, because a photo-eye or encoder position gives you a clean trigger and one image contains the whole feature of interest.

A line-scan camera builds an image one row at a time. If rows are clocked by an internal timer while the web speeds up and slows down, the image stretches and compresses, defect sizes become meaningless and repeating-defect period estimates fall apart.

The line trigger must come from a web encoder so that one image row equals a fixed distance of material regardless of speed. That single decision makes measurements speed-independent and makes defect positions reportable in real machine units.

• Encoder mounted where it reads actual web motion, not motor shaft position upstream of slip.
• Line rate budget checked against the maximum web speed at the chosen down-web resolution.
• Illumination intensity sized for the shortest line exposure at maximum speed, not at nominal.
• Encoder scaling verified against a measured length of material during commissioning.

Area scan gives you one resolution figure in both axes. Line scan gives you two, and they are set by different mechanisms.

Cross-web resolution

Web width ÷ sensor pixel count. Fixed by optics and sensor length; the only way to improve it is a longer sensor, more cameras, or a narrower field.

Down-web resolution

Web speed ÷ line rate, set by the encoder scaling. Adjustable in configuration, but bounded by the camera's maximum line rate and available light.

Square pixels

Usually desirable for defect shape and area metrics; achieved by matching down-web spacing to cross-web pixel size.

Illustrative sizing

A 1.0 m web on an 8k sensor gives ~125 µm cross-web. At 2 m/s with square pixels you need a line rate of about 16 kHz — which then sets the illumination requirement.

Line-scan illumination must be intense and uniform along a narrow strip across the entire web width, including the edges. Uniformity that is acceptable for an area-scan field is often unacceptable across a metre of web, because a gradual falloff toward the edges becomes a false-defect gradient or a blind zone.

Practical consequences: high-output line lights, careful alignment to the scan line, verification of flat-field uniformity across the full width, and a plan for lamp aging. Where transmissive defects matter — pinholes, coating voids — a backlit line is often the more informative geometry than a front light.

Line scan generates data continuously. An 8k monochrome sensor at 16 kHz is on the order of 128 MB/s per camera before any processing — sustained, not in bursts. That constrains interface choice, storage policy and how much history you can retain.

• Interface sized for sustained throughput, not peak: CoaXPress or Camera Link for high line rates, GigE/10 GigE where the budget allows.
• Processing distributed so the defect decision happens at rate; buffering only defers the problem.
• Image retention policy defined explicitly — typically defect crops plus context, never every frame.
• Multi-camera systems: overlap regions and stitching handled deterministically so a defect at a seam is reported once.

On discrete parts, the useful output is a verdict per part. On a web, the useful output is a map: defect class, size, cross-web position and down-web position in machine units, plus the repeat period when a defect recurs.

That map is what lets finishing skip a bad section, what tells a press operator that a defect repeating every 640 mm implicates a specific cylinder, and what turns inspection from an alarm into a process tool.

• Registration and print-to-print alignment measured on a discrete repeat where one image captures the full pattern.
• Sheet-fed work with generous cycle time, where a strobed area-scan image is simpler to stand up and easier to maintain.
• Low web speeds and modest resolution requirements, where a line-scan system's cost and complexity buy nothing.
• Applications needing color fidelity with a straightforward color area-scan sensor rather than trilinear line-scan calibration.
• Existing plant standardization on area-scan hardware and spares, where a second architecture adds a maintenance burden.