Polarization used to be a specialist's measurement. Now it is a camera feature. Dividerless polarization sensors — image sensors with micropolarizer arrays bonded directly onto the pixels — ship in machine-vision cameras for glare removal, stress birefringence inspection of plastics and glass, and material classification in robotic sorting. The sensor reads four polarization angles per pixel group and reconstructs the full Stokes vector of the scene. Everything about the image chain assumes one thing: that the optics between the scene and the sensor do not change that polarization state. Most window glass does exactly that.

Where the polarization error comes from
Stress birefringence is the main culprit. Glass that was cooled too fast in annealing, cut with residual edge stress, or mounted against an overtightened flange develops internal stress, and stressed glass is birefringent: the two polarization components travel at slightly different speeds and accumulate a phase difference. In an ordinary camera this is invisible. In a polarization camera it appears directly in the data — as a false retardance map, as degree-of-polarization errors, or as angle artifacts in a stress-inspection application where the sample under test is measured through the very window that is itself stressed.
The irony is sharp in industrial inspection. A polarization camera is often mounted behind a protective window to inspect moulded parts for the same kind of stress. If the window contributes retardance comparable to the sample's, the instrument reports ghost stress that moves when the ambient temperature moves.
The numbers that matter
Glass grades are specified with a stress birefringence limit, classically stated in nanometres of optical path difference per centimetre of path length. Standard annealing may permit 10 nm/cm; precision annealing tightens this to under 1 nm/cm for the grades used in polarization-critical optics. A 3 mm window of ordinary glass can therefore contribute 3 nm or more of retardance; the same window in a fine-annealed grade contributes a few tenths of a nanometre. Polarization sensors with division-of-focal-plane architecture typically need window contributions well under a nanometre across the aperture to keep measurement uncertainty in the useful range.
Mounting stress adds on top of the material limit. A window clamped rigidly by steel screws will pick up birefringence at the contact points even if the blank was perfect. Elastomeric sealing, compliant retainers and generous clearance at the edge are not luxuries in this application; they are part of the polarization budget.
Coating choices interact with the budget
An anti-reflection coating changes the game twice. First, reflection at uncoated surfaces partially polarizes off-axis light, adding angle-dependent error at the field edges. A good AR coating suppresses that. Second, coating stress during deposition can bend the substrate slightly, and a bent window in a mounted assembly is a stressed window. Coating houses that regularly supply polarization optics tune their layer stacks not only for transmission but for near-zero net stress, and they verify the result by measuring retardance on the coated part, not the bare blank.
How to specify, in practice
- Substrate grade with a written stress-birefringence class, measured per blank or per melting lot, not assumed from the glass catalogue.
- Total retardance of the finished, coated, mounted window — measured at the working wavelength — rather than the material spec alone.
- Mounting design review: compliant sealing, controlled screw torque, no hard point contact on the optical aperture.
- Homogeneity data if the window is large, because bulk striae behave like localized birefringence.
- A polarization calibration procedure in the instrument that separates window contribution from sensor contribution, so the window can be changed without recalibrating from scratch.
Why this matters now
Polarization imaging is expanding fastest exactly where protective windows are standard: factory floors, outdoor robotics, automotive sensing. The suppliers who already control fine-annealed glass and low-stress coating will define the component baseline for these instruments, and the difference between a random window and a polarization-qualified window is the difference between data and decoration.
Product page: beam-splitter and polarization optics — specifications, coating options and lead times.