Speckle, and how to get rid of it

Illuminate a rough surface with a laser and the image breaks into bright and dark grain. That grain is not noise in the usual sense and no amount of filtering will remove it, because it is the correct interference pattern for the surface, the wavelength and your viewing geometry.

Why averaging is the only route

Speckle is deterministic. Change the illumination geometry slightly and you get a different but equally valid pattern. Average enough independent patterns together and the grain washes out toward the smooth intensity distribution you wanted. Contrast falls roughly as one over the square root of the number of independent patterns.

The word that matters is independent. A static diffuser in the beam gives you exactly one new pattern, which is no improvement. The pattern has to keep changing.

Driven diffusers

An electrically driven diffuser moves the scattering structure continuously, generating a stream of independent patterns. As long as the motion cycles faster than the detector integration time, the camera sees the average and the speckle is gone.

The failure mode is a short exposure. Freeze the diffuser mid-cycle with a fast shutter and the speckle returns, because you sampled one pattern rather than many.

What you give up

You are deliberately destroying spatial coherence. That is exactly what you want for illumination and exactly what you must not do to a beam you intend to use interferometrically. Split the beam first if you need both.