A control signal (k_lum) derived from the luminance component (Y) of a video signal is used to adaptively control a temporal noise reduction filter (10) according to the level of motion in a video image. To compress the control signal for storage in a memory (30), the control signal is averaged over each 22 pixel area, and then a non-linear compression function is applied. The non-linear compression function preferably selects quantization values of the control signal which correspond to a perceptually substantially linear response in the noise reduction factor (NRF) of the noise filter (10).