Psycho-visual redundancy is information in a video that the human eye won't actually notice — and therefore something a codec can safely throw away. It's distinct from spatial redundancy (neighbouring pixels are similar) and temporal redundancy (neighbouring frames are similar), which the codec exploits before encoding. Psycho-visual redundancy is about exploiting the limits of human perception itself.
The known weaknesses of human vision that codecs lean on. Eyes are far more sensitive to brightness changes than to colour changes — so chroma can be stored at half resolution (chroma-subsampling) with no visible loss. Eyes are more sensitive to low-frequency information (the overall shape of a face) than high-frequency information (individual pores) — so codecs quantize high-frequency dct coefficients more aggressively. Eyes can't track fine detail in busy or fast-moving areas because of motion masking — so encoders deliberately spend fewer bits there. Eyes adapt to ambient brightness and are bad at noticing small changes in shadows during bright scenes — so HDR transfer functions allocate code values according to perception.
For a product team, psycho-visual redundancy is the theoretical backbone of why modern video looks essentially identical to uncompressed at 1/300th the size. It's also the reason "perceptual quality metrics" like vmaf correlate so much better with viewer satisfaction than raw error metrics like PSNR — VMAF measures what the eye actually sees, while PSNR measures things the eye is biologically incapable of noticing. Encoder choices that target psycho-visual redundancy (perceptual quantization, masking-aware rate control, content-adaptive encoding) routinely save 15–25 % bitrate at the same visible quality.
