A reduced-reference metric is the middle ground between full- and no-reference: it does not carry the whole original, only a compact set of features extracted from it, a few numbers small enough to travel alongside the stream through a low-bandwidth side channel. At the far end the metric extracts the same features from the impaired video and compares the two fingerprints, like checking a copy against a short tamper-evident checksum rather than the whole document. The NTIA pioneered this paradigm on perceptual features such as edges and motion energy, standardized as the Fast Low Bandwidth Model in ITU-T J.249, with J.246 covering multimedia over cable. The catch is logistical: a reduced-reference deployment needs a side channel connecting the measurement point back to the originating source, and most operators cannot make that connection, so it is rarely used in practice. Teams use full-reference in the lab and no-reference in production, squeezing out the middle, which makes reduced-reference more important for understanding the taxonomy than for daily work.
