A method and apparatus for coding and decoding the fingerprint of a multimedia item such as video or audio is disclosed. A multimedia content temporal, such as a video segment or audio segment, is described by a coarse fingerprint and a plurality of fine fingerprints, each fine fingerprint corresponding to a temporal sub-interval of said temporal interval, said temporal sub-interval typically being smaller than said temporal interval. One or more fine fingerprints are encoded in a non-predictive way, with no reference to the temporally neighboring signatures, and one or more fine fingerprints are encoded in a predictive way, from the temporally neighboring signatures. The predictive encoding entails computing the difference between neighboring fine fingerprints to make up a prediction difference matrix, scanning said prediction difference matrix into a one dimensional vector by vectorising along rows or along columns or along diagonals or along any suitable scanning pattern, and performing lossless encoding on the one dimensional vector by an appropriate method, preferably selected, at least in part, based on the scanning method used.