The invention discloses a method and a
system for protecting portrait
feature data of multilayer
space mapping, which are used for scenes such as user portraits / article portraits and the like. According to the method, a three-layer mapping architecture of a scalar input layer (S)-a middle random layer (R)-a target output layer (C) is adopted, and multi-stage irreversible transformation
processing is carried out on original scalar data. The method comprises the following steps: firstly, mapping a scalar into an intermediate vector by utilizing a one-way function of
cryptography security and combining a secret key and an offset; secondly, distance features of the intermediate vector are extracted, direction features are discarded, and the distance features are converted into two-dimensional space coordinates through a deterministic
mapping algorithm; according to the method, the irreversibility of reversely deducing the scalar quantity from the coordinates is ensured by erasing the angle information of the middle layer; the space isolation in different business scenes is realized by introducing the offset; according to the method, a deterministic rejection sampling
algorithm based on HMAC is preferably adopted, the problem of center aggregation existing in a traditional mapping method is solved, uniform distribution of coordinates in a set area is achieved, sensitive feature value scalar quantities of all dimensions in portrait feature vectors are mapped into two-dimensional coordinate points, and the two-dimensional coordinate points are aggregated to form an object portrait
feature matrix; the method is used for cross-
system analysis, clustering, similar retrieval and visual display, and meanwhile
plaintext feature value leakage can be avoided.