Vector quantization
The vector quantization is a method of compressing or identification of records.
The data records are summarized in feature vectors . The principle of the method consists in assigning to these feature vectors that vector from a table which is most similar to the feature vector under consideration. Instead of saving all data of the feature vector, only the index of this most similar vector is required (see also database index ).
Vector quantization consists of two steps. In the first step, the training, the table (code book) must be created with frequently occurring feature vectors. In the second step, the code book vector with the smallest distance is determined for further vectors. Only the index of the code book vector is required for data transmission, which can also be a vector if the code book is multi-dimensional. The corresponding decoder must have the same code book and can then generate an approximation of the original vector from the index .
Another possible application is the assignment of data records to specific patterns in the same way as speech recognition . In this case, the distance between the feature vector and the code book vector is used to decide whether the data record under consideration corresponds to a pattern.