| 摘要 |
PURPOSE: A method of generating a vector quantization codebook, wherein the entries in the codebook are used with substantially equal frequency when compressing data whose statistical properties mirror those of the training data used to construct the codebook is provided to mitigate disadvantages of conventional vector quantization codebooks. CONSTITUTION: A method of constructing a vector quantization codebook from at least one archetypal data array composed of a set of data values is comprised of: selecting from the data array(s) a first multiplicity of n-dimension sample vectors, each sample vector consisting of a set of data values which are contiguous in the array(s) and each sample vector defining a point in a finite n-dimensional space; partitioning the space into a predetermined number of regions, each region containing substantially the same number of sample vectors; assigning to each region a unique index, where the indices are selected to codify the partitioning process previously; determining for substantially all possible points within the space, the regions in which these points are located, and constructing a look-up index table mapping substantially all possible points to the respective region indices; selecting from the data array(s) a second multiplicity of n-dimension sample vectors, each sample vector consisting of a set of contiguous data values, and replacing each of these sample vectors with the associated region index obtained by looking up the index table to create a further data array or arrays; repeating prior steps for the further data array(s) and any subsequently generated further data array(s), wherein each index generated in the final iteration is derived from a set of nm dimension sample vectors in the archetypal data array(s), where m is the number of iterations carried out; for each index generated in the final iteration, creating an nm dimension reference vector which is representative of the associated set of nm dimension sample vectors in the archetypal data array; and constructing a codebook containing the reference vectors, where each index generated in the final iteration points to the location in the codebook of the corresponding reference vector. |
