Compression of MQDF classifier using flexible sub-vector grouping
AssigneeMicrosoft Corporation
InventorsQi Zhang, Wei Yu
Life of this patent
Estimating the term…
Abstract
Systems and methods to compress MQDF data are disclosed herein. A plurality of eigenvectors is identified. Each eigenvector in the plurality of eigenvectors can correspond to a pattern to be recognized. Each eigenvector in the plurality of eigenvectors can be split into sub-vectors. The sub-vectors can then be grouped into one or more groups according to a location of the sub-vectors within each of the eigenvectors. Each group can be associated with location data of the sub-vectors in the group. At least one group can be compressed according to a codebook. The codebook can be identifiable via the location data.
Claims20
A computer-implemented method of compressing Modified Quadratic Discriminant Function (MQDF) data, comprising:
- identifying a plurality of eigenvectors, each eigenvector in the plurality of eigenvectors corresponding to a class in a plurality of classes, each class corresponding to a pattern to be recognized;
- splitting each eigenvector in the plurality of eigenvectors into a number of sub-vectors, wherein all eigenvectors in the plurality of eigenvectors are split into the same number of sub-vectors;
- grouping the sub-vectors into one or more groups, the sub-vectors being grouped according to a location of the sub-vectors within each of the eigenvectors such that each group in the one or more groups is associated with location data;
- and
- compressing at least two of the groups of the one or more groups to form compressed groups, each one of the at least two compressed groups being compressed according to a different codebook among multiple codebooks, each different codebook being specific to the group being compressed and identifiable via the location data for each compressed group.
A computer-implemented method of compressing MQDF data, comprising:
- identifying a plurality of eigenvectors, each eigenvector in the plurality of eigenvectors corresponding to a class in a plurality of classes, each class corresponding to a pattern to be recognized;
- splitting each eigenvector in the plurality of eigenvectors into at least a first sub-vector, a second sub-vector, and a third sub-vector;
- grouping the first sub-vector of each eigenvector in the plurality of eigenvectors into a first group;
- grouping the second sub-vector of each eigenvector in the plurality of eigenvectors into a second group;
- grouping the third sub-vector of each eigenvector in the plurality of eigenvectors into a third group;
- and
- compressing the first group, the second group, and the third group using multiple codebooks, wherein compressing includes:
- compressing the first group using a first codebook;
- compressing the second group using a second codebook;
- and
- compressing the third group using a third codebook.
A computer-implemented method of compressing MQDF data, comprising:
- identifying a plurality of eigenvectors, each eigenvector in the plurality of eigenvectors corresponding to a class in a plurality of classes, each class corresponding to a pattern to be recognized;
- splitting each eigenvector in the plurality of eigenvectors into at least a first sub-vector, and a second sub-vector;
- grouping the first sub-vector of each eigenvector in the plurality of eigenvectors into a first group;
- grouping the second sub-vector of each eigenvector in the plurality of eigenvectors into a second group;
- leaving the first group uncompressed;
- and
- compressing the second group using a codebook.
Dependencies are inferred from the claim text.