| US 7,502,495 B2 | ||
| Method and system for incrementally learning an adaptive subspace by optimizing the maximum margin criterion | ||
| Benyu Zhang, Beijing (China); Hua-Jun Zeng, Beijing (China); Jun Yan, Beijing (China); Wei-Ying Ma, Beijing (China); and Zheng Chen, Beijing (China) | ||
| Assigned to Microsoft Corporation, Redmond, Wash. (US) | ||
| Filed on Mar. 01, 2005, as Appl. No. 11/70,382. | ||
| Prior Publication US 2006/0204081 A1, Sep. 14, 2006 | ||
| Int. Cl. G06K 9/00 (2006.01) | ||
| U.S. Cl. 382—118 [382/224] | 20 Claims |
| 1. A method in a computer system for incrementally updating a projection matrix for projecting data from a high dimensional
space to a low dimensional space, the method comprising:
using a computer to carry out the steps of:
establishing an objective function based on a maximum margin criterion matrix;
providing data samples, each data sample being in the high dimensional space and having a class; and
for each data sample,
calculating a vector that is the maximum margin criterion matrix times the eigenvector, the calculated vector being derived
incrementally from the vector calculated for the previous data sample;
deriving the leading eigenvector of the maximum margin criterion matrix from the calculated vector; and
for each subsequent eigenvector,
successively subtracting the projection of the previously derived eigenvector from the data sample;
calculating a vector that is the maximum margin criterion matrix times the subsequent eigenvector, the calculated vector being
derived incrementally from the previously calculated vector; and
deriving the subsequent eigenvector of the maximum margin criterion matrix from the calculated vector
wherein the derived eigenvectors compose the projection matrix.
|