System for analyzing and synthesis of multi-factor data

• US 6,549,899 B1
• Filed: 11/14/1997
• Issued: 04/15/2003
• Est. Priority Date: 11/14/1997
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Tennenbaum, Joshua B., Freeman, William T.
• Primary Examiner(s): Corrielus, Jean M.
• Application Number:
  US08/970,824
• Time in Patent Office: 1,978 Days
• Field of Search: 707/1, 707/503, 707/6, 364/728.5, 364/736.3, 364/724.11, 364/723, 428/141, 358/1-5-, 385/59, 709/219
• US Class Current: 1/1
• CPC Class Codes: G06F 18/21 Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/279 Recognition of textual entities
• CPC Class Codes: Y10S 707/99931 Database or file accessing
• CPC Class Codes: Y10S 707/99936 Pattern matching access

Abstract

The present invention provides a system for analyzing multi-factor data according to a separable mixture model to classify, extrapolate, and translate data. A separable mixture model represents each data point as a function of underlying factors and corresponding parameters. With two factors, the factors can be referred to as style and content. In particular, a particular data point can be represented according to a separable mixture model, as a combination of the product of parameter vectors for each of the factors. The system accommodates symmetric and asymmetric models of interactions of parameter vectors, which are selected based upon the known data and the desired analysis. Using a selected model and known data, the system generates relevant parameters describing the known data. The parameter vector information can then be used to analyze and synthesize test data. In classification, unknown data in a new style is classified as to content, or vice versa. In extrapolation, complete content in a new style is generated from a set of observations in the new style. In translation, data having both new style and content is used to generate complete sets of content and style corresponding to the new data. The system is applicable to many types of multi-factor or 2-factor data, including typography, voice recognition, face recognition with varying poses or illuminations, and color constancy.

Claims

1. A computer-based apparatus for analysis and synthesis of multi-factor data comprising:

means for storing a plurality of observed data values, wherein the data is definable by at least two separable factors;

parameter means for determining a parameter vector for each of the at least two separable factors based upon the observed data; and

matrix means for determining at least one combination matrix representing interaction between the at least two factors, based upon the observed data.

2. The apparatus of claim 1, wherein each of the observed data values includes a plurality of components, and wherein the matrix means determined a combination matrix corresponding to each component of the data values.

3. The apparatus of claim 1, further comprising:

means for receiving at least one test data;

means for determining at least one parameter vector for at least one factor for the at least one test data.

4. The apparatus of claim 1, further comprising:

means for receiving a plurality of test data having a different factor value than the observed data for at least one factor;

means for determining a test parameter vector for the at least one factor of the test data; and

means for generating new data based upon the test parameter vector.

5. A computer-based apparatus for analysis and synthesis of multi-factor data comprising:

means for storing a plurality of observed data values, wherein the data is definable by at least two separable factors;

parameter means for determining a parameter vector for at least one separable factor based upon the observed data; and

matrix means for determining a parameter matrix for at least one other separable factor based upon the observed data.

6. A method for analyzing multi-factor data, comprising the steps of:

representing parameters of known multi-factor data as a bilinear model; and

analyzing unknown multi-factor data based on the represented parameters to determine one of a factor of data included in the unknown multi-factor data and a factor of data missing from the unknown multi-factor data.

7. A method according to claim 6, wherein:

each parameter represents a product of linear forms corresponding to a first factor and a second factor, different than the first factor, of data included in the known multi-factor data; and

the determined factor is one of the first factor and the second factor.

8. A method according to claim 7, wherein the first factor is content and the second factor is style.

9. A method according to claim 6, wherein the analyzing of unknown multi-factor data includes at least one of:

categorizing parameters of the unknown multi-factor data based on the represented parameters of the known multi-factor data to determine the factor;

extrapolating parameters of the unknown multi-factor data from the represented parameters of the known multi-factor data to determine the factor; and

translating parameters of the unknown multi-factor data from the represented parameters of the known multi-factor data to determine the factor.

10. A method according to claim 9, further comprising the step of:

modifying the bilinear model to represent one of the categorized parameters, the extrapolated parameters, and the translated parameters of the unknown multi-factor data.

11. A method according to claim 6, wherein the bilinear model is one of a symmetric bilinear model and an asymmetric bilinear model.

12. A system for analyzing multi-factor data, comprising:

a memory configured to store a bilinear model representing parameters of known multi-factor data; and

a processor configured to receive unknown multi-factor data and to determine one of (i) a factor of data included in the unknown multi-factor data and (ii) a factor of data missing from the unknown multi-factor data, based on the stored bilinear model.

13. A system according to claim 12, wherein:

each of the parameters represented in the bilinear model represents a product of linear forms corresponding to a first factor and a second factor, different than the first factor, of data included in the known multi-factor data; and

the determined factor is one of the first factor and the second factor.

14. A system according to claim 13, wherein the first factor is content and the second factor is style.

15. A system according to claim 12, wherein the processor is further configured to determine the factor of the unknown multi-factor data by at least one of:

categorizing parameters of the unknown multi-factor data based on the represented parameters of the known multi-factor data to determine the factor;

extrapolating parameters of the unknown multi-factor data from the represented parameters of the known multi-factor data to determine the factor; and

translating parameters of the unknown multi-factor data from the represented parameters of the known multi-factor data to determine the factor.

16. A system according to claim 15, wherein the processor is further configured to modify the bilinear model to represent one of the categorized parameters, the extrapolated parameters, and the translated parameters of the unknown multi-factor data.

17. A system according to claim 12, wherein the bilinear model is one of a symmetric bilinear model and an asymmetric bilinear model.