Image recognition using hidden markov models and coupled hidden markov models

• US 7,171,043 B2
• Filed: 10/11/2002
• Issued: 01/30/2007
• Est. Priority Date: 10/11/2002
• Status: Expired due to Fees

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Nefian, Ara V.
• Primary Examiner(s): Mancuso; Joseph
• Assistant Examiner(s): DULANEY, KATHLEEN YUAN
• Application Number:
  US10/269,333
• Publication Number:
  US 20040071338A1
• Time in Patent Office: 1,572 Days
• Field of Search: 382/159
• US Class Current: 382/159
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

An image processing system useful for facial recognition and security identification obtains an array of observation vectors from a facial image to be identified. A Viterbi algorithm is applied to the observation vectors given the parameters of a hierarchical statistical model for each object, and a face is identified by finding a highest matching score between an observation sequence and the hierarchical statistical model.

Claims

1. An image processing method, comprising:

forming from multiple images a hierarchical statistical model for each object to be identified in an image training database, the hierarchical statistical model having a parent layer with multiple supernodes associated with a first image direction and a child layer having multiple nodes associated with each supernode of the parent layer and a second image direction, wherein the parent layer is formed from a hidden Markov model (HMM) and the child layer is formed from a coupled hidden Markov model (CHMM), or the parent layer is formed from a CHMM and the child layer is formed from an HMM;

obtaining an array of observation vectors from an image to be identified; and

applying a Viterbi algorithm to the observation vectors given parameters of the hierarchical statistical model for each object, and identifying an object by finding a highest matching score between an observation sequence and hierarchical statistical model.

2. The method according to claim 1, wherein the objects are faces.

3. The method according to claim 1, further comprising estimating the parameters of the hierarchical statistical model based on the multiple images.

4. An article comprising a storage medium having stored thereon instructions that when executed by a machine result in:

forming from multiple images a hierarchical statistical model for each object to be identified in an image training database, the hierarchical statistical model having a parent layer with multiple supernodes associated with a first image direction and a child layer having multiple nodes associated with each supernode of the parent layer and a second image direction, wherein the parent layer is formed from a hidden Markov model (HMM) and the child layer is formed from a coupled hidden Markov model (CHMM), or the parent layer is formed from a CHMM and the child layer is formed from an HMM;

obtaining an array of observation vectors from an image to be identified; and

applying a Viterbi algorithm to the observation vectors given parameters of the hierarchical statistical model for each object, and identifying an object by finding a highest matching score between an observation sequence and hierarchical statistical model.

5. The article of claim 4, wherein the objects are faces.

6. The article of claim 4, wherein the instructions that when executed by a machine further result in estimating the parameters of the hierarchical statistical model based from the multiple images.

7. An image processing system comprising:

an image training database having a hierarchical statistical model for each object to be identified, the hierarchical statistical model having a parent layer with multiple supernodes associated with a first image direction and a child layer having multiple nodes associated with each supernode of the parent layer and a second image direction, wherein the parent layer is formed from a hidden Markov model (HMM) and the child layer is formed from a coupled hidden Markov model (CHMM), or the parent layer is formed from a CHMM and the child layer is formed from an HMM; and

a classification module that obtains an array of observation vectors from an image to be identified and tests it for identity against the image training database by applying a Viterbi algorithm to the observation vectors given parameters of the hierarchical statistical model for each object, and identifying an object by finding a highest matching score between an observation sequence and the hierarchical statistical model in the image training database.

8. The image processing system according to claim 7, wherein the objects are faces.

9. The image processing system according to claim 7, further comprising a training module to estimate the parameters of the hierarchical statistical model based on the image training database.