Collation apparatus and collation method

• US 8,295,614 B2
• Filed: 04/12/2007
• Issued: 10/23/2012
• Est. Priority Date: 04/14/2006
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Ishiyama, Rui
• Primary Examiner(s): Bhatnagar, Anand
• Assistant Examiner(s): Park, Soo
• Application Number:
  US12/297,136
• Publication Number:
  US 20090180672A1
• Time in Patent Office: 2,021 Days
• Field of Search: 382/154, 382/209, 382215-223
• US Class Current: 382/215
• CPC Class Codes: G06T 2207/30168 Image quality inspection
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection
• CPC Class Codes: G06V 40/168 Feature extraction; Face representation
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A 2D model fitting means 1201 estimates values of parameters optimum to generate a probe model image similar to a probe image and a gallery model image similar to a gallery image with an image variation model 1204. At that time, among a plurality of parameters of the image variation model 1204, the value of a parameter of which sameness between the probe image and the gallery image is to be judged as a target parameter is set to be the same for both images. A model goodness-of-fit evaluation means 1202 computes a model goodness of fit for the probe model image and the gallery model image to the probe image and the gallery image under the estimated parameters. A collation means 1203 compares the model goodness of fit with a threshold value to judge the sameness between the probe image and the gallery image.

Claims

1. A collation apparatus for collating target data of an image or three-dimensional data of an object with reference data, the collation apparatus characterized by comprising:

storage means for storing a variation model capable of generating model data of an image or three-dimensional data of an object by adjusting values of a plurality of parameters;

model fitting means for defining, among the plurality of parameters, a parameter representing a judgement target of which sameness between the target data and the reference data is to be judged as a target parameter and other parameters as an external parameter, and for generating target model data that are most similar to the target data and reference model data that are most similar to the reference data based on the variation model under a condition in which the target parameter is set to be the same for the target data and the reference data; and

model goodness-of-fit evaluation means for computing a model goodness of fit of the target model data and the reference model data to the target data and the reference data.

2. The collation apparatus as recited in claim 1, characterized by comprising collation means for comparing the computed model goodness of fit with a predetermined threshold value to judge the sameness between a judgement target of the target data and a judgement target of the reference data.

3. The collation apparatus as recited in claim 2, characterized in that the target parameter is a parameter describing a kind of an object, and the collation means is means for judging whether or not an object of the target data is the same as an object of the reference data.

4. The collation apparatus as recited in claim 2, characterized in that the variation model is a model for generating an image or three-dimensional data of a face, the target parameter is a parameter describing an individuality of a person, and the collation means is means for determining whether or not a person of the target data is the same as a person of the reference data.

5. The collation apparatus as recited in claim 2, characterized in that the target parameter is a parameter specifying a position or a pose of an object, and the collation means is means for determining whether or not a position or a pose of an object of the target data is the same as a position or a pose of an object of the reference data.

6. The collation apparatus as recited in claim 2, characterized in that the variation model is a model for generating an image or three-dimensional data of a face, the target parameter is a parameter specifying a facial expression of a person, and the collation means is means for determining whether or not a facial expression of the target data is the same as a facial expression of the reference data.

7. The collation apparatus as recited in claim 1, characterized by comprising collation means for repeating processes of the model fitting means and the model goodness-of-fit evaluation means on each combination of each of plural pieces of reference data and the target data and for setting a value of the target parameter in a combination having the best model goodness of fit as an estimate value of the target parameter for the target data.

8. The collation apparatus as recited in claim 1, characterized by comprising collation means for repeating processes of the model fitting means and the model goodness-of-fit evaluation means on each combination of each of plural pieces of reference data and the target data and for setting a value computed from the target parameter in a plurality of combinations having a model goodness of fit better than a predetermine threshold value as an estimate value of the target parameter for the target data.

9. The collation apparatus as recited in claim 1, characterized in that:

the model goodness-of-fit evaluation means;

computes an error between the generated target model data and the target data and an error between the generated reference model data and the reference data, and

computes the model goodness of fit from the computed errors.

10. The collation apparatus as recited in claim 9, characterized in that the model fitting means estimates values of the parameters optimum to generate target individual model data similar to the target data from the variation model and values of the parameters optimum to generate reference individual model data similar to the reference data from the variation model, under a condition in which the value of the target parameter is not necessarily the same for the target data and the reference data, and

the model goodness-of-fit evaluation means computes the model goodness of fit with additional consideration of at least one of an error between the target individual model data and the target data, an error between the reference individual model data and the reference data, an error between the target model data and the target individual model data, and an error between the reference model data and the reference individual model data.

11. The collation apparatus as recited in claim 1, characterized in that:

the model goodness-of-fit evaluation means;

computes a parameter confidence factor representing certainty of the estimated values of the parameters with reference to predetermined parameter reference information, and

computes the model goodness of fit from the computed parameter confidence factor.

12. The collation apparatus as recited in claim 11, characterized in that the parameter reference information includes an allowable range of the value of the parameter.

13. The collation apparatus as recited in claim 11, characterized in that the parameter reference information includes a correct value of the parameter.

14. The collation apparatus as recited in claim 11, characterized in that the parameter reference information includes a probability distribution of the value of the parameter.

15. The collation apparatus as recited in claim 11, characterized in that the model fitting means estimates values of the parameters optimum to generate target individual model data similar to the target data from the variation model and values of the parameters optimum to generate reference individual model data similar to the reference data from the variation model, under a condition in which the value of the target parameter is not necessarily the same for the target data and the reference data, and

the model goodness-of-fit evaluation means sets the estimated values of the parameters for the target individual model data and the estimated values of the parameters for the reference individual model data as the parameter reference information.

16. The collation apparatus as recited in claim 1, characterized in that:

the model goodness-of-fit evaluation means;

computes an error between the generated target model data and the target data and an error between the generated reference model data and the reference data and computes a parameter confidence factor representing certainty of the estimated values of the parameters with reference to predetermined parameter reference information, and

computes the model goodness of fit from the computed errors and parameter confidence factor.

17. The collation apparatus as recited in claim 1, characterized in that the variation model is an image variation model including three-dimensional models of various objects for selecting a three-dimensional model of a specified object according to a parameter specifying the object, deforming the selected three-dimensional model according to a parameter such as a state or a pose of the object, an illumination condition, or a characteristic of a camera, and generating an image under a condition the parameter represents by using computer graphics.

18. The collation apparatus as recited in claim 1, characterized in that the variation model is an image variation model for computing a three-dimensional model of an object in a state according to a parameter specifying an individuality or a state of the object, deforming the computed three-dimensional model according to a parameter such as a pose of the object, an illumination condition, or a characteristic of a camera, and generating an image under a condition the parameter represents by using computer graphics.

19. The collation apparatus as recited in claim 1, characterized in that the variation model is a three-dimensional data variation model including three-dimensional models of various objects for selecting a three-dimensional model of a specified object according to a parameter specifying the object and deforming the selected three-dimensional model according to a parameter such as a state or a pose of the object, an illumination condition, or a characteristic of a camera.

20. The collation apparatus as recited in claim 1, characterized in that the variation model is a three-dimensional data variation model for computing a three-dimensional model of an object in a state according to a parameter specifying an individuality or a state of the object and deforming the computed three-dimensional model according to a parameter such as a pose of the object, an illumination condition, or a characteristic of a camera.

21. A collation method of collating target data of an image or three-dimensional data of an object with use of a computer, the collation method characterized by including:

a first step for the computer to use a variation model capable of generating model data of an image or three- dimensional data of an object by adjusting values of a plurality of parameters, to define, among the plurality of parameters, a parameter representing a judgement target of which sameness between the target data and the reference data is to be judged as a target parameter and other parameters as an external parameter, and to generate target model data that are most similar to the target data and reference model data that are most similar to the reference data based on the variation model under a condition in which the target parameter is set to be the same for the target data and the reference data; and

a second step for the computer to compute a model goodness of fit of the target model data and the reference model data to the target data and the reference data.

22. The collation method as recited in claim 21, characterized by further including a third step for the computer to compare the computed model goodness of fit with a predetermined threshold value to judge the sameness between a judgement target of the target data and a judgement target of the reference data.

23. The collation method as recited in claim 21, characterized by further including a third step for the computer to repeat the first and second steps on each combination of each of plural pieces of reference data and the target data and to set a value of the target parameter in a combination having the best model goodness of fit as an estimate value of the target parameter for the target data.

24. The collation method as recited in claim 21, characterized by further including a third step for the computer to repeat the first and second steps on each combination of each of plural pieces of reference data and the target data and to set a value computed from the target parameter in a plurality of combinations having a model goodness of fit better than a predetermine threshold value as an estimate value of the target parameter for the target data.

25. The collation method as recited in claim 21, characterized in that the second step comprises computing an error between the generated target model data and the target data and an error between the generated reference model data and the reference data, and computing the model goodness of fit from the computed errors.

26. The collation method as recited in claim 25, characterized in that the computer:

estimates values of the parameters optimum to generate target individual model data similar to the target data from the variation model and values of the parameters optimum to generate reference individual model data similar to the reference data from the variation model, under a condition in which the value of the target parameter is not necessarily the same for the target data and the reference data, and

computes the model goodness of fit with additional consideration of at least one of an error between the target individual model data and the target data, an error between the reference individual model data and the reference data, an error between the target model data and the target individual model data, and an error between the reference model data and the reference individual model data.

27. The collation method as recited in claim 21, characterized in that the second step comprises computing a parameter confidence factor representing certainty of the estimated values of the parameters with reference to predetermined parameter reference information, and computing the model goodness of fit from the computed parameter confidence factor.

28. The collation method as recited in claim 27, characterized in that the parameter reference information includes an allowable range of the value of the parameter.

29. The collation method as recited in claim 27, characterized in that the parameter reference information includes a correct value of the parameter.

30. The collation method as recited in claim 27, characterized in that the parameter reference information includes a probability distribution of the value of the parameter.

31. The collation method as recited in claim 27, characterized in that the computer estimates values of the parameters optimum to generate target individual model data similar to the target data from the variation model and values of the parameters optimum to generate reference individual model data similar to the reference data from the variation model, under a condition in which the value of the target parameter is not necessarily the same for the target data and the reference data, and

the computer sets the estimated values of the parameters for the target individual model data and the estimated values of the parameters for the reference individual model data as the parameter reference information.

32. The collation method as recited in claim 21, characterized in that the second step comprises computing an error between the generated target model data and the target data and an error between the generated reference model data and the reference data, computing a parameter confidence factor representing certainty of the estimated values of the parameters with reference to predetermined parameter reference information, and computing the model goodness of fit from the computed errors and parameter confidence factor.

33. A program embodied in a non-transitory computer-readable medium for providing a computer for collating target data of an image or three-dimensional data of an object with reference data with functions, the computer having storage means for storing a variation model capable of generating model data of an image or three-dimensional data of an object by adjusting values of a plurality of parameters, the functions including:

model fitting means for using the variation model stored in the storage means, for defining, among the plurality of parameters, a parameter representing a judgement target of which sameness between the target data and the reference data is to be judged as a target parameter and other parameters as an external parameter, and for generating target model data that are most similar to the target data and reference model data that are most similar to the reference data from the variation model under a condition in which the target parameter is set to be the same for the target data and the reference data; and

model goodness-of-fit evaluation means for computing a model goodness of fit of the target model data and the reference model data to the target data and the reference data.

34. The program as recited in claim 33, further providing the computer with a function as collation means for comparing the computed model goodness of fit with a predetermined threshold value to judge the sameness between a judgement target of the target data and a judgement target of the reference data.

35. The program as recited in claim 33, further providing the computer with a function as collation means for repeating processes of the model fitting means and the model goodness-of-fit evaluation means on each combination of each of plural pieces of reference data and the target data and for setting a value of the target parameter in a combination having the best model goodness of fit as an estimate value of the target parameter for the target data.

36. The program as recited in claim 33, further providing the computer with a function as collation means for repeating processes of the model fitting means and the model goodness-of-fit evaluation means on each combination of each of plural pieces of reference data and the target data and for setting a value computed from the target parameter in a plurality of combinations having a model goodness of fit better than a predetermine threshold value as an estimate value of the target parameter for the target data.

37. The program as recited in claim 33, characterized in that:

the model goodness-of-fit evaluation means;

computes an error between the generated target model data and the target data and an error between the generated reference model data and the reference data, and

computes the model goodness of fit from the computed errors.

38. The program as recited in claim 37, characterized in that the model fitting means estimates values of the parameters optimum to generate target individual model data similar to the target data from the variation model and values of the parameters optimum to generate reference individual model data similar to the reference data from the variation model, under a condition in which the value of the target parameter is not necessarily the same for the target data and the reference data, and

the model goodness-of-fit evaluation means computes the model goodness of fit with additional consideration of at least one of an error between the target individual model data and the target data, an error between the reference individual model data and the reference data, an error between the target model data and the target individual model data, and an error between the reference model data and the reference individual model data.

39. The program as recited in claim 33, characterized in that the model goodness-of-fit evaluation means computes a parameter confidence factor representing certainty of the estimated values of the parameters with reference to predetermined parameter reference information, and computes the model goodness of fit from the computed parameter confidence factor.

40. The program as recited in claim 39, characterized in that the parameter reference information includes an allowable range of the value of the parameter.

41. The program as recited in claim 39, characterized in that the parameter reference information includes a correct value of the parameter.

42. The program as recited in claim 39, characterized in that the parameter reference information includes a probability distribution of the value of the parameter.

43. The program as recited in claim 39, characterized in that the model fitting means estimates values of the parameters optimum to generate target individual model data similar to the target data from the variation model and values of the parameters optimum to generate reference individual model data similar to the reference data from the variation model, under a condition in which the value of the target parameter is not necessarily the same for the target data and the reference data, and

the model goodness-of-fit evaluation means sets the estimated values of the parameters for the target individual model data and the estimated values of the parameters for the reference individual model data as the parameter reference information.

44. The program as recited in claim 33, characterized in that:

the model goodness-of-fit evaluation means;

computes an error between the generated target model data and the target data and an error between the generated reference model data and the reference data and computes a parameter confidence factor representing certainty of the estimated values of the parameters with reference to predetermined parameter reference information, and

computes the model goodness of fit from the computed errors and parameter confidence factor.