Image recognition system and recognition method thereof, and program

• US 20030161504A1
• Filed: 02/26/2003
• Published: 08/28/2003
• Est. Priority Date: 02/27/2002
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Inoue, Akira
• Granted Patent: US 7,532,745 B2
• Time in Patent Office: Days
• US Class Current: 382/115
• CPC Class Codes: G06F 18/213 Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 10/507 Summing image-intensity values; Histogram projection analysis
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

A task is to correctly classify an input image regardless of a fluctuation in illumination and a state of occlusion of the input image.

Input image sub-region extraction means 2 extracts a sub-region of the input image. Inter-pattern distance calculation means 3 calculates an inter-pattern distance between this sub-region and a sub-region of a registration image pre-filed in dictionary filing means 5 for each sub-region. Region distance value integration means 10 integrates the inter-pattern distances obtained for each sub-region. This is conducted for the registration image of each category. Identification means 4 finds a minimum value out of its integrated inter-pattern distances, and in the event that its minimum value is smaller than a threshold, outputs a category having its minimum distance as a recognition result.

Claims

1. An image recognition system including:

calculation means for calculating each inter-pattern distance between a sub-region of an input image, and a sub-region of a registration image that corresponds to said sub-region of said input image; and

identification means for identifying said input image based on said each inter-pattern distance.

2. The image recognition system set forth in claim 1, said image recognition system further having:

means for calculating one integrated distance value from said inter-pattern distance of said each sub-region; and

means for employing said integrated distance value to identify said input image.

3. The image recognition system set forth in claim 1, wherein said calculation means is:

in the event that said registration image is a registration image string consisting of a plurality of images, means for calculating an inter-pattern distance between said sub-region of said input image, and a sub-region common in said registration image string that corresponds to said sub-region of said input image; and

in the event that said input image and said registration image are an input image string and a registration image string that consist of a plurality of images, respectively, means for calculating an inter-pattern distance between a sub-region common in said input image string, and a sub-region common in said registration image string that corresponds to said sub-region of said input image string.

4. The image recognition system set forth in claim 1, wherein said identification means is means for employing only said inter-pattern distance smaller than a predetermined threshold to identify said input image.

5. An image recognition method including:

a calculation step of calculating an inter-pattern distance value between a sub-region of an input image, and a sub-region of a registration image that corresponds to said sub-region of said input image; and

an identification step of identifying said input image based on said inter-pattern distance of each sub-region.

6. The image recognition method set forth in claim 5, said image recognition method including the steps of:

calculating one integrated distance value from said inter-pattern distance of said each sub-region; and

employing said integrated distance value to identify said input image.

7. The image recognition method set forth in claim 5, wherein said calculation step is:

in the event that said registration image is a registration image string consisting of a plurality of images, a step of calculating an inter-pattern distance between said sub-region of said input image, and a sub-region common in said registration image string that corresponds to said sub-region of said input image; and

in the event that said input image and said registration image are an input image string and a registration image string that consist of a plurality of images, respectively, a step of calculating an inter-pattern distance between a sub-region common in said input image string, and a sub-region common in said registration image string that corresponds to said sub-region of said input image string.

8. The image recognition method set forth in claim 5, wherein said identification step includes a step of employing only said inter-pattern distance smaller than a predetermined threshold to identify said input image.

9. A program for causing:

calculation means for calculating an inter-pattern distance between a sub-region of an input image, and a sub-region of a registration image that corresponds to said sub-region of said input image; and

identification means for identifying said input image based on said inter-pattern distance of each sub-region to function.

10. The program set forth in claim 9, said program causing;

means for calculating one integrated distance value from said inter-pattern distance of said each sub-region; and

means for employing said integrated distance value to identify said input image to function.

11. The program set forth in claim 9, wherein said calculation means is:

in the event that said registration image is a registration image string consisting of a plurality of images, means for calculating an inter-pattern distance between said sub-region of said input image, and a sub-region common in said registration image string that corresponds to said sub-region of said input image; and

in the event that said input image and said registration image are an input image string and a registration image string that consist of a plurality of images, respectively, means for calculating an inter-pattern distance between a sub-region common in said input image string, and a sub-region common in said registration image string that corresponds to said sub-region of said input image string.

12. The program set forth in claim 9, wherein said identification means is means for employing only said inter-pattern distances smaller than a predetermined threshold to identify said input image.

13. A record medium in which the program set forth in claim 9 was filed.

14. An image recognition method of collating and identifying an input image and a registration image, said image recognition method including:

an input image sub-region extraction step of employing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels to generate a by-region input feature vector sub-region by sub-region of the input image;

an inter-pattern distance calculation step of employing the by-region input feature vector, and dictionary data, which was registered, to calculate P inter-pattern distance values sub-region by sub-region;

a region distance value integration step of calculating one integrated distance value, based on the P inter-pattern distance values obtained in said inter-pattern distance calculation step; and

an identification step of employing said integrated distance value to determining a category to which the input image belongs.

15. The image recognition method set forth in claim 14, said image recognition method including:

a registration image sub-region extraction step of generating a by-region dictionary feature vector by sub-region from the registration image, based on the arrangement information of the sub-region;

a dictionary data generation step of generating the dictionary data from the by-region dictionary feature vector; and

a registration step of registering said dictionary data.

16. The image recognition method set forth in claim 14, wherein, in the event that said dictionary data includes a by-region dictionary feature vector extracted from the registration image, said inter-pattern distance calculation step includes a step of calculating an inter-vector normed distance between said by-region dictionary feature vector and said by-region input feature vector.

17. The image recognition method set forth in claim 14, wherein, in the event that said dictionary data includes by-region principal-component data generated by a by-region dictionary feature vector group extracted from a registration image group, said inter-pattern distance calculation step includes a step of calculating a subspace projection distance between a subspace formed by K eigenvectors to be included in said by-region principal-component data, and said by-region input feature vector.

18. The image recognition method set forth in claim 14, wherein said input image sub-region extraction step includes a posture compensation step of converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

19. The image recognition method set forth in claim 14, wherein said registration image sub-region extraction step includes a fluctuation image generation step of converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

20. The image recognition method set forth in claim 14, wherein said region distance value integration step includes a step of calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

21. The image recognition method set forth in claim 14, wherein said region distance value integration step includes a step of calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

22. The image recognition method set forth in claim 14, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

23. The image recognition method set forth in claim 14, wherein an identification step object is a human face.

24. The image recognition method set forth in claim 19, wherein, in said fluctuation image generation step, the registration image is converted so that a face in the registration image differs in visualization due to a variation in posture or a variation in illumination, or a variation in expression of the face.

25. An image recognition method of collating and identifying an input image string and a registration image string, said image recognition method including:

an input image string smoothing step of averaging said input image string in a pixel unit to generate one sheet of an input image;

an input image sub-region extraction step of employing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels to generate a by-region input feature vector sub-region by sub-region from said input image;

an inter-pattern distance calculation step of employing the by-region input feature vector, and dictionary data, which was registered, to calculate P inter-pattern distance values sub-region by sub-region;

a region distance value integration step of calculating one integrated distance value, based on the P inter-pattern distance values obtained in said inter-pattern distance calculation step; and

an identification step of employing said integrated distance value to determining a category to which the input image belongs.

26. The image recognition method set forth in claim 25, said image recognition method including:

a registration image sub-region extraction step of generating a by-region dictionary feature vector by sub-region from the registration image, based on the arrangement information of the sub-region;

a dictionary data generation step of generating the dictionary data from the by-region dictionary feature vector; and

a registration step of registering said dictionary data.

27. The image recognition method set forth in claim 25, wherein, in the event that said dictionary data includes a by-region dictionary feature vector extracted from the registration image, said inter-pattern distance calculation step includes a step of calculating an inter-vector normed distance between said by-region dictionary feature vector and said by-region input feature vector.

28. The image recognition method set forth in claim 25, wherein, in the event that said dictionary data includes by-region principal-component data generated by a by-region dictionary feature vector group extracted from a registration image group, said inter-pattern distance calculation step includes a step of calculating a subspace projection distance between a subspace formed by K eigenvectors to be included in said by-region principal-component data, and said by-region input feature vector.

29. The image recognition method set forth in claim 25, wherein said input image sub-region extraction step includes a posture compensation step of converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

30. The image recognition method set forth in claim 25, wherein said registration image sub-region extraction step includes a fluctuation image generation step of converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

31. The image recognition method set forth in claim 25, wherein said region distance value integration step includes a step of calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

32. The image recognition method set forth in claim 25, wherein said region distance value integration step includes a step of calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

33. The image recognition method set forth in claim 25, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

34. The image recognition method set forth in claim 25, wherein an identification step object is a human face.

35. The image recognition method set forth in claim 30, wherein, in said fluctuation image generation step, the registration image is converted so that a face in the registration differs in visualization due to a variation in posture or a variation in illumination, or a variation in expression of the face.

36. An image recognition method of collating and identifying an input image string and a registration image string, said image recognition method including:

an input image sub-region extraction step of employing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels to generate a by-region input feature vector sub-region by sub-region from said input image string;

an input principal-component generation step of generating by-region input principal-component data from the by-region input feature vector obtained from said input image string;

an inter-pattern distance calculation step of employing the by-region input principal-component data and by-region dictionary principal-component data, which was registered, to calculate P inter-subspace distance values sub-region by sub-region;

a region distance value integration step of calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

an identification step of employing said integrated distance value to determining a category to which the input image belongs.

37. The image recognition method set forth in claim 36, said image recognition method including:

a registration image sub-region extraction step of generating a by-region dictionary feature vector by sub-region from the registration image string, based on the arrangement information of the sub-region;

a dictionary data generation step of generating by-region dictionary principal-component data from a by-region dictionary feature vector group; and

a registration step of registering said dictionary data.

38. The image recognition method set forth in claim 36, wherein said input image sub-region extraction step includes a posture compensation step of converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

39. The image recognition method set forth in claim 36, wherein said registration image sub-region extraction step includes a fluctuation image generation step of converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

40. The image recognition method set forth in claim 36, wherein said region distance value integration step includes a step of calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

41. The image recognition method set forth in claim 36, wherein said region distance value integration step includes a step of calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

42. The image recognition method set forth in claim 36, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

43. The image recognition method set forth in claim 36, wherein an identification step object is a human face.

44. The image recognition method set forth in claim 39, wherein, in said fluctuation image generation step, the registration image is converted so that a face in the registration image differs in visualization due to a variation in posture or a variation in illumination, or a variation in expression of the face.

45. An image recognition system for collating and identifying an input image and a registration image, said image recognition system including:

sub-region information storage means for storing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels;

input image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region input feature vector sub-region by sub-region from the input image;

inter-pattern distance calculation means for employing said by-region input feature vector, and dictionary data, which was registered, to calculate P inter-pattern distance values sub-region by sub-region;

region distance value integration means for calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

identification means for employing said integrated distance value to determining a category to which the input image belongs.

46. The image recognition system set forth in claim 45, said image recognition system including:

dictionary filing means for filing the dictionary data;

registration image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region dictionary feature vector sub-region by sub-region from the registration image;

dictionary data generation means for generating the dictionary data from said by-region dictionary feature vector; and

a registration section for registering the dictionary data in said dictionary filing means.

47. The image recognition system set forth in claim 45, wherein, in the event that said dictionary data includes a by-region dictionary feature vector extracted from the registration image by employing information of the sub-region information storage means, said inter-pattern distance calculation means has means for calculating an inter-vector normed distance between said by-region dictionary feature vector and said by-region input feature vector.

48. The image recognition system set forth in claim 45, wherein, in the event that said dictionary data includes by-region principal-component data generated by a by-region dictionary feature vector group extracted from a registration image group by employing information of the sub-region information storage means, said inter-pattern distance calculation means has means for calculating a subspace projection distance between a subspace formed by K eigenvectors to be included in said by-region principal-component data, and said by-region input feature vector.

49. The image recognition system set forth in claim 45, wherein said input image sub-region extraction means includes posture compensation means for converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

50. The image recognition system set forth in claim 45, wherein said registration image sub-region extraction means includes fluctuation image generation means for converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

51. The image recognition system set forth in claim 45, wherein said region distance value integration means is means for calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

52. The image recognition system set forth in claim 45, wherein said region distance value integration means is means for calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

53. The image recognition system set forth in claim 45, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

54. The image recognition system set forth in claim 45, wherein an identification object is a human face.

55. An image recognition system for collating and identifying an input image and a registration image, said image recognition system including:

an input image string smoothing means for averaging an input image string in a pixel unit to generate one sheet of an input image;

sub-region information storage means for storing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels;

input image sub-region extraction means for employing information of the sub-region information storage means to generate a by-region input feature vector sub-region by sub-region from the input image;

inter-pattern distance calculation means for employing said by-region input feature vector and dictionary data, which was registered, to calculate P inter-pattern distance values sub-region by sub-region;

region distance value integration means for calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

identification means for employing said integrated distance value to determining a category to which the input image belongs.

56. The image recognition system set forth in claim 55, said image recognition system including:

dictionary filing means for filing dictionary data;

registration image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region dictionary feature vector sub-region by sub-region from the registration image;

dictionary data generation means for generating the dictionary data from said by-region dictionary feature vector; and

a registration section for registering the dictionary data in said dictionary filing means.

57. The image recognition system set forth in claim 55, wherein, in the event that said dictionary data includes a by-region dictionary feature vector extracted from the registration image by employing information of the sub-region information storage means, said inter-pattern distance calculation means has means for calculating an inter-vector normed distance between said by-region dictionary feature vector and said by-region input feature vector.

58. The image recognition system set forth in claim 55, wherein, in the event that said dictionary data includes by-region principal-component data generated by a by-region dictionary feature vector group extracted from a registration image group by employing information of the sub-region information storage means, said inter-pattern distance calculation means has means for calculating a subspace projection distance between a subspace formed by K eigenvectors to be included in said by-region principal-component data, and said by-region input feature vector.

59. The image recognition system set forth in claim 55, wherein said input image sub-region extraction means includes posture compensation means for converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

60. The image recognition system set forth in claim 55, wherein said registration image sub-region extraction means includes fluctuation image generation means for converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

61. The image recognition system set forth in claim 55, wherein said region distance value integration means is means for calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

62. The image recognition system set forth in claim 55, wherein said region distance value integration means is means for calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

63. The image recognition system set forth in claim 55, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

64. The image recognition system set forth in claim 55, wherein an identification object is a human face.

65. An image recognition system for collating and identifying an input image and a registration image, said image recognition system including:

sub-region information storage means for storing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels;

input image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region input feature vector sub-region by sub-region from the input image string;

an input principal-component generation section for generating by-region input principal-component data from the by-region input feature vector obtained from said input image string;

inter-pattern distance calculation means for employing said by-region input principal-component data, and by-region dictionary principal-component data, which was registered, to calculate P inter-subspace distance values sub-region by sub-region;

region distance value integration means for calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

identification means for employing said integrated distance value to determining a category to which the input image belongs.

66. The image recognition system set forth in claim 65, said image recognition system including:

dictionary filing means for filing dictionary data;

registration image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region dictionary feature vector sub-region by sub-region from the registration image;

dictionary data generation means for generating the dictionary data from said by-region dictionary feature vector; and

a registration section for registering the dictionary data in said dictionary filing means.

67. The image recognition system set forth in claim 65, wherein said input image sub-region extraction means includes posture compensation means for converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

68. The image recognition system set forth in claim 65, wherein said registration image sub-region extraction means includes fluctuation image generation means for converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

69. The image recognition system set forth in claim 65, wherein said region distance value integration means is means for calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

70. The image recognition system set forth in claim 65, wherein said region distance value integration means is means for calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

71. The image recognition system set forth in claim 65, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

72. The image recognition system set forth in claim 65, wherein an identification object is a human face.

73. A program of an image recognition system for collating and identifying an input image and a registration image, said program causing:

input image sub-region extraction means for employing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels to generate a by-region input feature vector sub-region by sub-region from said input image;

inter-pattern distance calculation means for employing said by-region input feature vector and dictionary data, which was registered, to calculate P inter-pattern distance values sub-region by sub-region;

region distance value integration means for calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

identification means for employing said integrated distance value to determine a category to which the input image belongs to function.

74. The program set forth in claim 73, said program causing:

dictionary filing means for filing the dictionary data;

registration image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region dictionary feature vector sub-region by sub-region from the registration image;

dictionary data generation means for generating the dictionary data from said by-region dictionary feature vector; and

a registration section for registering the dictionary data in said dictionary filing means to function.

75. The program set forth in claim 73, wherein, in the event that said dictionary data includes a by-region dictionary feature vector extracted from the registration image, said inter-pattern distance calculation means is means for calculating an inter-vector normed distance between said by-region dictionary feature vector and said by-region input feature vector.

76. The program set forth in claim 73, wherein, in the event that said dictionary data includes by-region principal-component data generated by a by-region dictionary feature vector group extracted from a registration image group, said inter-pattern distance calculation means is means for calculating a subspace projection distance between a subspace formed by K eigenvectors to be included in said by-region principal-component data, and said by-region input feature vector.

77. The program set forth in claim 73, wherein said input image sub-region extraction means includes posture compensation means for converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

78. The program set forth in claim 73, wherein said registration image sub-region extraction means includes fluctuation image generation means for converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

79. The program set forth in claim 73, wherein said region distance value integration means includes means for calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

80. The program set forth in claim 73, wherein said region distance value integration means includes means for calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

81. The program set forth in claim 73, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

82. The program set forth in claim 73, wherein an identification object is a human face.

83. The program set forth in claim 78, wherein, in the event that an identification object is a human face, said fluctuation image generation means is means for converting the registration image into an image in which the face in the registration image differs in visualization due to a variation in posture or a variation in illumination, or a variation in expression of the face.

84. A record medium in which the program set forth in claim 78 was filed.

85. A program of an image recognition system for collating and identifying an input image and a registration image, said program causing:

input image string smoothing means for averaging an input image string in a pixel unit to generate one sheet of the input image;

input image sub-region extraction means for employing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels to generate a by-region input feature vector sub-region by sub-region from the input image;

inter-pattern distance calculation means for employing said by-region input feature vector, and dictionary data, which was registered, to calculate P inter-pattern distance values sub-region by sub-region;

region distance value integration means for calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

identification means for employing said integrated distance value to determine a category to which the input image belongs to function.

86. The program set forth in claim 85, said program causing:

dictionary filing means for filing the dictionary data;

registration image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region dictionary feature vector sub-region by sub-region from the registration image;

dictionary data generation means for generating the dictionary data from said by-region dictionary feature vector; and

a registration section for registering the dictionary data in said dictionary filing means to function.

87. The program set forth in claim 85, wherein, in the event that said dictionary data includes a by-region dictionary feature vector extracted from the registration image, said inter-pattern distance calculation means is means for calculating an inter-vector normed distance between said by-region dictionary feature vector and said by-region input feature vector.

88. The program set forth in claim 85, wherein, in the event that said dictionary data includes by-region principal-component data generated by a by-region dictionary feature vector group extracted from a registration image group, said inter-pattern distance calculation means is means for calculating a subspace projection distance between a subspace formed by K eigenvectors to be included in said by-region principal-component data, and said by-region input feature vector.

89. The program set forth in claim 85, wherein said input image sub-region extraction means includes posture compensation means for converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

90. The program set forth in claim 85, wherein said registration image sub-region extraction means includes fluctuation image generation means for converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

91. The program set forth in claim 85, wherein said region distance value integration means includes means for calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

92. The program set forth in claim 85, wherein said region distance value integration means includes means for calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

93. The program set forth in claim 85, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

94. The program set forth in claim 85, wherein an identification object is a human face.

95. The program set forth in claim 90, wherein, in the event that an identification object is a human face, said fluctuation image generation means is means for converting the registration image into an image in which the face in the registration image differs in visualization due to a variation in posture or a variation in illumination, or a variation in expression of the face.

96. A record medium in which the program set forth in claim 85 was filed.

97. A program of an image recognition system for collating and identifying an input image string and a registration image string, said program causing:

input image sub-region extraction means for employing arrangement information of P (P is an integer equal to or more than

     2) sub-regions consisting of plural pixels to generate a by-region input feature vector sub-region by sub-region from the input image string;

input principal-component generation means for generating by-region input principal-component data from the by-region input feature vector obtained from said input image string;

inter-pattern distance calculation means for employing said by-region input principal-component data and by-region dictionary principal-component data, which was registered, to calculate P inter-subspace distance values sub-region by sub-region;

region distance value integration means for calculating one integrated distance value, based on the obtained P inter-pattern distance values; and

identification means for employing said integrated distance value to determine a category to which the input image belongs to function.

98. The program set forth in claim 97, said program causing:

dictionary filing means for filing dictionary data;

registration image sub-region extraction means for employing information of said sub-region information storage means to generate a by-region dictionary feature vector sub-region by sub-region from the registration image string;

dictionary data generation means for generating the dictionary data from said by-region dictionary feature vector group; and

a registration section for registering the dictionary data in said dictionary filing means to function.

99. The program set forth in claim 97, wherein said input image sub-region extraction means includes posture compensation means for converting the input image or the registration image so that a body of the input image or the registration image points to a pre-determined direction.

100. The program set forth in claim 97, wherein said registration image sub-region extraction means includes fluctuation image generation means for converting the registration image into a plurality of images in which a direction of a body of the registration image is different, or into a plurality of images in which the body differs in visualization due to a variation in posture or a variation in illumination.

101. The program set forth in claim 97, wherein said region distance value integration means includes means for calculating an integrated distance value as a weighted mean value of the P inter-pattern distance values.

102. The program set forth in claim 97, wherein said region distance value integration means includes means for calculating an integrated distance value as a mean value of P′

(P′

is an integer less than P) inter-pattern distance values, of which the distance is nearer, out of P.

103. The program set forth in claim 97, wherein said by-region input feature vector, or said by-region dictionary feature vector is a vector that takes a pixel value of the sub-region as an element.

104. The program set forth in claim 97, wherein an identification object is a human face.

105. The program set forth in claim 100, wherein, in the event that an identification object is a human face, said fluctuation image generation means is means for converting the registration image into an image in which the face in the registration image differs in visualization due to a variation in posture or a variation in illumination, or a variation in expression of the face.

106. A record medium in which the program set forth in claim 97 was filed.