Method for detecting similarity between standard information and input information and method for judging the input information by use of detected result of the similarity

• US 7,006,970 B2
• Filed: 09/11/2001
• Issued: 02/28/2006
• Est. Priority Date: 09/13/2000
• Status: Expired due to Fees

• Current Assignee: Michihiro Jinnai
• Original Assignee: Entropy Software Laboratory Incorporated
• Inventors: Yamaguchi, Hiroshi, Jinnai, Michihiro
• Primary Examiner(s): Ometz, David L.
• Assistant Examiner(s): JACKSON, JAKIEDA R
• Application Number:
  US09/951,211
• Publication Number:
  US 20020055839A1
• Time in Patent Office: 1,631 Days
• Field of Search: 704/239, 704/243, 704/254, 704/246, 381/128, 381/43, 395/2.5, 340/146.3, 382/181
• US Class Current: 704/239
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06V 10/761 Proximity, similarity or dissimilarity measures
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

Disclosed is a method for obtaining a precise detected value of a similarity between voices or the like. Standard and input pattern matrices, each having a voice feature amount as a component, are prepared (S1 and S2). A reference shape having a variance different for each specified component of the pattern matrices is prepared, and positive and negative reference pattern vectors, each having a value of the reference shape as a component, are prepared. Then, while the specified component (a center of the reference shape) being made to move to each component position j₁=1 to m₁, j₂=1 to m₂ of the standard pattern matrix, a shape change between the standard and input pattern matrices is substituted for shape changes of the positive and negative reference pattern vectors. And, an amount of change in kurtosis of each reference pattern vector is numerically evaluated to obtain a shape change amount Dj₁j₂ (S3). Then, a value of a geometric distance between the pattern matrices is calculated from Dj₁j₂ (S4).

Claims

1. A method for detecting a similarity between voices, comprising the steps of:

(a) preparing a standard pattern matrix with a feature amount of a standard voice as a component and an input pattern matrix with a feature amount of an input voice as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrices as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrices, obtaining a length between the specified component and a component in each of the pattern matrices, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is greater than the component value of the standard pattern matrix, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is smaller than the component value of the standard pattern matrix;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrices being made to move to the position of each component in calculating the value of the difference between the kurtoses; and

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix and the input pattern matrix.

2. The method for detecting a similarity between voices according to claim 1, wherein the reference distribution is a normal distribution.

3. A method for recognizing a voice, comprising the steps of:

(a) preparing a standard pattern matrix with a feature amount of a standard voice as a component and an input pattern matrix with a feature amount of an input voice as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrices as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrices, obtaining a length between the specified component and a component in each of the pattern matrices, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is greater than the component value of the standard pattern matrix, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is smaller than the component value of the standard pattern matrix;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrices being made to move to the position of each component in calculating the value of the difference between the kurtoses;

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix and the input pattern matrix; and

(g) comparing the obtained value of the geometric distance with an arbitrarily set acceptable value, judging that the input voice is not the standard voice when the value of the geometric distance exceeds the acceptable value, and judging that the input voice is the standard voice when the value of the geometric distance is within the acceptable value.

4. The method for recognizing a voice according to claim 3, wherein the reference distribution is a normal distribution.

5. A method for detecting a similarity between vibration waves, comprising the steps of:

(a) preparing a standard pattern matrix with a feature amount of a standard vibration wave as a component and an input pattern matrix with a feature amount of an input vibration wave as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrices as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrices, obtaining a length between the specified component and a component in each of the pattern matrices, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is greater than the component value of the standard pattern matrix, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is smaller than the component value of the standard pattern matrix;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrices being made to move to the position of each component in calculating the value of the difference between the kurtoses; and

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix and the input pattern matrix.

6. The method for detecting a similarity between vibration waves according to claim 5, wherein the reference distribution is a normal distribution.

7. A method for judging an abnormality in a machine, comprising the steps of:

(a) preparing a standard pattern matrix with a feature amount of a standard vibration wave as a component and an input pattern matrix with a feature amount of an input vibration wave as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrices as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrices, obtaining a length between the specified component and a component in each of the pattern matrices, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is greater than the component value of the standard pattern matrix, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is smaller than the component value of the standard pattern matrix;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrices being made to move to the position of each component in calculating the value of the difference between the kurtoses;

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix and the input pattern matrix; and

(g) comparing the obtained value of the geometric distance with an arbitrarily set acceptable value, judging that the machine is abnormal when the value of the geometric distance exceeds the acceptable value, and judging that the machine is normal when the value of the geometric distance is within the acceptable value.

8. The method for judging an abnormality in a machine according to claim 7, wherein the reference distribution is a normal distribution.

9. A method for detecting a similarity between images, comprising the steps of:

(a) preparing a standard pattern matrix with a feature amount of a standard image as a component and an input pattern matrix with a feature amount of an input image as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrices as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrices, obtaining a length between the specified component and a component in each of the pattern matrices, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is greater than the component value of the standard pattern matrix, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is smaller than the component value of the standard pattern matrix;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrices being made to move to the position of each component in calculating the value of the difference between the kurtoses; and

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix and the input pattern matrix.

10. The method for detecting a similarity between images according to claim 9, wherein the reference distribution is a normal distribution.

11. A method for recognizing an image, comprising the steps of:

(a) preparing a standard pattern matrix with a feature amount of a standard image as a component and an input pattern matrix with a feature amount of an input image as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrices as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrices, obtaining a length between the specified component and a component in each of the pattern matrices, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is greater than the component value of the standard pattern matrix, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix and the standard pattern matrix when the component value of the input pattern matrix is smaller than the component value of the standard pattern matrix;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrices being made to move to the position of each component in calculating the value of the difference between the kurtoses;

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix and the input pattern matrix; and

(g) comparing the obtained value of the geometric distance with an arbitrarily set acceptable value, judging that the input image is not the standard image when the value of the geometric distance exceeds the acceptable value, and judging that the input image is the standard image when the value of the geometric distance is within the acceptable value.

12. The method for recognizing an image according to claim 11, wherein the reference distribution is a normal distribution.

13. A method for detecting a similarity between solids, comprising the steps of:

(a) preparing a standard pattern matrix layer with a feature amount of a standard solid as a component and an input pattern matrix layer with a feature amount of an input solid as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrix layers as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrix layers, obtaining a length between the specified component and a component in each of the pattern matrix layers, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is greater than the component value of the standard pattern matrix layer, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is smaller than the component value of the standard pattern matrix layer;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrix layers being made to move to the position of each component in calculating the value of the difference between the kurtoses; and

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix layer and the input pattern matrix layer.

14. The method for detecting a similarity between solids according to claim 13, wherein the reference distribution is a normal distribution.

15. A method for recognizing a solid, comprising the steps of:

(a) preparing a standard pattern matrix layer with a feature amount of a standard solid as a component and an input pattern matrix layer with a feature amount of an input solid as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrix layers as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrix layers, obtaining a length between the specified component and a component in each of the pattern matrix layers, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is greater than the component value of the standard pattern matrix layer, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is smaller than the component value of the standard pattern matrix layer;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrix layers being made to move to the position of each component in calculating the value of the difference between the kurtoses;

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix layer and the input pattern matrix layer; and

(g) comparing the obtained value of the geometric distance with an arbitrarily set acceptable value, judging that the input solid is not the standard solid when the value of the geometric distance exceeds the acceptable value, and judging that the input solid is the standard solid when the value of the geometric distance is within the acceptable value.

16. The method for recognizing a solid according to claim 15, wherein the reference distribution is a normal distribution.

17. A method for detecting a similarity between moving pictures, comprising the steps of:

(a) preparing a standard pattern matrix layer with a feature amount of a standard moving picture as a component and an input pattern matrix layer with a feature amount of an input moving picture as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrix layers as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrix layers, obtaining a length between the specified component and a component in each of the pattern matrix layers, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is greater than the component value of the standard pattern matrix layer, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is smaller than the component value of the standard pattern matrix layer;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrix layers being made to move to the position of each component in calculating the value of the difference between the kurtoses; and

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix layer and the input pattern matrix layer.

18. The method for detecting a similarity between moving pictures according to claim 17, wherein the reference distribution is a normal distribution.

19. A method for recognizing a moving picture, comprising the steps of:

(a) preparing a standard pattern matrix layer with a feature amount of a standard moving picture as a component and an input pattern matrix layer with a feature amount of an input moving picture as a component;

(b) preparing an arbitrarily-shaped distribution such as a rectangle having a variance different for each specified component of the pattern matrix layers as a reference distribution, and preparing a positive reference pattern vector and a negative reference pattern vector, each having a value of the reference distribution as a component;

(c) with regard to each component of the pattern matrix layers, obtaining a length between the specified component and a component in each of the pattern matrix layers, calculating numbers of the components of the positive reference pattern vector and the negative reference pattern vector, the components being proximate to positions apart by the length from centers of the positive reference pattern vector and the negative reference pattern vector, increasing a value of the component of the number in the positive reference pattern vector by an absolute value of a difference between component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is greater than the component value of the standard pattern matrix layer, and increasing a value of the component of the number in the negative reference pattern vector by the absolute value of the difference between the component values of the input pattern matrix layer and the standard pattern matrix layer when the component value of the input pattern matrix layer is smaller than the component value of the standard pattern matrix layer;

(d) calculating a value of a difference between a kurtosis of the positive reference pattern vector and a kurtosis of the negative reference pattern vector;

(e) obtaining a value of a difference between the kurtoses while the specified component of the pattern matrix layers being made to move to the position of each component in calculating the value of the difference between the kurtoses;

(f) obtaining any one of a sum of squares of the values of the differences between the kurtoses and a square root of the sum of the squares as a value of a geometric distance between the standard pattern matrix layer and the input pattern matrix layer; and

(g) comparing the obtained value of the geometric distance with an arbitrarily set acceptable value, judging that the input moving picture is not the standard moving picture when the value of the geometric distance exceeds the acceptable value, and judging that the input moving picture is the standard moving picture when the value of the geometric distance is within the acceptable value.

20. The method for recognizing a moving picture according to claim 19, wherein the reference distribution is a normal distribution.