Similar time series detection method and apparatus, program and recording medium

• US 7,035,775 B2
• Filed: 07/03/2003
• Issued: 04/25/2006
• Est. Priority Date: 07/09/2002
• Status: Active Grant

• Current Assignee: Sony Corporation (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.)
• Inventors: Nishiguchi, Masayuki, Abe, Mototsugu
• Primary Examiner(s): Raymond, Edward
• Application Number:
  US10/613,789
• Publication Number:
  US 20040098225A1
• Time in Patent Office: 1,027 Days
• Field of Search: 702/189, 702/176
• US Class Current: 702/189
• CPC Class Codes: G06V 10/75 Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

A method, an apparatus and a program for detecting similarity between two time-domain signals, at a high speed, and a recording medium on which is recorded such program. The process for detecting the similarity is split into a stage for calculating the similarity between feature vectors at each discrete time t and a stage for counting the number of similar vectors in the entire time series. In the calculating stage, the distances between the components for the same frame numbers of the feature vector time series g_t, f_t are hierarchically integrated. If the ultimate integrated value is lower than the distance threshold value S, the feature vector time series g_t, f_t are determined to be similar. If the integrated value exceeds the threshold value part way in the course of the hierarchical integration, the feature vector time series g_t, f_t are determined to be dissimilar and the integrating calculations for the remaining components are discontinued. In the counting stage, the number of the feature vectors f determined to be similar and the number D of the feature vectors f determined to be dissimilar are hierarchically counted with respect to each frame. If the ultimate number of dissimilar vectors D is smaller than the threshold value (T-U), the feature vector time series g_t, f_t are determined to be similar. If the number of dissimilar vectors D exceeds the threshold value (T-U) part way in the integrating calculations, the feature vector time series g_t, f_t are determined to be dissimilar and the similarity calculations for the remaining frames are discontinued.

Claims

1. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being a time series of vectors, said method comprising:

a vector decision step of calculating a degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting,

wherein,

said degree of similarity is the distance between said paired vectors; and

said vector decision step includes (a) a hierarchical distance calculating step of hierarchically calculating the distance between said paired vectors, (b) a threshold value comparing step of comparing an integrated distance value calculated in each hierarchy of said hierarchical distance calculating step to a first threshold value pertinent to the distance, and (c) a decision step of determining whether or not said paired vectors are similar to each other in response to the results of comparison in said threshold comparing step

if, in said threshold comparing step, the integrated distance value calculated up to the last hierarchy is smaller than said first threshold value, said paired vectors are determined to be similar in said decision step, and

if said integrated distance value calculated up to a given hierarchy exceeds said first threshold value, said paired vectors are determined to be dissimilar, and the distance calculations are discontinued for the remaining hierarchies.

2. The method for detecting similar time series according to claim 1, wherein,

in said hierarchical distance calculating step, the distance between components making up the paired vectors is calculated hierarchically, and

if, in said threshold comparing step, the integrated distance value calculated up to a given hierarchy is lower than the first threshold value, the next calculations of the distance between the components are carried out.

3. The method for detecting similar time series according to claim 2, wherein

said vector decision step further includes a transform step of applying a preset transform to said paired vectors;

and

in said hierarchical distance calculating step, the distance between the paired vectors, transformed by said transform step, is calculated in a preset sequence which is based on said preset transform.

4. The method for detecting similar time series according to claim 3, wherein

said preset transform is a transform of re-arraying said components making up said paired vectors in the order of the magnitudes of the variance of said components;

and

in said hierarchical distance calculating step, the distance between the paired vectors, transformed by said transform step, is calculated in the order of the decreasing magnitudes of the variance of the components.

5. The method for detecting similar time series according to claim 3, wherein

said preset transform is the discrete cosine transform or the discrete Fourier transform;

and

in said hierarchical distance calculating step, the distance calculations between the paired vectors transformed by said transform step are carried out sequentially beginning from the low frequency components.

6. The method for detecting similar time series according to claim 3, wherein

said preset transform is the Walsh-Hadamard transform;

and

in said hierarchical distance calculating step, the distance calculations between the paired vectors transformed by said transform step are carried out sequentially beginning from the low sequency components.

7. The method for detecting similar time series according to claim 3, wherein

said preset transform is the Karhunen-Loeve transform;

and

in said hierarchical distance calculating step, the distance calculations between the paired vectors transformed by said transform step are carried out sequentially beginning from components of high characteristic value.

8. The method for detecting similar time series according to claim 3, wherein

said vector decision step further includes a splitting step of taking out components making up each of said paired vectors, transformed by said transform step, in said preset sequence, and splitting said component into a plurality of hierarchical partial vectors;

in said hierarchical distance calculating step, distance calculations between the respective components forming partial vectors are carried out hierarchically beginning from the partial vector of the uppermost hierarchy; and

when the integrated value of the distance calculated for the totality of the components making up the partial vectors up to a given hierarchy is lower than said first threshold value, the distance calculations between the components forming the partial vectors of the hierarchy lower by one hierarchy are carried out.

9. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating a degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said time series decision step includes (a) a hierarchical counting step of hierarchically counting the number of said similar vectors and/or said dissimilar vectors, (b) a threshold value comparing step of comparing an integrated value of the number of said similar vectors and/or said dissimilar vectors, obtained in each hierarchy of said hierarchical counting step, to a second threshold value of the number of similar vectors, and (c) a decision step of determining the similarity of said paired time series patterns, responsive to the results of comparison in said threshold value comparing step,

if, in said threshold value comparing step, the integrated value of the number of said similar vectors calculated up to the last hierarchy is larger than said second threshold value, said paired time series patterns are determined to be similar in said decision step,

if the integrated value of the number of the dissimilar vectors, calculated up to a given hierarchy, exceeds the total number of said vectors in said time series pattern less said second threshold value, said paired time series patterns are determined to be dissimilar, and control is performed to discontinue the calculations for the remaining hierarchies.

10. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating a degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector is the vectorized features of audio signals, and

said time series pattern is a time series of vectorized power spectrum coefficients in a preset time domain of said audio signals.

11. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating a degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector is the vectorized features of audio signals, and

said time series pattern is a time series of vectorized linear predictive coefficients in a preset time domain of said audio signals.

12. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating a degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector is the vectorized features of encoded audio signals, and

said time series pattern is a time series of vectorized parameters representing the intensity of the frequency components in each frame of said encoded audio signals.

13. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector is the vectorized features of video signals, and

said time series pattern is a time series of vectorized signal values of a representative picture in a preset time domain of said video signals, an average picture of frame pictures in said preset time domain or a small-sized picture, obtained on splitting said representative picture or said average picture in terms of a preset block as a unit.

14. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector is the vectorized features of video signals, and

said time series pattern is a time series of vectorized histograms for luminance and/or chroma in a frame picture in a preset time domain of said video signals.

15. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector is the vectorized features of encoded video signals, and

said time series pattern is a time series of vectorized signal values of DC components of respective blocks as encoding units of an intra-frame encoded picture in direct proximity of a preset time domain of said encoded video signals.

16. A method for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said method comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

one of said paired time series patterns is sliced from a time series of a vector longer than said time series patterns and wherein similarity is sequentially checked as the slicing position is updated.

17. An apparatus for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said apparatus comprising:

vector decision means for calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

time series decision means for counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision means and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said vector decision means includes (a) hierarchical distance calculating means for hierarchically calculating the distance between said paired vectors, (b) threshold value comparing means for comparing an integrated distance value calculated in each hierarchy of said hierarchical distance calculating means to a first threshold value pertinent to the distance, and (c) decision means for determining whether or not said paired vectors are similar to each other, responsive to the results of comparison in said threshold comparing means

said degree of similarity is the distance between said paired vectors, and

said decision means being configured such that, if, as a result of comparison by said threshold comparing means, the integrated distance value calculated up to the last hierarchy is smaller than said first threshold value, said paired vectors are determined to be similar, and

if an integrated distance value calculated up to a given hierarchy exceeds said first threshold value, said paired vectors are determined to be dissimilar, and the distance calculations for the remaining hierarchies is ended.

18. The apparatus for detecting similar time series according to claim 17 wherein

said hierarchical distance calculating is means configured to calculate the distance between components making up the paired vectors hierarchically, and

if, as a result of comparison by said threshold comparing means, the integrated distance value calculated up to a given hierarchy is lower than the first threshold value, the next calculations of the distance between the components are carried out.

19. The apparatus for detecting similar time series according to claim 18, wherein

said vector decision means further includes a transform means for applying a preset transform to said paired vectors, and

said hierarchical distance calculating means is configured to calculate the distance between the paired vectors, transformed by said transform means, in a preset sequence which is based on said preset transform.

20. The apparatus for detecting similar time series according to claim 19, wherein

said vector decision means further includes splitting means for taking out components making up each of said paired vectors, transformed by said transform means, in said preset sequence, and splitting said components into a plurality of hierarchical partial vectors,

said hierarchical distance calculating means is configured to execute distance calculations between the respective components forming partial vectors hierarchically beginning from the partial vector of the uppermost hierarchy, and

the distance calculations between the components forming the partial vectors of the hierarchy lower by one hierarchy are carried out when the integrated value of the distance calculated for the totality of the components making up the partial vectors up to a given hierarchy is lower than said first threshold value.

21. An apparatus for detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said apparatus comprising:

vector decision means for calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

time series decision means for counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision means and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting, wherein,

said time series decision means includes (a) hierarchical counting means for hierarchically counting the number of said similar vectors and/or said dissimilar vectors, (b) threshold value comparing means for comparing an integrated value of the number of said similar vectors and/or said dissimilar vectors, obtained in each hierarchy of said hierarchical counting means, to a second threshold value of the number of similar vectors, and (c) decision means for determining the similarity of said paired time series patterns, responsive to the results of comparison in said threshold value comparing means,

if, as a result of comparison by said threshold value comparing means, the integrated value of the number of said similar vectors calculated up to the last hierarchy is larger than said second threshold value, said decision means determines the paired time series patterns to be similar, and

if the integrated value of the number of the dissimilar vectors, calculated up to a given hierarchy, exceeds the total number of said vectors in said time series pattern less said second threshold value, said decision means determines the paired time series patterns to be dissimilar, and control is performed to discontinue the distance calculations for the remaining hierarchies.

22. A program for having a computer execute the processing of detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said program comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said degree of similarity is the distance between said paired vectors,

said vector decision step includes (a) a hierarchical distance calculating step of hierarchically calculating the distance between said paired vectors, (b) a threshold value comparing step of comparing an integrated distance value calculated in each hierarchy of said hierarchical distance calculating step to a first threshold value pertinent to the distance and (c) a decision step of determining whether or not said paired vectors are similar to each other, responsive to the results of comparison in said threshold comparing step;

if, in said threshold comparing step, the integrated distance value calculated up to the last hierarchy is smaller than said first threshold value, said paired vectors are determined to be similar in said decision step, and

if an integrated distance value calculated up to a given hierarchy exceeds said first threshold value, said paired vectors are determined to be dissimilar, and control is performed to discontinue the distance calculations for the remaining hierarchies.

23. The program for detecting similar time series according to claim 22, wherein,

in said hierarchical distance calculating step, the distance between components making up the paired vectors is calculated hierarchically;

and,

if, in said threshold comparing step, the integrated distance value calculated up to a given hierarchy is lower than the first threshold value, the next calculations of the distance between the components are carried out.

24. The program for detecting similar time series according to claim 23, wherein

said vector decision step further includes a transform step of applying a preset transform to said paired vectors;

and

in said hierarchical distance calculating step, the distance between the paired vectors, transformed by said transform step, is calculated in a preset sequence which is based on said preset transform.

25. The program for detecting similar time series according to claim 24, wherein

said vector decision step further includes a splitting step of taking out components making up each of said paired vectors, transformed by said transform step, in said preset sequence, and splitting said component into a plurality of hierarchical partial vectors;

and

in said hierarchical distance calculating step, distance calculations between the respective components forming partial vectors are carried out hierarchically beginning from the partial vector of the uppermost hierarchy;

when the integrated value of the distance calculated for the totality of the components making up the partial vectors up to a given hierarchy is lower than said first threshold value, the distance calculations between the components forming the partial vectors of the hierarchy lower by one hierarchy are carried out.

26. A program for having a computer execute the processing of detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said program comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said time series decision step includes (a) a hierarchical counting step of hierarchically counting the number of said similar vectors and/or said dissimilar vectors, (b) a threshold value comparing step of comparing an integrated value of the number of said similar vectors and/or said dissimilar vectors, obtained in each hierarchy of said hierarchical counting step, to a second threshold value of the number of similar vectors, and (c) a decision step of determining the similarity of said paired time series patterns, responsive to the results of comparison in said threshold value comparing step,

if, in said threshold value comparing step, the integrated value of the number of said similar vectors calculated up to the last hierarchy is larger than said second threshold value, said paired time series patterns are determined to be similar in said decision step, and

if the integrated value of the number of the dissimilar vectors, calculated up to a given hierarchy, exceeds the total number of said vectors in said time series pattern less said second threshold value, said paired time series patterns are determined to be dissimilar, and control is performed to discontinue the calculations for the remaining hierarchies.

27. A computer-controllable recording medium having recorded thereon a program for having a computer execute the processing of detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said program comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said degree of similarity is the distance between said paired vectors,

said vector decision step includes (a) a hierarchical distance calculating step of hierarchically calculating the distance between said paired vectors, (b) a threshold value comparing step of comparing an integrated distance value calculated in each hierarchy of said hierarchical distance calculating step to a first threshold value pertinent to the distances and (c) a decision step of determining whether or not said paired vectors are similar to each other, responsive to the results of comparison in said threshold comparing step;

if, in said threshold comparing step, the integrated distance value calculated up to the last hierarchy is smaller than said first threshold value, said paired vectors are determined to be similar in said decision step, and

if an integrated distance value calculated up to a given hierarchy exceeds said first threshold value, said paired vectors are determined to be dissimilar, and control is performed to discontinue the distance calculations for the remaining hierarchies.

28. The recording medium according to claim 27, wherein,

in said hierarchical distance calculating step, the distance between components making up the paired vectors is calculated hierarchically, and

if, in said threshold comparing step, the integrated distance value calculated up to a given hierarchy is lower than the first threshold value, the next calculations of the distance between the components are carried out.

29. The recording medium according to claim 27, wherein

said vector decision step further includes a transform step of applying a preset transform to said paired vectors;

and

in said hierarchical distance calculating step, the distance between the paired vectors, transformed by said transform step, is calculated in a preset sequence which is based on said preset transform.

30. The recording medium according to claim 29, wherein

said vector decision step further includes a splitting step of taking out components making up each of said paired vectors, transformed by said transform step, in said preset sequence, and splitting said component into a plurality of hierarchical partial vectors;

and

in said hierarchical distance calculating step, distance calculations between the respective components forming partial vectors are carried out hierarchically beginning from the partial vector of the uppermost hierarchy;

when the integrated value of the distance calculated for the totality of the components making up the partial vectors up to a given hierarchy is lower than said first threshold value, the distance calculations between the components forming the partial vectors of the hierarchy lower by one hierarchy are carried out.

31. A computer-controllable recording medium having recorded thereon a program for having a computer execute the processing of detecting similar time series of a pair of time series patterns, with the time series patterns being time series of vectors, said program comprising:

a vector decision step of calculating the degree of similarity between a pair of vectors of said paired time series patterns and providing a decision as to whether or not said paired vectors are similar to each other, based on said degree of similarity; and

a time series decision step of counting the number of similar vectors decided to be similar and/or dissimilar vectors decided to be dissimilar in said vector decision step and providing a decision as to whether or not the paired time series patterns are similar, based on the results of the counting wherein,

said time series decision step includes (a) a hierarchical counting step of hierarchically counting the number of said similar vectors and/or said dissimilar vectors, (b) a threshold value comparing step of comparing an integrated value of the number of said similar vectors and/or said dissimilar vectors, obtained in each hierarchy of said hierarchical counting step, to a second threshold value of the number of similar vectors, and (c) a decision step of determining the similarity of said paired time series patterns, responsive to the results of comparison in said threshold value comparing step,

if, in said threshold value comparing step, the integrated value of the number of said similar vectors calculated up to the last hierarchy is larger than said second threshold value, said paired time series patterns are determined to be similar in said decision step, and

if the integrated value of the number of the dissimilar vectors, calculated up to a given hierarchy, exceeds the total number of said vectors in said time series pattern less said second threshold value, said paired time series patterns are determined to be dissimilar, and control is performed to discontinue the calculations for the remaining hierarchies.