Full-search-equivalent method for matching data and a vector quantizer utilizing such method

• US 4,958,225 A
• Filed: 06/09/1989
• Issued: 09/18/1990
• Est. Priority Date: 06/09/1989
• Status: Expired due to Fees

• Current Assignee: Utah State University Foundation UMC
• Original Assignee: Utah Valley University
• Inventors: Bi, Qi, Stiles, Gardiner S., Huang, Chien M.
• Primary Examiner(s): Groody, James J.
• Assistant Examiner(s): Kostak, Victor R.
• Application Number:
  US07/363,554
• Time in Patent Office: 466 Days
• Field of Search: 358/133, 358/135, 358/136, 358/105, 375/122, 341/106
• US Class Current: 375/240.22
• CPC Class Codes: G06T 9/008 Vector quantisation
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94
• CPC Class Codes: H04N 19/94 Vector quantisation

Abstract

A method for compressing data employing vector quantization is achieved by calculating the norm of an input vector and identifying a reference codebook vector which has a norm which is closest to the norm of the input vector. The distance between the input vector and the reference codebook vector selected is computed and employed to identify a vector space about the reference vector containing a subset of codebook vectors one or more of which may be closer to the input vector than the initially selected reference vector. The closest codebook vector is selected iteratively without the necessity of searching every vector in the codebook.

Claims

1. A method for searching a set of representative elements to locate one element which closely matches an input element, the method comprising:

(a) calculating the value of a characteristic measure for the input element,

(b) identifying a reference element, A_j, having a value of the characteristic measure which satisfies a first criterion,

(c) computing a distance, h_I,j, between the input element and the reference element A_j,

(d) identifying a subject, S, of representative elements for which a measure of the representative elements in the subset S satisfies a second criterion with respect to the reference element,

(e) selecting a representative element from the subset S which closely matches the input element.

2. A method for compressing data comprising the steps of:

(a) generating a set of codebook vectors, each codebook vector identified by a unique identification code,

(b) organizing the data as a set of input vectors,

(c) for each input vector;

(i) calculating the value of a characteristic measure for the input vector,

(ii) identifying a reference codebook vector, A_j, having a value of a measure which satisfies a first criterion,

(iii) computing a distance, h_I,j , between the input vector. and the reference codebook vector A_j,

(iv) identifying a subset, S, of codebook vectors for which a measure of codebook vectors of the subset S satisfies a second criterion with respect to the reference vector,

(v) selecting a codebook vector from the subset S which closely matches the input vector,

(d) expressing each input vector as the identification code of the selected, closely matching codebook vector.

3. A method in accordance with claim 2 wherein the characteristic measure for the input vector comprises a vector norm.

4. A method in accordance with claim 2 wherein identifying a reference codebook vector, A_j, comprises the step of:

identifying a reference codebook vector, A_j, having a norm which most nearly equals the norm of the input vector.

5. A method in accordance with claim 2 wherein the step of calculating the characteristic measure of an input vector executes fewer computational steps than calculating the distance between an input vector and another vector.

6. A method in accordance with claim 2 wherein identifying a reference codebook vector, A_j, comprises the step of:

comparing the value of the characteristic measure of the input vector to calculated and stored values of the characteristic measures of the codebook vectors.

7. A method in accordance with claim 2 wherein the codebook vectors of the subset S have norms in the range |I|-h_I,j to |I|+h_I,j where |I| is the norm of the input vector and h_I,j is the distance between the reference codebook vector A_j and the input vector I.

8. A method in accordance with claim 2 wherein the distance between the reference codebook vector A_j and codebook vectors of the subset S is less than twice the distance h_I,j between the reference codebook vector A_j and the input vector.

9. A method in accordance with claim 2 wherein

(a) the codebook vectors of the subset S have norms in the range |I|-h_I,j to |I|+h_I,j where |I| is the norm of the input vector and h_I,j is the distance between the reference codebook vector A_j and the input vector I, and

(b) the distance between the reference codebook vector A_j and codebook vectors of the subset S is less than twice the distance h_I,J between the reference codebook vector A_j and the input vector.

10. A method in accordance with claim 2 wherein the distance measure satisfies the following condition:

space="preserve" listing-type="equation">d(x,A.sub.i)≦

d(x,A.sub.j)+d(A.sub.j,A.sub.i)

where d(x,A_i) is the distance measure, and X_i is an input vector, and A_j and A_i are codebook vectors.

11. A method in accordance with claim 2 further comprising the step of:

(a) calculating the distances among codebook vectors in advance of expressing the data as a set of input vectors, and (b) comparing the distances among codebook vectors calculated in advance with the condition of the second criterion.

12. A method in accordance with claim 2 wherein selecting a codebook vector from the subset S comprises the step of:

selecting the codebook vector from the subset S which has the minimum distance to the input vector.

13. A method in accordance with claim 2 wherein selecting a codebook vector from the subset S comprises the steps of:

(a) computing a distance h_I,k between the input vector and a codebook vector A_k of the subset S,

(b) eliminating the element A_k from the set S if the distance h_I,k between the input vector and the codebook vector is greater than or equal to the distance h_I,j between the input vector and the reference codebook vector,

(c) forming a new subset S'"'"' by eliminating the element A_j from the subset S and designating A_k to be the new reference vector if the distance h_I,k between the input vector and the codebook vector is less than the distance h_I,j between the input vector and the reference codebook vector, and

(d) repeating steps (a) through (c) until all but one codebook vector has been eliminated.

14. A method in accordance with claim 13 wherein forming a new set S'"'"' further comprises the step of:

eliminating from the subset S those codebook vectors whose norm is outside the range |I|-h_I,j to |I| +h_I,j where h_I,j is the distance between the reference codebook vector A_j and the input vector I.

15. A method in accordance with claim 13 wherein forming a new set S'"'"' further comprises the step of:

eliminating from the subset S those codebook vectors whose distance to the new reference vector A_k is greater than or equal to twice the distance h_I,k between the input vector and the new reference vector.

16. A method in accordance with claim 13 wherein forming a new set S'"'"' further comprises the step of:

retaining in the subset S those codebook vectors for which;

(a) the norm is in the range |I|-h_I,j to |I|+h_I,j where h_I,j is the distance between the reference codebook vector A_j and the input vector I, and

(b) the distance to the new reference vector A_k is greater than or equal to twice the distance h_I,k between the input vector and the new reference vector.

17. An apparatus for compressing data in the form of an input vector into an identification code of one of a set of codebook vectors, said apparatus comprising:

(a) input means for receiving the input vector and codebook vectors,

(b) memory means connected to the input means for storing the input vector and codebook vectors,

(d) vector processor means connected to the memory means, said processor means comprising;

(i) measuring means responsive to the input vector for calculating the value of a characteristic measure for the input vector;

(ii) identification means responsive to the calculated characteristic measure for the input vector and to codebook vectors for identifying a reference codebook vector, A_j, having a value of the characteristic measure which satisfies a first criterion;

(iii) computing means responsive to the identified reference codebook vector, A_j, and to the input vector for computing a distance, h_I,j, between the input vector and the reference codebook vector A_j ;

(iv) partitioning means responsive to the computed distance h_I,j and to codebook vectors having identifying a subset, S, of codebook vectors having measures which satisfy a second criterion with repsect to the input vector,

(v) comparing means responsive to the subset S of codebook vectors and to the input vector for selecting a codebook vector from the subset S, which closely matches the input vector,

(e) output means connected to the processor means for generating an output signal, said signal output representative of the identification code of the closely matching codebook vector.

18. An apparatus in accordance with claim 17 wherein said memory means further comprises characteristic measure memory means for storing values of the characteristic measures of codebook vectors.

19. An apparatus in accordance with claim 17 wherein said memory means further comprises distance matrix memory means for storing values of the distances between codebook vectors.

20. A system for compressing and regenerating input vector data using codebook vectors, said system comprising:

(a) a source of input vector data;

(c) a vector quantizer connected to said source for quantizing said data, said vector quantizer comprising;

(i) input means for receiving an input vector,

(11) memory means connected to the input means for storing said input vector and codebook vectors,

(iii) vector processor means connected to the memory means, said processor means comprising;

(A) measuring means responsive to the input vector for calculating the value of a characteristic measure for the input vector;

(B) identification means responsive to the calculated characteristic measure for the input vector and to codebook vectors for identifying a reference codebook vector, A_j, having a value of the characteristic measure which satisfies a first criterion;

(C) computing means responsive to the identified reference codebook vector, A_j, and to the input vector for computing a distance, h_I,j, between the input vector and the reference codebook vector A_j ;

(D) partitioning means responsive to the computed distance h_i,j and to codebook vectors for identifying a subset, S, of codebook vectors having measures which satisfy a second criterion with respect to the reference vector,

(E) comparing means responsive to the subset S of codebook vectors and to the input vector for selecting a codebook vector from the subset S, which closely matches the input vector;

(iv) output means connected to the processor means for generating a compressed signal, said compressed signal representative of the identification code of the closely matching codebook vector;

(e) regeneration means for approximately regenerating said data from said compressed signal.

21. An apparatus in accordance with claim 20 wherein said memory means further comprises characteristic measure memory means for storing values of the characteristic measures of codebook vectors.

22. An apparatus in accordance with claim 20 wherein said memory means further comprises distance matrix memory means for storing values of the distances between codebook vectors.

23. An apparatus for compressing data in the form of an input vector into an identification code of one of a set of codebook vectors, said apparatus comprising:

(a) input means for receiving the input means for storing the input data vector and for storing codebook vectors,

(b) memory means connected to the input means for storing the input data vector and for storing codebook vectors,

(d) vector processor means connected to the memory means for calculating the value of a characteristic measure for an input vector;

identifying a reference codebook vector, A_j, having a value of the characteristic measure which satisfies a first criterion;

computing a distance, h_I,j, between the input vector and the reference codebook vector A_j ;

identifying a subset, S, of codebook vectors for which a measure of the codebook vectors satisfies a second criterion with respect to the reference vector; and

selecting a codebook vector from the subset S which closely matches the input vector,

(e) output means connected to the vector processor means for generating an output signal representative of the identification code of the closely matching codebook vector.