Zero-search, zero-memory vector quantization

• US 8,185,390 B2
• Filed: 04/23/2009
• Issued: 05/22/2012
• Est. Priority Date: 06/26/2003
• Status: Active Grant

• Current Assignee: Promptu Systems Corporation
• Original Assignee: Promptu Systems Corporation
• Inventors: Printz, Harry
• Primary Examiner(s): JACKSON, JAKIEDA R
• Application Number:
  US12/429,122
• Publication Number:
  US 20090208120A1
• Time in Patent Office: 1,125 Days
• Field of Search: 704/236
• US Class Current: 704/236
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 19/00 Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis in musical instruments G10H
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94
• CPC Class Codes: H04N 19/94 Vector quantisation

Abstract

The invention comprises a method for lossy data compression, akin to vector quantization, in which there is no explicit codebook and no search, i.e. the codebook memory and associated search computation are eliminated. Some memory and computation are still required, but these are dramatically reduced, compared to systems that do not exploit this method. For this reason, both the memory and computation requirements of the method are exponentially smaller than comparable methods that do not exploit the invention. Because there is no explicit codebook to be stored or searched, no such codebook need be generated either. This makes the method well suited to adaptive coding schemes, where the compression system adapts to the statistics of the data presented for processing: both the complexity of the algorithm executed for adaptation, and the amount of data transmitted to synchronize the sender and receiver, are exponentially smaller than comparable existing methods.

Claims

1. A computer-implemented method for data compression comprising the steps of:

via a computational device, receiving at least one signal in any of digitized and non-digitized form, said received at least one signal representing at least one data object to be compressed;

responsive to receiving said signal in non-digitized form, via a computational device, digitizing said at least one received signal;

via a computational device, extracting at least one information element from said at least one digitized signal;

via a computational device, establishing an implicit codebook comprising a set of code points for representing said at least one information element extracted from said at least one digitized signal, each of said at least one information element constituting a vector, said set of code points implicitly represented by a hypercube radius vector;

via a computational device, inspecting the signs of each information element to determine in which orthant said each information element lies to determine the implicit code point to represent said each information element;

via a computational device, determining an index of the implicit code point associated with said each information element; and

via a computational device, transmitting or storing said each resulting index as the compressed representation of said each information element.

2. The method of claim 1, for decompressing an index obtained via compression with respect to the given hypercube radius vector, further comprising the steps of:

via a computational device, inspecting said index to determine in which orthant the corresponding implicit code point lies; and

via a computational device, outputting the corresponding implicit code point to produce at least a partial facsimile of said each information element.

3. The method of claim 2, for decompressing an index i, obtained via compression with respect to a hypercube radius vector α

=α

₁,α

₂, . . . , α

_D, said method further comprising the steps of;

via a computational device, obtaining an index i for decompression;

via a computational device, setting

4. The method of claim 1, further comprising the step of:

via a computational device, finding an optimal D-hypercube codebook, with respect to typical data.

5. The method of claim 1, wherein said data compression comprises compressing digitized human speech;

and

further comprising the steps of;

via a computational device, transmitting or storing said compressed, digitized human speech;

via a computational device, receiving or retrieving said compressed, digitized human speech;

via a computational device, decompressing said compressed, digitized human speech; and

via a computational device, processing said digitized human speech with an automatic speech recognition system.

6. The method of claim 1, further comprising the step of:

via a computational device, applying an invertible transform to optionally translated information elements to be compressed, said transform determined so that an aggregate of typical data to be compressed will approximate a uniform distribution within a D-dimensional sphere, centered at the origin.

7. The method of claim 6, further comprising the step of:

via a computational device, finding an optimal D-hypercube codebook, with respect to typical transformed data.

8. The method of claim 7, further comprising the step of:

via a computational device, determining an optimal hypercube radius vector α

=α

₁,α

₂, . . . , α

_D, this vector defining said hypercube codebook, that yields minimal mean square coding distortion among all D-hypercube codebooks, with respect to typical transformed data.

9. The method of claim 8, further comprising the step of:

via a computational device, computing an optimal hypercube radius vector α

=α

₁,α

₂, . . . , α

_D from an example data collection E comprised of E=|E| typical data vectors ν

, to which has been applied the invertible transform of claim 5, to yield a collection U comprised of U=|U|=|E|=E typical data vectors u, by computing for each dimension i=1, . . . , D the quantity

10. The method of claim 6, said invertible transform comprising a symmetrizing transform.

11. The method of claim 10, said symmetrizing transform comprising a whitening transform.

12. The method of claim 11, further comprising the steps of:

via a computational device, projecting all data into a D−

1-dimensional space;

via a computational device, processing a resulting data set; and

wherein only D−

1 bits are transmitted for each vector processed.

13. The method of claim 11, further comprising the step of:

via a computational device, representing the whitening transform in a memory efficient manner as the product of a diagonal matrix, said matrix represented by only its D non-zero elements, and an orthogonal matrix;

the inverse of this transform therefore obtainable via a computational device as the product, in the opposite order, of a diagonal matrix of D reciprocals of the original D non-zero elements, and the transpose of the original orthogonal matrix.

14. The method of claim 11, further comprising the step of:

via a computational device, determining an optimal hypercube radius vector α

=α

₁,α

₂, . . . , α

_D that defines said hypercube codebook and yields minimal mean square coding distortion among all D-hypercube codebooks for typical data to be compressed.

15. The method of claim 14, further comprising the step of:

via a computational device, computing an optimal hypercube radius vector α

=α

₁,α

₂, . . . , α

_D from an example data collection E comprised of E=|E| typical data vectors ν

, by computing for each dimension i=1, . . . , D the quantity

16. The method of claim 7, for compressing a vector ν

, further comprising the steps of;

via a computational device, obtaining a vector ν

for compression;

via a computational device, computing u=Tν

, where u is denoted u₁,u₂, . . . , u_p;

where T is the symmetrizing transform;

via a computational device, forming a D-bit binary number i as a bitwise concatenation i=m(u_D)m(u_D-1) . . . m(u₂)m(u₁);

where the j th bit of i is 0 if u_j is zero or positive, and 1 if it is negative; and

via a computational device, transmitting or storing i.

17. The method of claim 16, for decompressing an index i, obtained via compression with respect to a hypercube radius vector α

=α

₁,α

₂, . . . , α

_D, said method further comprising the steps of;

via a computational device, obtaining an index i for decompression;

via a computational device, setting

18. The compression method of claim 10, further comprising the step of:

via a computational device, incorporating a rotation in the symmetrizing transform to lower distortion.

19. The method of claim 18, further comprising the step of:

via a computational device, finding an optimal hypercube radius vector with respect to rotated symmetrized typical data.

20. The method of claim 1, for compressing a vector ν

=ν

₁,ν

₂, . . . , ν

_D, further comprising the steps of;

via a computational device, obtaining a vector ν

=ν

₁,ν

₂, . . . , ν

_D for compression;

via a computational device, forming a D-bit binary number i as a bitwise concatenation

i=m(ν

_D)m(ν

_D-1) . . . m(ν

₂)m(ν

₁);

where the j th bit of i is 0 if ν

_j is zero or positive, and 1 if it is negative; and

via a computational device, transmitting or storing i.

21. The compression method of claim 1, further comprising the steps of:

via a computational device, increasing a number of implicit code points by increasing the number of hypercubes, wherein compression occurs with respect to a family of hypercube codebooks.

22. The compression method of claim 21, using a family of hypercubes A, each hypercube determined by its associated hypercube radius vector α

=α

₁,α

₂, . . . , α

_D, and using a symmetrizing transform T, said symmetrizing transformation comprising one of;

the identity transform;

a whitening transform;

or

some other symmetrizing transform;

further comprising the steps of;

given vector ν

to compress finding u=Tν

;

via a computational device, finding the orthant of u, encoded as i=m(u_D)m(u_D-1) . . . m(u₂)m(u₁);

via a computational device, finding, via explicit search within the orthant, a hypercube index k of the closest hypercube α

^kε

A; and

via a computational device, transmitting or storing a result of said search, in the form of the said hypercube index k so determined, along with the identity i of the orthant.

23. The method of claim 22, using a multiplicity of hypercubes A;

varying with respect to one another in hypercube radius vector, in orientation, or both;

said orientations being expressed by a rotation R^k associated to each hypercube radius vector α

^k, said rotation possibly being the identity; and

further comprising the steps of;

given vector ν

to compress finding each u^k=R^kTν

;

via a computational device, finding the orthant index i of u^k;

via a computational device, finding, via explicit search among the multiplicity of hypercubes, the hypercube index k yielding the minimal distortion; and

via a computational device, transmitting or storing a result of said search, in the form of the said hypercube index k so determined, along with the identity i of the orthant.

24. The method of claim 23, further comprising the steps of:

via a computational device, decompressing the pair comprised of hypercube index k and orthant index i;

by using hypercube index k to select an appropriate hypercube radius vector α

^k; and

via a computational device, inspecting the coded orthant i to yield an appropriate vertex of the α

^k hypercube;

wherein said vertex is taken as ũ

, from which {tilde over (ν

)}=T<sup>−

1</sup>(R^k)<sup>−

1</sup>ũ

is computed, and the value {tilde over (ν

)} returned as the result.

25. The method of claim 22, further comprising the steps of:

via a computational device, decompressing the pair comprised of hypercube index k and orthant index i;

by using hypercube index k to select an appropriate hypercube radius vector α

^k; and

via a computational device, inspecting the coded orthant i to yield an appropriate vertex of the α

^k hypercube;

wherein said vertex is taken as ũ

, from which {tilde over (ν

)}=T<sup>−

1</sup>ũ

is computed, and the value {tilde over (ν

)} returned as the result.

26. The method of claim 1, wherein said data compression comprises compressing at least one image;

and

further comprising the steps of;

via a computational device, transmitting or storing said at least one compressed, digitized image;

via a computational device, receiving or retrieving said at least one compressed, digitized image;

via a computational device, decompressing said at least one compressed, digitized image; and

via a computational device, processing said at least one digitized image with an automatic image-processing system.

27. A computer-implemented method for data compression and subsequent decompression comprising the steps of:

via a computational device, receiving at least one signal representing at least one data object in any of digitized and non-digitized form;

responsive to receiving said at least one signal in non-digitized form, via a computational device, digitizing said at least one received signal;

via a computational device, extracting at least one information element from said digitized signal;

via a computational device, establishing an implicit codebook comprising a set of code points for representing at least one information element extracted from said digitized signal, each of said at least one information element constituting a vector, said set of code points implicitly represented by a hypercube radius vector;

via a computational device, inspecting the signs of each information element to determine in which orthant said each information element lies to determine the implicit code point to represent said each information element;

via a computational device, determining an index of the implicit code point associated with said each information element;

via a computational device, transmitting or storing each resulting index as the compressed representation of each said information element; and

via a computational device, applying a decompression function to said compressed information elements to produce at least a partial facsimile of said at least one information element.

28. A computer-implemented method for finding a family A of K hypercube codebooks, comprising the steps of:

via a computational device, receiving at least one signal representing at least one data object in any of digitized and non-digitized form;

responsive to said at least one signal being received in non-digitized form, via a computational device, digitizing said at least one received signal;

via a computational device, extracting an example dataset U from said digitized signal;

beginning with a fixed number K of desired hypercubes, and said example dataset U;

via a computational device, mapping each element uε

U, where u=u₁, . . . , u_D, to the positive

orthant P^D, via the map p;

u₁, . . . , u_D→

|u₁|, |u₂|, . . . , |u_D|, yielding the set

U⁺={p(u)|uε

U};

via a computational device, selecting an initial set of K radius vectors A⁽⁰⁾={ α

₀⁽⁰⁾, . . . , α

_K-1⁽⁰⁾};

via a computational device, setting an iteration count i to 0;

via a computational device, establishing a termination condition τ

, which depends upon one or more of;

the number of iterations executed;

the closeness of match between a current radius vector collection A⁽ⁱ⁾ and U⁺; and

the improvement of a statistic over a previous iteration, wherein said dependence is expressed as τ

(i, A⁽ⁱ⁾,U⁺);

via a computational device, testing τ

(i,A⁽ⁱ⁾,U⁺); and

if the termination condition is satisfied, via a computational device, outputting A⁽ⁱ⁾ as the desired radius vector collection; and

via a computational device, stopping;

else, if the termination condition is not satisfied, via a computational device, computing a new radius vector collection A⁽ⁱ⁺¹⁾ as follows;

via a computational device, partitioning U⁺ into K sets S₀ . . . S_K-1, where

29. A computer-implemented method for data compression, comprising the steps of:

via a computational device, receiving at least one signal representing at least one data object in any of digitized and non-digitized form;

responsive to said at least one signal being received in non-digitized form, via a computational device, digitizing said at least one received signal;

via a computational device, extracting at least one information element from said digitized signal;

via a computational device, computing a compression function for said at least one information element extracted from said digitized signal by;

via a computational device, first applying a symmetrizing transform;

via a computational device, inspecting the signs of said transformed information elements to determine in which quadrant of an implicit codebook a corresponding implicit code point lies;

via a computational device, determining an index of an associated implicit code point for each said information element; and

via a computational device, transmitting or storing each said resulting index as the compressed representation of each said information element.

30. The method of claim 29, further comprising the steps of:

via a computational device, rotating the data to be compressed, to lower distortion, and then compressing the rotated data.

31. The method of claim 29, further comprising the step of:

via a computational device, incorporating a rotation in the symmetrizing transform, to lower distortion, and then compressing.

32. The method of claim 29, further comprising the step of:

via a computational device, establishing an implicit codebook comprising an implicitly defined set of vectors, hereafter called code points, which are symmetrically placed with respect to the origin;

wherein said code points are used for representing information elements.

33. The method of claim 32, further comprising the step of:

via a computational device, increasing a number of code points by increasing the number of hypercubes, wherein compression occurs with respect to a family of hypercube codebooks, and the preferred hypercube is found by explicit search, once the orthant of the vector to be compressed has been determined.

34. The method of claim 32, further comprising the step of:

via a computational device, increasing a number of code points by increasing the number of hypercubes, wherein compression occurs with respect to a family A of hypercube codebooks, the hypercubes varying with respect to one another in hypercube radius vector, in orientation, or both;

via a computational device, the selected hypercube and orthant index being found by explicit search among a set consisting of the preferred vertex of each hypercube, the preferred vertex of each hypercube being the one that lies in the same orthant of the vector to be compressed, for each hypercube.

35. A computer-implemented method for data compression comprising the steps of:

via a computational device, receiving at least one signal representing at least one spoken utterance in any of non-digitized and digitized form;

responsive to receiving said at least one signal in non-digitized form, via a computational device, digitizing said at least one received signal;

via a computational device, extracting at least one information element from said digitized signal;

via a computational device, establishing an implicit codebook comprising a set of code points for representing said at least one information element extracted from said digitized signal, each of said at least one information element constituting a vector, said set of code points implicitly represented by a hypercube radius vector;

via a computational device, inspecting the signs of each information element to determine in which orthant said each information element lies to determine the implicit code point to represent said each information element;

via a computational device, determining an index of the implicit code point associated with said each information element; and

via a computational device transmitting or storing said each resulting index as the compressed representation of said each information element.

36. A computer-implemented method for data compression comprising the steps of:

via a computational device, receiving at least one signal representing at least one image in any of digitized and non-digitized form;

responsive to receiving said signal in non-digitized form, via a computational device, digitizing said at least one received signal;

via a computational device, extracting at least one information element from said digitized signal;

via a computational device, establishing an implicit codebook comprising a set of code points for representing said at least one information element extracted from said digitized signal, each of said at least one information element constituting a vector, said set of code points implicitly represented by a hypercube radius vector;

via a computational device, inspecting the signs of each information element to determine in which orthant said each information element lies to determine the implicit code point to represent said each information element;

via a computational device, determining an index of the implicit code point associated with said each information element; and

via a computational device, transmitting or storing said each resulting index as the compressed representation of said each information element.