Progressive transmission of vector quantized data

• US 5,255,090 A
• Filed: 11/19/1991
• Issued: 10/19/1993
• Est. Priority Date: 11/19/1991
• Status: Expired due to Term

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: Scientific-Atlanta Incorporated (Cisco Systems, Inc.)
• Inventors: Israelsen, Paul D.
• Primary Examiner(s): Britton, Howard W.
• Application Number:
  US07/794,493
• Time in Patent Office: 700 Days
• Field of Search: 358/86, 358/133, 358/135, 358/136, 358/138, 358/105, 341/67
• US Class Current: 375/240.12
• CPC Class Codes: G06T 9/008 Vector quantisation
• CPC Class Codes: H04N 19/154 Measured or subjectively estimated visual quality after decoding, e.g. measurement of distortion use of rate-distortion criteria H04N19/147
• CPC Class Codes: H04N 19/39 involving multiple description coding [MDC], i.e. with separate layers being structured as independently decodable descriptions of input picture data
• CPC Class Codes: H04N 19/60 using transform coding
• CPC Class Codes: H04N 19/94 Vector quantisation

Abstract

A distortion adaptive vector quantization method is employed to select, for each input vector, several codevectors from a tree structured codebook. The first codevector defines a basis signal, and each subsequent codevector defines an enchancement signal. The basic signal is indicative of the full codebook address from which the first codevector was selected. Each enhancement signal is indicative of only so much of the additional address bits that are required, in combination with the basic signal, to identify the address of the codebook from which the subsequent codevectors were selected. The basic signal and each enhancement signal are transmitted to a decoder which may employ only the basic signal, or the basic signal and one or more of the enhancement signals to reproduce the input vector. The reproduction quality of the input vector increases with the number of enhancement signals chosen.

Claims

1. In a tree search, variable rate vector quantization data compression system of the type the organizes an image frame into blocks of input data, and converts each block of input data to a multi-dimensional input vector, and having a codebook with plural levels of codevectors, each codevector being representative of a possible input vector, the codevectors at each successive level representing possible input vectors with greater accuracy than codevectors at a preceding level, each codevector having an associated address, a method comprising the steps of:

a) for each input vector, automatically selecting a plurality of codevectors, each codevector being selected from a different level of the codebook;

b) for a first one of the selected codevectors, transmitting data indicative of the address associated therewith; and

,

c) for each of the other selected codevectors, transmitting data indicative of at least a portion of the address of each other selected codevectors that, in combination with previously transmitted address data, uniquely identifies each other selected codevector.

2. Method according to claim 1 wherein a compression code is assigned to each possible address and each possible portion of an address, each compression code having a length that is substantially inversely proportional to a predetermined probability that a particular codevector will be selected in step (a), and step (b) comprises transmitting the compression code for the address associated with the first one of the selected codevectors, and step (c) comprises transmitting the compression code for each said portion.

3. Method according to claim 1 wherein the address data to be transmitted has variable length and is temporarily stored then transmitted at a substantially fixed rate.

4. Method according to claim 1 further comprising the steps of:

a'"'"') providing another substantially identical tree search vector quantization codebook at a remote receiving location;

b'"'"') receiving at the receiving location the data transmitted for each input vector and obtaining therefrom the address for the first selected codevector and the address portions for other selected codevectors;

c'"'"') combining the address of the first selected codevector and at least one of the address portions to uniquely identify one of the other selected codevectors;

d'"'"') retrieving from the codebook at the receiving location the codevector uniquely identified in step (c'"'"'); and

e'"'"') employing the retrieved codevector to provide at the receiving location a substantial representation of the block of input data.

5. Method according to claim 1 wherein there are a plurality of first and second threshold values and step (a) comprises:

(i) selecting an initial level of the codebook and a first one of each of first and second threshold values;

(ii) initially selecting the codevector at the selected level that most closely resembles a representation of the input vector;

(iii) obtaining a measure of difference between the initially selected codevector in step (ii) and the selected one of the first threshold values;

(iv) defining the initially selected codevector in step (ii) as a finally selected codevector if either (i) the measure of difference is not greater than the selected one of the first threshold values or (ii) the selected level is not less than the selected one of the second threshold values, then transmitting data indicative of the address of the finally selected codevector and proceeding to step (vi), but otherwise proceeding to step (v);

(v) selecting subsequent levels in the codebook and repeating steps (ii)-(iv) for each subsequent level until address data has been transmitted;

(vi) selecting next subsequent codebook levels and subsequent ones of each of the first and second threshold values;

(vii) repeating steps (ii)-(vi) so as to transmit address data for finally selected codevectors from plural levels of the codebook.

6. Method according to claim 5 wherein a mean value of the input vector is determined and the mean value is removed from the input vector to obtain a residual vector, and the representation of the input vector is the residual vector.

7. Method according to claim 1 wherein the plurality of codevectors are selected according to a plurality of different threshold values.

8. Method according to claim 7 wherein there are a plurality of first and second threshold values and wherein each first threshold value is indicative of a tolerable distortion between a representation of the input vector and a codebook vector, and each second threshold value is indicative of a level of the codebook from which a codevector should be selected if the distortion is less than the first threshold value.

9. Method according to claim 7 wherein the addresses have a variable length that increases with each level of codevectors, further comprising the step of periodically adjusting the threshold values based upon an average length of data transmitted in steps (b) and (c).

10. Method according to claim 7 wherein there are a plurality of buffers for temporarily storing the address data to be transmitted, further comprising the step of storing the address data for the first one of the selected codevectors in a first one of the buffers and storing the said at least portions of address data for each of the other selected codevectors in other associated ones of the buffers, each buffer corresponding to one of the threshold values, each threshold value being individually adjusted based upon a measure of unused capacity of its corresponding buffer.

11. In a tree search, variable rate vector quantization data compression system of the type that organizes an image frame into blocks of input data, and converts each block of input data to a multi-dimensional input vector, and having a codebook with plural levels of codevectors, each codevector being representative of a possible input vector, the codevectors at each successive level representing possible input vectors with greater accuracy than codevectors at a preceding level, each codevector having an associated address, the addresses increasing in length with each subsequent level of the codebook, a method comprising the steps of:

a) defining first and second threshold values wherein the first threshold value is indicative of a tolerable distortion between a representation of the input vector and a codebook vector and the second threshold value is indicative of a level of the codebook from which a codevector should be selected if the distortion is greater than the first threshold value;

b) providing first and second buffers, there being a buffer corresponding to each threshold value;

c) for each input vector, automatically selecting, based upon the threshold values, a plurality of codevectors, each codevector being selected from a different level of the codebook, each buffer being assigned to one of the selected codevectors;

d) for a first codevector selected for each input vector, storing data indicative of the address associated therewith in the buffer assigned to the first selected codevector;

e) for each subsequent codevector selected for each input vector, storing, in its assigned buffer, data indicative of at least a portion of the address associated therewith that, in combination with address data stored in other buffers, uniquely identifies the subsequent codevectors; and

,

f) transmitting the address data from the buffers while periodically adjusting the threshold values based upon a measure of unused capacity of their corresponding buffer.

12. Method according to claim 11 further comprising the steps of:

a'"'"') providing another substantially identical tree search vector quantization codebook at a remote receiving location;

b'"'"') receiving at the receiving location the data transmitted for each input vector and obtaining therefrom the address for the first selected codevector and the address portions for the subsequently selected codevectors;

c'"'"') combining the address of the first selected codevector and at least one of the address portions to uniquely identify one of the subsequently selected codevectors;

d'"'"') retrieving from the codebook at the receiving location the codevector uniquely identified in step (c'"'"'); and

e'"'"') employing the retrieved codevector to provide at the receiving location a substantial representation of the block of input data.

13. Method according to claim 11 wherein there is a pair of first threshold values and a pair of second threshold values and wherein step (c) comprises:

(i) selecting an initial level of the codebook and one of the threshold values in each of the first and second pair of threshold values;

(ii) initially selecting the codevector at the selected level that most closely resembles the representation of the input vector;

(iii) obtaining a measure of difference between the initially selected codevector in step (ii) and the first threshold value in the selected pair of threshold values;

(iv) defining the initially selected codevector in step (ii) as a finally selected codevector if either (i) the measure of difference is not greater than the first threshold value in the selected pair of threshold values or (ii) the selected level is not less than the second threshold value in the selected pair of threshold values, then transmitting data indicative of the address of the finally selected codevector and proceeding to step (vi), but otherwise proceeding to step (v);

(v) selecting subsequent levels in the codebook and repeating steps (ii)-(iv) for each subsequent level until address data has been transmitted;

(vi) selecting next subsequent codebook levels and subsequent pairs of the threshold values;

(vii) repeating steps (ii)-(vi) so as to transmit address data for finally selected codevectors from plural levels of the codebook.

14. Method according to claim 13 wherein a compression code is assigned to each possible address and each possible portion of an address, each compression code having a length that is substantially inversely proportional to a predetermined probability that a particular codevector will be selected in step (c), and step (d) comprises storing the compression code for the address associated with the first one of the selected codevectors, and step (e) comprises storing the compression code for each said portion.

15. In a tree search, variable rate vector quantization data compression system of the type that converts a block of input data to a multi-dimensional input vector, and having a codebook with plural levels of codevectors, each codevector being representative of a possible input vector, the codevectors at each successive level representing possible input vectors with greater accuracy than codevectors at a preceding level, each codevector having an associated address, each address having a length that increases with each subsequent level of the codebook, a method comprising the steps of:

a) defining a plurality of distortion threshold values and an equal plurality of codebook level threshold values, there being a codebook level threshold value corresponding to each distortion threshold value;

b) selecting a first one of the distortion threshold values and the corresponding one of the codebook level threshold values, and defining the same as current distortion threshold and current codebook level threshold values, respectively;

c) receiving and processing an input vector, selecting an initial level of the codebook, and defining the initial level as a current level;

d) comparing the processed input vector to the codevectors at the current level of the codebook and selecting the codevector at the current level that most closely resembles the processed input vector;

e) obtaining a measure of distortion between the processed input vector and the selected codevector;

f) storing an indication of the address associated with the selected codevector only if the measure of distortion is less than the current distortion threshold value, but otherwise selecting a next level of the codebook;

g) redefining the current level as the next level selected in step (f);

h) repeating steps (d)-(g) until either the measure of distortion is less than the current distortion threshold value or a level of the codebook subsequent to the current level is not less than the current codebook level threshold value, then storing the indication;

i) selecting a next one of the distortion threshold values and the corresponding one of the codebook level threshold values and redefining the current distortion threshold and codebook level threshold values as the selected distortion threshold and codebook level threshold values, respectively;

j) selecting a next level of the codebook and redefining the current level as the next level selected in this step (j);

k) comparing the processed input vector to the codevectors at the current level of the codebook and selecting the codevector at the current level that most closely resembles the processed input vector;

l) obtaining a measure of distortion between the processed input vector and the selected codevector;

m) proceeding to step (n) if either the measure of distortion is less than the current distortion threshold value or a level of the codebook subsequent to the current level is not less than the current codebook level threshold, but otherwise repeating steps (j) through (l);

n) storing an indication of only a portion of the address of the codevector selected at the current level that, in combination with previously stored data, is required to uniquely identify the codevector selected at the current level; and

o) transmitting the stored data while periodically adjusting the distortion threshold and codebook level threshold values based upon the amount of stored data.

16. Method according to claim 15 wherein a compression code is assigned to each possible address and each possible portion of an address, each compression code having a length that is substantially inversely proportional to a predetermined probability that a particular codevector will be finally selected in either step (d) or (k), and wherein step (f) comprises storing the compression code for the address associated with the codevector finally selected in step (d), and step (n) comprises storing the compression code for the portion of the address associated with the codevector selected in step (k).

17. Method according to claim 15 wherein the stored address data has variable length and is transmitted at a substantially fixed data rate and adjustment of the threshold values ensures that there is no overflow or empty storage condition as a result of storing address data at a variable rate while transmitting the same at a substantially fixed rate.

18. Method according to claim 15 wherein, prior to performing step (d), a mean value of the input vector is determined and the mean value is removed from the input vector to obtain a residual vector, and the residual vector is employed to select the codevectors in steps (d) and (i), the residual vector being said processed version of the input vector.

19. Method according to claim 15 wherein the compression codes are Huffman codes.

20. Method according to claim 15 wherein the system receives input data to be compressed and the input data comprises image frames representing moving images to be displayed on a television set, each image frame comprising a matrix of pixels and each pixel within each image frame having at least an associated intensity value, and at least one of the input vectors are representative of the intensity values of a block of pixels in the image.

21. Method according to claim 20 further comprising the steps of:

a'"'"') providing another substantially identical tree search vector quantization codebook at a locale of the television set;

b'"'"') receiving the transmitted indications and obtaining therefrom the address of the first selected codevector and the portions of addresses of the other selected codevectors;

c'"'"') combining the address of the first selected codevectors and at least one of the portions to uniquely identify one of the other selected codevectors;

d'"'"') retrieving from the codebook at the locale of the television set the codevector uniquely identified in step (c'"'"'); and

e'"'"') employing the retrieved codevector to substantially recreate the image data for display on the television set.

22. Method comprising the steps of:

a) providing, at an earth station, a tree search variable rate vector quantization data compression system of the type that organizes an image frame into blocks of input data and that converts each block of input data to a multi-dimensional input vector, and having a codebook with plural levels of codevectors, each codevector being representative of a possible input vector, the codevectors at each successive level representing possible input vectors with greater accuracy than codevectors at a preceding level, each codevector having an associated address, the addresses increasing in length with each subsequent level of the codebook;

b) defining, in the compression system, a plurality of pairs of first and second threshold values wherein each first threshold value is indicative of a tolerable distortion between a representation of the input vector and a codebook vector and each second threshold value is indicative of a level of the codebook from which a codevector should be selected if the distortion is greater than the first threshold value;

c) providing, in the compression system, a plurality of buffers, there being a buffer corresponding to each pair of threshold values;

d) in the compression system;

(i) for each input vector, automatically selecting, based upon the threshold values, a plurality of codevectors, each codevector being selected from a different level of the codebook, each buffer being assigned to one of the selected codevectors;

(ii) for a first codevectors selected for each input vector, storing data indicative of the address associated therewith in the buffer assigned to the first selected codevector;

(iii) for each subsequent codevector selected for each input vector, storing, in its assigned buffer, data indicative of at least a portion of the address associated therewith that, in combination with address data stored in other buffers, uniquely identifies the subsequent codevector;

e) transmitting from the earth station, via satellite, to a receiving location, the address data from the buffers while periodically adjusting each pair of threshold values based upon a measure of unused capacity of its corresponding buffer;

f) at the receiving location;

i) providing another tree search vector quantization codebook that is substantially identical to that in the compression system;

(ii) receiving the data transmitted for each input vector and obtaining therefrom the address for the first selected codevector and the address portions for the subsequently selected codevectors;

(iii) combining the address of the first selected codevector and at least one of the address portions to uniquely identify one of the subsequently selected codevectors;

(iv) retrieving from the codebook at the receiving location the codevector uniquely identified in step (F)(iii); and

(v) employing the retrieved codevector to provide at the receiving location a substantial representation of the block of input data.

23. Method according to claim 22 wherein the receiving location comprises a DBS subscriber.

24. Method according to claim 22 wherein the receiving location comprises a cable head end and a plurality of subscribers.

25. Method according to claim 24 wherein steps (f)(i) through (f)(v) are performed at the cable head end and NTSC encoded picture data is transmitted from the cable head end to the subscribers.

26. Method according to claim 24 wherein the cable head end receives the transmitted address data and retransmits the same to the subscribers, and steps (f)(i) through (f)(v) are performed at a locale of each subscriber.

27. Data compression apparatus comprising:

first means for receiving a multi-dimensional input vector representative of a block of data to be compressed;

a tree search vector quantization codebook having plural levels of codevectors, each codevector being representative of a possible input vector and the codevectors at each successive level representing possible input vectors in greater detail than codevectors at a preceding level, each codevector having a unique address associated therewith;

a controller coupled to the first means and to the tree search codebook for maintaining a plurality of threshold values and for repeatedly and automatically comparing the input vector to the codevectors at successive levels of the codebook and for repeatedly and automatically selecting a plurality of codevectors for the input vector, each codevector being selected from a different level based on a comparison of one of the threshold values with a measure of difference between the codevector and the input vector; and

,

a plurality of transmitters each coupled to the controller, a first one of the transmitters for transmitting an indication of the address associated with a firs tone of the selected codevectors, the indication of the address associated with the first one of the selected codevectors being representative of a basic signal, each other transmitter for transmitting an indication of at least a portion of the address associated with a one of the remaining selected codevectors, each indication of at least a portion of the address associated with a one of the remaining selected codevectors being representative of an enhancement signal.

28. Apparatus according to claim 27 further comprising a plurality of lossless compression encoders, a first one of the encoders being coupled to the first one of the transmitters for assigning a compression code to the address of the first one of the selected codevectors, each other encoder being coupled to a respective one of the other transmitters for assigning a compression code to the portion of the address of the one of the remaining selected codevectors, the compression codes having a length that is substantially inversely proportional to a probability that the selected codevector would be selected, the transmitters each transmitting an indication of the respective compression code.

29. Apparatus according to claim 27 wherein the data compression apparatus is embodied as a portion of a pay television encoder.

30. Apparatus according to claim 27 wherein the indication of the address of the first one of the selected codevectors is representative of a basic compression signal and wherein the indication of each of the portions of the addresses of the remaining selected codevectors is representative of an enhancement signal.

31. Apparatus according to claim 27 further comprising calculator means for determining a mean value of the input vector and for removing the mean value from the input vector to obtain a residual vector, the controller employing the residual vector to select the codevectors.

32. Apparatus according to claim 31 wherein the first one of the transmitters further transmits an indication of the mean value.

33. Apparatus according to claim 27 further comprising a plurality of data buffers, each data buffer being coupled to a respective one of the transmitters for temporarily storing indications to be transmitted by the transmitter.

34. Apparatus according to claim 33 wherein at least one of the threshold values is periodically adjusted based upon a measure of unused capacity of at least one of the data buffers.

35. Apparatus according to claim 33 further comprising a plurality of feedback loops respectively coupling each of the data buffers to the controller for supplying the threshold values.

36. Apparatus according to claim 33 wherein the addresses of the selected codevectors have variable length and wherein the data buffers each comprise a first-in first-out (FIFO) buffer and further wherein the indications of the selected addresses and portions of addresses are transmitted form the FIFOs at a fixed data rate.

37. A decoder for use by subscribers of a pay television system employing, at a head end thereof, a tree search vector quantization data compressor for compressing digital data indicative of at least intensity values of pixels of images to be transmitted to the subscribers, and having a first codebook having plural levels of codevectors representative of possible input vectors and each successive level representing codevectors with a greater accuracy than a preceding level of codevectors, the head end transmitting a basic signal comprising data indicative of the address of a first selected codevector found to most closely resemble the input vector and successively transmitting a plurality of enhancement signals each indicative of a portion of an address of a subsequently selected codevector, each portion of an address being such that, in combination with previously transmitted data, the portion uniquely identifies the selected codevector, the decoder comprising:

a second codebook being substantially identical in content to the first codebook;

first means for receiving the basic signal and each successive enhancement signal, and for combining the basic signal and a selected number of the enhancement signals to obtain an address for addressing the second codebook and retrieving a codevector therefrom residing at the obtained address, the retrieved codevector being a function of the number of enhancement signals combined with the basic signal;

second means for converting the retrieved codevectors to NTSC image data and for providing the NTSC image data at an output thereof for substantially reproducing images input to the head end.