Apparatus and method for compressing information

• US 5,412,741 A
• Filed: 01/22/1993
• Issued: 05/02/1995
• Est. Priority Date: 01/22/1993
• Status: Expired due to Term

• Current Assignee: MediaTek, Inc.
• Original Assignee: David Sarnoff Research Center, Inc.
• Inventors: Shapiro, Jerome M.
• Primary Examiner(s): Razavi, Michael T.
• Assistant Examiner(s): FOX, DAVID
• Application Number:
  US08/007,694
• Time in Patent Office: 830 Days
• Field of Search: 382/56, 382/37, 348/398, 358/426, 358/261.3, 358/262.1, 358/432
• US Class Current: 382/232
• CPC Class Codes: H03M 7/30 Compression speech analysis-synthesis for redundancy reduction G10L19/00; for image communication H04N; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
• CPC Class Codes: H04N 19/64 characterised by ordering of coefficients or of bits for transmission
• CPC Class Codes: H04N 19/647 using significance based coding, e.g. Embedded Zerotrees of Wavelets [EZW] or Set Partitioning in Hierarchical Trees [SPIHT]

Abstract

An apparatus and method for signal, image, or video compression that achieves high compression efficiency in a computationally efficient manner and corresponding decoder apparatus and methods are disclosed. This technique uses zerotree coding of wavelet coefficients in a much more efficient manner than previous techniques. The key is the dynamic generation of the list of coefficient indices to be scanned, whereby the dynamically generated list only contains coefficient indices for which a symbol must be encoded. This is a dramatic improvement over the prior art in which a static list of coefficient indices is used and each coefficient must be individually checked to see whether a) a symbol must be encoded, or b) it is completely predictable. Additionally, using dynamic list generation, the greater the compression of the signal, the less time it takes to perform the compression. Thus, using dynamic list generation, the computational burden is proportional to the size of the output compressed bit stream instead of being proportional to the size of the input signal or image.

Claims

1. In a digital compression encoder responsive to an array of a relatively large number of multibit coefficients defining image data at a relatively high resolution that has been hierarchical subband decomposed by a wavelet transform at a given plurality of lower resolutions, whereby the lowest low (LL) subband comprises an array having the smallest number of multibit coefficients of all the wavelet-transform subband arrays, each LL subband multibit coefficient comprises an ancestor of all multibit coefficients of all higher subbands descended from that LL subband multibit coefficient, and each of said multibit coefficients of said LL subband and said multibit coefficients of said descendents thereof includes a most significant bit;

wherein said encoder includes coding means for deriving a data-compressed code comprising (a) a ZEROTREE symbol at first given coordinates of of said LL subband or at coordinates of descendents of said first given coordinates of said LL subband which are occupied by coefficients that are zerotree roots, (b) an ISOLATED ZERO symbol at second given coordinates of of said LL subband or at coordinates of descendents of said second given coordinates of said LL subband which are occupied by isolated-zero coefficients, and (c) at least one type of non-zero symbol at third given coordinates of said LL subband or at coordinates of descendents of said third given coordinates of said LL subband which are occupied by non-zero coefficients;

the improvement for efficiently deriving said data-compressed code wherein said encoder comprises;

means responsive to each multibit coefficient of said LL subband and said multibit coefficients of all of its descendents of said wavelet transform for deriving a zerotree map, wherein said zerotree map includes a zerotree-map multibit coefficient corresponding to each coefficient of said LL subband array, and each zerotree-map coefficient has a binary "1" value in a bit position thereof only if either its corresponding LL-subband-array coefficient or any of the descendents of its corresponding LL-subband-array coefficient has its most significant bit in that bit position; and

said coding means includes means that starts with a list of coordinates of said relatively small number of coefficients of said LL subband of said array and a relatively high initial threshold value for dynamically deriving in sequence said data-compressed code from the multibit coefficients of the wavelet-transform subband and the multibit coefficients of the zerotree map first at said relatively high initial threshold value and then at one or more successively lower threshold values by ignoring from consideration in said sequential dynamic derivation all coordinates of coefficients which are descendents of coordinates which have already been found in said sequential dynamic derivation to be occupied by coefficients that are zerotree roots.

2. The encoder defined in claim 1, wherein:

at least certain ordinally-arrranged bits of each multibit coefficient of said LL subband and each multibit coefficient of its descendents defines the magnitude of that coefficient, a first of said ordinally-arrranged bits corresponds to the magnitude of said most significant bit, and each other of of said ordinally-arrranged bits corresponds to one-half the magnitude of its immediately preceding bit;

said initial threshold value is equal to one-half the largest zerotree-map value as represented by the most significant bit of all the bits of all the zerotree-map multibit coefficients; and

each of said successively lower threshold values is equal to one-half of its immediately preceding threshold value.

3. The encoder defined in claim 1, wherein said coding means comprises:

first cyclically-operated logic means successively responsive to each of dominant lists of coordinates of multibit coefficients and each of said threshold values for deriving a subordinate list of coordinates of multibit coefficients during each cycle of operation thereof, wherein a dominant list is defined as a list that contains the coordinates of those multibit coefficients which (1) have not yet been found to be significant and a symbol is to be generated during that cycle of operation or (2) if a coefficient has been previously found to be significant, one of its descendants has not yet been found to be significant and a symbol is to be generated during that cycle of operation, and wherein a subordinate list is defined as a list that contains the coordinates of those multibit coefficients which have been previously found to be significant with respect to previous, larger, thresholds;

second cyclically-operated logic means successively responsive to each of subordinate lists of coordinates of multibit coefficients derived by said first cyclically-operated logic means and a current one of of said threshold values for deriving a dominant list of coordinates of multibit coefficients during each cycle of operation thereof, which derived dominant list is for use during a subsequent cycle of operation of said first cyclically-operated logic means;

arithmetic coding means for continuously generating a more accurate model of said data-compressed code in response to data supplied thereto which is derived during each cycle of operation of said first logic means and in response to data supplied thereto which is derived during each cycle of operation of said second logic means; and

said list of coordinates of said relatively small number of coefficients of said LL subband of said array comprises a starting dominant list of coordinates of multibit coefficients and said relatively high initial threshold value comprises the threshold value which are used during the first cycle of operation of said first logic means.

4. The encoder defined in claim 3, wherein each wavelet transform multibit coefficient includes a significance flag bit that is initially set to signify FALSE, and said first logic means comprises:

means for comparing, in turn, the absolute magnitude |x| of each individual coefficient addressed by said dominant list of coordinates during the current cycle of operation of said first logic means with said threshold value T during that current cycle of operation of said first logic means to (1), if |x|>

T, append the magnitude X of that coefficient to a subordinate list and then set the magnitude of that coefficient itself to zero and the significance flag bit to TRUE, (2), if |x|≦

T and the significance flag bit is FALSE, employing the corresponding zerotree-map value to define data indicative of whether that coefficient is an ISOLATED ZERO or a ZEROTREE ROOT, whereby that data may be used by said arithmetic coding means to (3) generate an ISOLATED ZERO symbol code in response to said first logic means determining a coefficient to be an ISOLATED ZERO and generate a ZEROTREE ROOT symbol code in response to said first logic means determining a coefficient to be a ZEROTREE ROOT, and (4) using the ISOLATED ZERO symbol code to update said model for an ISOLATED ZERO and using the ZEROTREE ROOT symbol code to update said model for a ZEROTREE ROOT;

index appending means operative, unless said first logic means has determined that coefficient to be a ZEROTREE ROOT, following the comparison by said first logic means during the first cycle operation thereof of each individual coefficient for appending the index of each of the three children of that coefficient of the starting dominant list to a separate one of three corresponding new dominant lists, and following the comparison by said first logic means during the each subsequent cycle operation thereof of each individual coefficient for appending the index of the children of that coefficient of the current dominant list to a new dominant list for use during the next cycle of operation of said first logic means; and

means effective after all the individual coefficients addressed by the starting dominant list of coordinates have been operated on during the first cycle of operation of said first logic means for concatenating the three corresponding new dominant lists into a single dominant list for use during the second cycle of operation of said first logic means, and effective after all the individual coefficients addressed by the current dominant list of coordinates have been operated on during each subsequent cycle of operation of said first logic means for cutting the threshold value to the next lower one of said successively lower threshold values and initiating a cycle of operation of said second logic means.

5. The encoder defined in claim 4, wherein each of said multibit coefficients of said array defining said image data includes a sign bit due to the fact that the image mean of the magnitudes of all the pixels defining the image has been subtracted from the magnitude of each individual one of these pixels, wherein said encoder and each of these multibit coefficients, and wherein:

said first logic means further comprises means effective in response to |x|>

T, to define data indicative of the state of said sign bit which may be used by said arithmetic coding means to generate either a POSITIVE symbol code or a NEGATIVE symbol code in accordance with the state of said sign bit and to update said model accordingly; and

said coding means further comprises means for encoding a bit stream header associated with said arithmetic coding means, which bit stream header includes an individual multibit coefficient which defines said image mean.

6. The encoder defined in claim 5, wherein:

said bit stream header also includes individual multibit coefficients which respectively define the number of wavelet scales, the initial threshold value and the dimensions of said image.

7. The encoder defined in claim 3, wherein said second logic means comprises:

means for establishing a bottom list of coordinates of multibit coefficients which is initially cleared; and

initial means for determining the magnitude W by which a first coefficient on the subordinate list exceeds the current threshold value.

8. The encoder defined in claim 3, wherein said second logic means further comprises:

first means for determining if the magnitude y by which each individual coefficient, in turn, addressed by said subordinate list of coordinates exceeds the current threshold value is equal to W, and, if not, (1) inserting the entire bottom list at the coordinate position of the subordinate list immediately before the coordinate position of that coefficient, (2) clearing the bottom list, and (3) setting W equal to y,

second means for comparing, in turn, the magnitude X of each individual coefficient addressed by said subordinate list of coordinates during the current cycle of operation of said second logic means with said threshold value T during that current cycle of operation of said second logic means to (1), if X>

T, to define data that may be used by said arithmetic coding means to generate a "1" symbol code and to update said model for that "1" symbol code, and (2), if X≦

T, to define data that may be used by said arithmetic coding means to generate a "0" symbol code, to update said model for that "0" symbol code and move that individual coefficient from the subordinate list to the bottom list; and

third means effective after all the individual coefficients addressed by the current subordinate list of coordinates have been operated on during each cycle of operation of said second logic means for appending the entire bottom list to the end of the subordinate list and then initiating the next cycle of operation of said first logic means.

9. The encoder defined in claim 3, wherein said coding means has a given bit budget and further comprises:

means associated with each of said first and second logic means for updating said bit budget on each cycle of operation thereof and for terminating operation thereof if the bit budget is exceeded.

10. In a digital compression decoder for reconstructing image data from a bit stream comprising data bits of arithmetic-coded symbols comprising a ZEROTREE symbol, an ISOLATED ZERO symbol, and at least one type of non-zero symbol that define coefficient models for an array of multibit coefficients of wavelet-transformed hierarchical decomposed subbands of said image data;

wherein said decoder has an a priori knowledge of the protocol for the order in which the wavelet-transformed multibit coefficients of said array were scanned on compression so that a list of the coordinates of the multibit coefficients of the lowest low (LL) wavelet-transformed subband of said array are predetermined;

wherein (a) a ZEROTREE symbol occurs at first given coordinates of of said LL subband or at coordinates of descendents of said first given coordinates of said LL subband which correspond to coefficients that are zerotree roots, (b) an ISOLATED ZERO symbol occurs at second given coordinates of of said LL subband or at coordinates of descendents of said second given coordinates of said LL subband which correspond to isolated-zero coefficients, and (c) at least one type of non-zero symbol at third given coordinates of said LL subband or at coordinates of descendents of said third given coordinates of said LL subband which correspond to non-zero coefficients;

the improvement for efficiently reconstructing said image data wherein said decoder comprises;

data decoding means responsive to said data bits applied thereto for deriving the data values of said wavelet-transformed multibit coefficients from said arithmetic-coded symbols that starts with a list of coordinates of said relatively small number of multibit coefficients of said LL subband of said array and a relatively high initial threshold value for dynamically deriving in sequence the values of said wavelet-transformed multibit coefficients first at said relatively high initial threshold value and then at one or more successively lower threshold values by ignoring from consideration in said sequential dynamic derivation all coordinates of coefficients which are descendents of coordinates which have already been found in said sequential dynamic derivation to be occupied by a ZEROTREE symbol.

11. The decoder defined in claim 10, wherein:

each of said successively lower threshold values is equal to one-half of its immediately preceding threshold value.

12. The decoder defined in claim 10, wherein said data decoding means comprises:

setting means for initially setting the magnitude of each of the values of said wavelet-transformed multibit coefficients of said array to zero;

arithmetic decoding means for decoding said arithmetic-coded symbols;

first cyclically-operated logic means successively responsive to each of dominant lists of coordinates of multibit coefficients, certain decoded symbols, and each of said threshold values for deriving a subordinate list of coordinates of multibit coefficients and updating models of the values of said wavelet-transformed multibit coefficients of said array during each cycle of operation thereof, wherein a dominant list is defined as a list that contains the coordinates occupied by those arithmetic-coded symbols that define multibit coefficients which (1) have not yet been found to be significant and a symbol is to be decoded during that cycle of operation or (2) if a coefficient defined by a decoded arithmetic-coded symbol has been previously found to be significant, one of its descendants has not yet been found to be significant and a symbol is to be decoded during that cycle of operation, and wherein a subordinate list is defined as a list that contains the coordinates of those multibit coefficients which have been previously found to be significant with respect to previous, larger, thresholds;

second cyclically-operated logic means successively responsive to each of subordinate lists of coordinates of multibit coefficients derived by said first cyclically-operated logic means, other decoded symbols, and a current one of of said threshold values for deriving a dominant list of coordinates of multibit coefficients and updating models of the values of said wavelet-transformed multibit coefficients of said array during each cycle of operation thereof during each cycle of operation thereof, which derived dominant list is for use during a subsequent cycle of operation of said first cyclically-operated logic means; and

said list of coordinates of said relatively small number of coefficients of said LL subband of said array comprises a starting dominant list of coordinates of multibit coefficients and said relatively high initial threshold value comprises the threshold value which are used during the first cycle of operation of said first logic means.

13. The decoder defined in claim 12, wherein each wavelet transform multibit coefficient of said array includes a significance flag bit that is initially set to signify FALSE by said setting means, and wherein said first logic means comprises:

means responsive to the setting of said significance flag bit being FALSE for determining, in turn, for the decoded symbol corresponding to each individual coefficient addressed by said dominant list of coordinates during the current cycle of operation of said first logic means is an ISOLATED ZERO symbol, a ZEROTREE ROOT symbol or is neither;

if neither, put a value equal to 3/2 of the threshold value T during that cycle in the corresponding coefficient of an output map, append coordinates of that coefficient to said subordinate list and set significance flag to TRUE;

if an ISOLATED ZERO symbol, update model for ISOLATED ZERO symbol; and

if a ZEROTREE ROOT symbol, update model for ZEROTREE ROOT symbol;

index appending means operative only if said significance flag is TRUE or the decoded symbol has been determined to be other than a ZEROTREE ROOT symbol for appending the index of each of the three children of that coefficient of the starting dominant list to a separate one of three corresponding new dominant lists, and following the comparison by said first logic means during the each subsequent cycle operation thereof of each individual coefficient for appending the index of the children of that coefficient of the current dominant list to a new dominant list for use during the next cycle of operation of said first logic means; and

means effective after all the individual coefficients addressed by the starting dominant list of coordinates have been operated on during the first cycle of operation of said first logic means for concatenating the three corresponding new dominant lists into a single dominant list for use during the second cycle of operation of said first logic means, and effective after all the individual coefficients addressed by the current dominant list of coordinates have been operated on during each subsequent cycle of operation of said first logic means for cutting the threshold value to the next lower one of said successively lower threshold values and initiating a cycle of operation of said second logic means.

14. The decoder defined in claim 13, wherein each wavelet transform multibit coefficient of said array includes a sign bit, and wherein:

said means for determining determines, in order, whether the decoded symbol corresponding to each individual coefficient addressed by said dominant list of coordinates during the current cycle of operation of said first logic means is a POSITIVE symbol, NEGATIVE symbol, ISOLATED ZERO symbol, or a ZEROTREE ROOT symbol, and the value equal to 3T/2 put in the corresponding coefficient of-said output map during that cycle is +3T/2 in response to a POSITIVE symbol and -3T/2 in response to a NEGATIVE symbol.

15. The decoder defined in claim 14, wherein said second logic means comprises:

means for establishing a bottom list of coordinates of multibit coefficients which is initially cleared; and

initial means for setting the initial magnitude S=y, where y is the magnitude value of the wavelet coefficient at the coordinates of the first element on the subordinate list.

16. The decoder defined in claim 15, wherein said second logic means further comprises:

first means for determining if the magnitude y of each individual wavelet coefficient, in turn, addressed by said subordinate list of coordinates is equal to S, and, if not, (1) inserting the entire bottom list at the coordinate position of the subordinate list immediately before the coordinate position of that coefficient, (2) clearing the bottom list, and (3) setting S equal to y

second means responsive to a decoded symbol from the arithmetic decoder of that individual wavelet coefficient for first determining whether the decoded symbol is that decoded symbol is an "end-of bit-stream" symbol, a "1" symbol, or said decoded symbol is neither, is neither, and in response to (1) an "end-of bit-stream" symbol, terminating operation of said decoder, (2) a "1" symbol,, updating model for a "1" symbol and adding T/2 to the magnitude of that coefficient in said output map, and (3) said decoded symbol being neither, updating model for a "0" symbol, moving that coefficient from the subordinate list to the bottom list and subtracting T/2 from the magnitude of that coefficient in said output map; and

third means effective after all the individual coefficients addressed by the current subordinate list of coordinates have been operated on during each cycle of operation of said second logic means for appending the entire bottom list to the end of the subordinate list and then initiating the next cycle of operation of said first logic means.

17. The decoder defined in claim 10, wherein said bit stream comprises both said data bits and header bits defining image dimensions, said initial threshold value and the number of wavelet scales, and said decoding means further comprises:

means for separating the header bits from said data bits and decoding said header bits to derive separate header multibit coefficients including ones that respectively define said image dimensions, said initial threshold value and the number of wavelet scales; and

means for applying said separate multibit coefficients as inputs to said data decoding means.

18. The decoder defined in claim 17, further comprising:

means responsive to said derived the data values of said wavelet-transformed multibit coefficients and said multibit coefficients that define said image dimensions and the number of wavelet scales for computing the coefficients of an inverse wavelet transform of said wavelet-transformed multibit coefficients.

19. The decoder defined in claim 18, wherein said data values of said wavelet-transformed multibit coefficients image data includes a sign bit due to the fact that the image mean of the magnitudes of all the pixels defining the image has been subtracted from the magnitude of each individual one of these pixels and said separate header multibit coefficients include one that defines said image mean, and said decoder further comprises:

means for adding said coefficient that defines said image mean to each of the coefficients of said inverse wavelet transform thereby to reconstruct said image data.

20. A method for deriving a zerotree map for use in digitally compression encoding an array of a relatively large number of multibit coefficients defining image data at a relatively high resolution that has been hierarchical subband decomposed by a wavelet transform at a given plurality of lower resolutions, whereby the lowest low (LL) subband comprises an array having the smallest number of multibit coefficients of all the wavelet-transform subband arrays, each LL subband multibit coefficient comprises an ancestor of all multibit coefficients of all higher subbands descended from that LL subband multibit coefficient, and each of said multibit coefficients of said LL subband and said multibit coefficients of said descendents thereof includes a most significant bit;

said method comprising the step of;

generating a separate zerotree-map multibit coefficient corresponding to each coefficient of said LL subband array such that each zerotree-map coefficient has a binary "1" value in a bit position thereof only if either its corresponding LL-subband-array coefficient or any of the descendents of its corresponding LL-subband-array coefficient has its most significant bit in that bit position.

21. A system comprising (1) an encoder for efficiently generating a digital compressed code in response to an array of a relatively large number of multibit coefficients defining image data at a relatively high resolution that has been hierarchical subband decomposed by a wavelet transform at a given plurality of lower resolutions, whereby the lowest low (LL) subband comprises an array having the smallest number of multibit coefficients of all the wavelet-transform subband arrays, each LL subband multibit coefficient comprises an ancestor of all multibit coefficients of all higher subbands descended from that LL subband multibit coefficient, and each of said multibit coefficients of said LL subband and said multibit coefficients of said descendents thereof includes a most significant bit;

wherein said encoder includes coding means for deriving said data-compressed code comprising (a) a ZEROTREE symbol at first given coordinates of of said LL subband or at coordinates of descendents of said first given coordinates of said LL subband which are occupied by coefficients that are zerotree roots, (b) an ISOLATED ZERO symbol at second given coordinates of of said LL subband or at coordinates or descendents of said second given coordinates of said LL subband which are occupied by isolated-zero coefficients, and (c) at least one type of non-zero symbol at third given coordinates of of said LL subband or at coordinates of descendents of said third given coordinates of said LL subband which are occupied by non-zero coefficients; and

(2) a decoder for efficiently decoding the digital compressed code generated by said encoder;

wherein;

said encoder includes (3) means responsive to each multibit coefficient of said LL subband and said multibit coefficients of all of its descendents of said wavelet transform for deriving a zerotree map, wherein said zerotree map includes a zerotree-map multibit coefficient corresponding to each coefficient of said LL subband array, and each zerotree map coefficient has a binary "1" value in a bit position thereof only if either its corresponding LL-subband-array coefficient itself or any of the descendents of its corresponding LL-subband-array coefficient has its most significant bit in that bit position, and (4) said coding means includes means that starts with a list of coordinates of said relatively small number of coefficients of said LL subband of said array and a relatively high initial threshold value for dynamically deriving in sequence said data-compressed code from the multibit coefficients of the wavelet-transform subband and the multibit coefficients of the zerotree map first at said relatively high initial threshold value and then at one or more successively lower threshold values by ignoring from consideration in said sequential dynamic derivation all coordinates of coefficients which are descendents of coordinates which have already been found in said sequential dynamic derivation to be occupied by coefficients that are zerotree roots; and

said decoder includes (5) data decoding means responsive to said data bits of said arithmetic-coded symbols of said data-compressed bit stream applied thereto for deriving the data values of said wavelet-transformed multibit coefficients from said arithmetic-coded symbols that starts with a list of a priori known coordinates of said relatively small number of multibit coefficients of said LL subband of said array and a relatively high initial threshold value for dynamically deriving in sequence the values of said wavelet-transformed multibit coefficients first at said relatively high initial threshold value and then at one or more successively lower threshold values by ignoring from consideration in said sequential dynamic derivation all coordinates of coefficients which are descendents of coordinates which have already been found in said sequential dynamic derivation to be occupied by a ZEROTREE symbol.