5,850,482
`
`13
`the data quantizer can quantize the transfomied data non-
`uniformly Without departing from the spirit and scope of the
`present invention.
`Once the transformed data has been quantized, the quan-
`tized significant coeflicienm can be encoded by the error
`resilient method and apparatus for encoding data as shown
`in block 35. The insignificant coellicients can be run length
`coded to generate run length values which are thereafter
`entropy encoded by the error resilient method and apparatus
`for encoding data of the present
`invention or by another
`method, such as Hnftiman coding, as shown in blocks 36 and
`37 of FIG. 2.
`Due to the transformation and quanlilation prowsses as
`described above, the quantized data can be well approxi-
`mated by a Laplacian or, more preferably, a generalized
`(iaussian distribution which is sharply peaked at the origin.
`A stylized generalized Gaussian distribution is shown in
`FIG. 5Afor purposes of illustration since actual distributions
`are typically noisy and are more sharply peaked at the origin.
`The data encoder 16 01' the present invention can encode
`the quantized data according to a predetermined codebock.
`In particular, both the quantized significant coeificients and
`their relative positions within the array of pixels are encoded
`tn thereby increase the compression performance by elimi-
`nating explicit coding of each insignificant coefiicient.
`The positions of the significant coellicients within the
`overall array oi‘ pixels can be encoded by at variety oi‘
`methods, including coefiicient maps, tree structures or run
`length coding. In one preferred embodiment, the numerous
`insignificant cocllicicnls are encoded by run lengths as
`shown in block 36. In run length coding, the number of
`insignificant coefiieients which occur consecutively between
`two significant coellicients is specified, thereby ellectively
`specifying the position of the second signiticant coellicient
`relative to the position of the first significant coefiicient.
`According to the present invention, an entropy encoder 16
`and, more preferably, code word generating means 26 gen-
`erates a plurality of code words which are representative of
`the quantized significant cocfficicnts. Accordingly, the plu-
`rality ol code words elleetively represent
`the quantized
`image data. Each code word includes at least a first portion
`(hereinafter termed a “prefix field”) and an associated sec-
`ond portion (hereinafter termed a "sul’fix field").
`Accordingly, the code word generating means preferably
`includes a prefix generating means 27 for generating the
`prefix field of each code word and a srtliix generating means
`28 for generating the associated sufix field of each code
`word. Since each code word is formed of two fields, namely,
`the prefix field and the suffix field, this method of coding will
`be termed “split field coding”.
`According to split field coding, the prefix field includes
`inforrriation representative of the associated sullix field,
`while the sullix field associated with the prefix field includes
`information representative of the respective significant
`coetticient, typically encoded according to a predetermined
`codebook. More specifically,
`the prefix field preferably
`includes information representative of the predetermined
`number of characters which form the associated sufiix field.
`The prefix field may also include infonnation representative
`of another predetermined characteristic of the associated
`sufiix field, such as the contiguous or consecutive range of
`coefficient values which the associated suffix field of the
`code Word may represent.
`Typically, each sufiix field is associated with a corre-
`sponding prefix field and is formed by a predetermined
`number of characters, such as a predetermined number of
`hits. Thus, the prefix field preferably includes information
`
`‘Jt
`
`ltl
`
`2t}
`
`I~.a-A
`
`40
`
`45
`
`{it}
`
`14
`representative of the prerieternztinetl number of characters
`which form the associated sufiix field.
`It has been widely observed that a decorrelating properties
`of the wavelet transform resuit in a distribution of coefficient
`values which is typically sharply peaked at zero and which
`decays more or less monotonically away from the peak at
`zero as depicted in FIG. 5A. This type of distribution
`dictates that similar coeilicient values typically have similar
`probabilities or frequencies of occurrence. As noted
`previously.
`the quantizer 14 maps all coeflicients whose
`values fall within a particular interval to a particular discrete
`symbol. As known to those skilled in the art, the statistics of
`the quantized coelliciunts can be characterized using a
`“histogram" which is a discrete distribution consisting of
`individual bins representing the [requency or probability of
`occurrence of the quantized coefficient values. Each bin is
`associated with a particular quantization interval and has a
`frequency defined by a count of the number of occurrences
`of quantized coellicients whose values fall within the asso-
`ciated quanti‘/.atit'm interval. A stylized representation of a
`quantized coefficient histogram is depicted in FIG. 5B.
`Because the counts within the histogram are dependent upon
`the original coefiicient distribution, the histogram also dem-
`onstrates that bins which represent coefficicnts having simi-
`lar values will typically have similar counts or probabilities
`oi" occurrence.
`As known to those skilled in the art, entropy coding
`achieves a reduction in the number of bits required to
`represent a data set by assigning shorter code words to
`symbols which occur frequently and longer code words to
`symbols which occur less frequently. Consequently, sym-
`bols with similar probabilities of occurrence should be
`presented by code words with similar lengths, and because
`quantized coeflicients with similar values typically share
`similar probabilities of occurrence as described above, they
`should also be represented by code words having similar
`code word lengths.
`According to one advantageous embodiment of the
`present invention, the prefix field includes information rep-
`resentative ol" the number of hits K which form the associ-
`ated sullix field. of the code word. Furthermore, the prefix
`field preferably includes information representative of a
`specific set of 2"‘ consecutive histogram bins of the quan-
`tized coefiicient histogram which are, in turn, associated
`with a corresponding set of 2K consecutive quantized coef-
`ficicnt values. The 2K possible values for the associated K hit
`sutftx field will each be associated by one-to—one correspon-
`dence with the 2K consecutive bins which are designated by
`the associated prefix field. In aggregate, the prefix and suffix
`field of each code word shall together include information
`representative of a specific symbol, associated with a spe-
`cific hin of the quantized coefiicient hi:~itog'ram. Accordingly,
`the quantized coellicicnt histogram shall be partitioned into
`sets of consecutively occurring bins which shall be referred
`to as "superbins". Each superbin is, in turn, associated with
`a unique value of the prefix field.
`FIG. SR shows the partitioning of an exemplary quantized
`cocficient histogram into sets termed “supcrbins", i.e., the
`dotted lines indicate the bounds of the superbins. It will be
`apparent
`to those skilled in the art, however,
`that
`the
`histogram can be partitioned in other manners without
`departing from the spirit and scope of the present invention.
`Accordingly, the partitioning of the exemplary histogram of
`FIG. 5B into superbins is depicted and discussed for pur-
`poses of illustration and not limitation.
`The central bin (shaded) of the exemplary histogram
`corresponds to the insignificant coefficients which are not
`
`Page 22 of 437
`
`Page 22 of 437
`
`

`
`5,850,482
`
`15
`directly encoded as quantized coellicients and are therefore
`not included in any superbin. On either side of the central
`bin, the histogram illustrates a superbin consisting of two
`bins. These superbins would preferably each be associated
`with a sullix field having 51 length of 1 bit which is sullicicnt
`to distinguish the 2 individual bins of the superbin. The
`histogram in FIG. 513 also shows superbins with widths of
`4 and 8 bins which would preferably be associated with
`sufix fields having lengths of2 and 3 bits respectively, so as
`to distinguish the individual bins within each of the
`superblns.
`The prefix generating means 27 can generate the prefix
`fields in a variety of manners, such as unary coding. As
`known to those skilled in the art, one type of unary coding
`represents an unsigned integer 3 using 1 bits consisting of
`(J—l) mros followed by a single one bit which terminates the
`code. In instances in which the largest value to be repre-
`sented by a unary code, such as JMAX, is known, the integer
`IMAX may be represented by (JMAX—l) zeros with no
`terminating one required to dillizrentiatc JMAX l'Ion'i larger
`integers. In instances in which the values to be represented
`by the code words are signed quantities, the prefix field can
`also inciudc an extra or leading bit which designates the sign
`of the quantized cocfiicicnt represented by the code word.
`According to the present invention, each possible value
`for the prefix lireld can be associated with a respective sttilix
`field having a predetermined number of characters. The
`sutfix generating means 28 can generate the sufiix fields in
`a variety of manners, but, in one preferred embodiment, the
`sullix fields are binary integers oi" length K. In this pre ferrcd
`embodiment, each possible combination of the K bit sufix
`field is associated with one of the 2K bins which form the
`superbin designated by the associated prefix field. For
`example, the two and three bit
`integers shown over the
`superbins in the histogram of FIG. 5B represent the prefix
`field codes for the respective superbins. The most significant
`bit is a sign bit, white the remaining bits are a unary code
`which specifies the respective supcrbin. As described above,
`the associated suffix fields for the supcrbins of width 2, 4,
`and S bins consist of 1, 2, or 3 bits, respectively. Note that
`the code words associated with each superbin all share the
`same code word length. As previously noted, the bins for
`similar valued eoefiicients typically have similar counts or
`probabiiities of occurrence, so that the proposed codes can
`result in ellicient codes by assigning common code word
`lengths to coeflicicnts with similar probabilities.
`In summary, the prefix field preferably includes informa-
`tion representative of El predetermined characteristic of the
`associated sufiix field, such as the predetermined number of
`characters which form the associated sulilx field of the code
`word. Furthermore, the prefix field may also include infor-
`rnution representative of another pl'Cdctt3t‘I'f]i1'tcd characteris-
`tic of the associated sullix field, such as the contiguous or
`consecutive range of coeffieient values or bins which the
`associated sufiix field of the code word may represent,
`wherein the contiguous or consecutive range corresponds to
`a supcrbin. In addition, the suffix fields include information
`representative of respective portions of the original data.
`such as specifically designating an individual bin within a
`superbin.
`Consequently, if the prefix held of a code word is decoded
`correctly, that is, without the occurrence of bit error, the
`method and apparatus of the present invention can correctly
`determine the length of the asstociatcd sulfix field and can
`also correctly determine the range of coefficient values to be
`represented by the associated sullix field. As a result, the
`associated sullix field will exhibit resilience to errors in two
`
`‘Jl
`
`ll)
`
`2[}
`
`to-A
`
`40
`
`45
`
`{st}
`
`16
`respects. First, one or more bit errors within the associated
`suftlx field shall not result in a loss of code word synchro-
`nization but, instead, the effects of those bit errors shall be
`isolated to that single code word. Second, the misdecoded
`coelficient value resulting from one or more bit errors within
`the associated suffix field shall be constrained to that con-
`tiguous range ol' coellicient values represented by the prefix
`field which corresponds to the range of the associated
`superbin.
`the method and apparatus of the present
`Accordingly,
`invention is suitable for use with unequal error protection
`means as known to those skilled in the art and as described,
`for example,
`in R. G. Gallager, “Infrinnatic)u Theory and
`Reliable Communication”, Wiley and Sons (1968).
`Specifically, the prefix fields of the encoded data are pref-
`erably channei encoded with an appropriately high level of
`error protection in order to provide a high probability that
`the prefix fields will be decoded correctly. Because the
`associated sullix fields are error resilient, however, the sullix
`fields may be channel encoded with a lower level of error
`protection or may not be channel encoded, thereby provid-
`ing no error protection. This unique error protection shall
`result in a reduction of storage requirements or a reduction
`in transmission bandwidth because the use of a lower level
`of error protection or no error protection will reduce the
`introduction of redundant data into the data link or storage
`medium while still providing error resiliency.
`As known to those skilled in the art, a variety of proposed
`codes can be separated into a prefix and suftix fields as
`described above. See, for example, E. R. Fiala and I). H.
`Greene, “Data Compression with Finite Windows," Com-
`mmricrrfrbrtv rJftireACM, Vol. 32, No. 4, pp. 490-505 (1989).
`However,
`the proposed codes have not previously been
`separated in order to provide error resiliency as provided by
`the method and apparatus of the present invention.
`It will be apparent to those skilled in the art that split field
`coding can be applied to data sets which are not character-
`ized by a wcll—behaverl distribution. This application can be
`accomplished by initially sorting the data set to produce a
`re-ordered monotonic distribution. This approach will result
`in error resilience in the sense that a bit error in the sullix
`field will not result in a loss of code word synchronization.
`However, the resulting error in the decoded value will not be
`constrained to a particular range since the sorting of the data
`set will destroy the contiguity of the supcrbins associated
`with specific prefix field values.
`The relative positions of the significant coeflicients can
`also be encoded in a variety of manners, such as ran length
`coding as described above. The resulting run length values
`can, in turn, also be entropy encoded using an approach such
`as Hufiman coding or the split field coding method described
`above. Alternatively, the positions of the significant coctli—
`cients can be encoded by other methods known to those
`skilled in the art, such as tree structures or coefficient maps,
`without departing from the spirit and scope of the present
`invention.
`Once the plurality of code words representative of the
`quantized cocflicients and the encoded run lengths have
`been generated, the run length code words and the prefix
`fields of the quantized coellicient code words are preferably
`error protected at an appropriately high level of error
`protection, as shown in block 38. The run length code words
`are preferably afforded protection because a misdecodcd run
`length value can potentially introduce catastrophic distortion
`into the reconstructed image. However, the suffix fields of
`the qpantized ooellicient code words are" preferably error
`protected at at reiatively lower level of error protection, it" at
`
`Page 23 of 437
`
`Page 23 of 437
`
`

`
`5,850,482
`
`17
`all. As shown schernztticaliy in FIG. 1, the data encoder 16
`can therefore include unequal crror protection means 2.9 for
`providing appropriate levels of error protection to the
`encoded data as described above.
`Regardless of the error protection means, error protection
`adds redundancy to the encoded data and increases the
`storage and transmission requirements. Accordingly, by pro-
`viding a
`reduced level of error protection or no error
`protection to the sulfur fields of the quantized coeflicient
`code words, the storage and transmission requirements can
`be reduced by the method and apparatus of the present
`invention While limiting the effects of bit errors incident
`upon the suflix fields of the quantized coetficicnt code
`words.
`Following the data compression process described above,
`the encoded data can be efficiently stored. For example, the
`run length code words and the prefix fields of the quantized
`coefficient code words can be stored in a first data block 66
`defined by a storage medium 18, such as a magnetic disk
`storage which is error protected as shown in FIG. 6. In
`addition, the respective sullix tields of the quantized coef-
`ficient code words can be stored in a second data block 68
`defined by a storage medium which includes a reduced level
`of error protection or no error protection. Thus, the sufix
`fields can he more efiiciently stored within the second data
`block.
`Likewise, the contpressed and encorlerl data can be elli-
`ciently transmitted, such as via lirst and second data links. In
`particular the error resilient method and apparatus of the
`present invention can include a transmitter 20 which trans-
`mits the respective run length code words and the prefix
`fields ofthe quantized coeficicnt code words via a first data
`link 22 which is error protected, and which transmits the
`respective sullix fields of the quantized coeflicient code
`words via a second data link 24 which is not error protected
`or is error protected to a lesser degree than the first data link.
`Thus, the suflix fields can be more efficiently transmitted
`(with reduced or no redundancy) using the second data link.
`Upon reception of the compressed data, the prefix fields
`of the quantized cneflicicnt code words can be decoded (as
`shown in FIG. 7) and the lengths of the suilix lields can be
`determined based on the decoded prefix lielrls. If one or
`more bit errors are incident upon the suflix field of a
`quantized coelficient code word, the code word synchroni-
`zation is not lost because the length of the sulfix field is
`known. As a result,
`the resulting error
`in the decoded
`coefficient value will be constrained to the range of coe.lli—
`cient values for the superbin corresponding to the associated
`prefix field. Atzcordingly,
`the effects of the error on the
`reconstructed image will be limited and will not be cata-
`strophic. Following the transmission of the encoded data and
`the possible detection and correction of any storage and
`transmission errors by means of channel decoding known to
`those skilled in the art, the cornprcsscd data, including both
`the quantized values for the significant coefficients and the
`relative positions of the significant coelftcients, is decoded,
`dequantized, and inverse transformed, as known to those
`skilled in the art, so as to provide a reconstructed image
`based upon the original image as shown in FIG. 7.
`The error resilient method and apparatus for compressing
`data, including the data transformer 12, the data quantiiaer
`14, the data encoder 16 and the unequal error protection
`means 29, are preferably implemented by a combination of
`hardware and software. For example, the method and appa-
`ratus for compressing data can be implemented by a com-
`puter having one or more controllers which operate under
`the control of software to provide the data transforrnation,
`quantization and encoding processes described above.
`
`18
`In the drawings and the specification, there has been set
`forth a preferred eanhodimcnt of the invention and, aithough
`specific terms are employed, the terms are used in a generic
`and descriptive sense only and not for purpose of limitation,
`the scope of the invention being set forth in the foilowing
`claims.
`That which is claimed is:
`1. An error resilient method of encoding data comprising
`the steps of:
`generating a plurality of code words representative of
`respective portions ofthe data, wherein each code word
`comprises a first portion and an associated second
`portion, and wherein said code word generating step
`comprist-3s the steps of:
`generating the llrst portion of each code word, wherein
`said first portion generating step comprises the step
`of including information within the first portion that
`is representative of a predetermined cha racteristic of
`the associated second portion; and
`generating the second portion of each code word,
`wherein said second portion generating step com-
`prises the step of including information within the
`second portion that is representative of the respective
`portion of the data; and
`least one of the first
`providing error protection to at
`portions of the plurality of code words while maintain-
`ing any error protection provided to the respective
`second portion associated with the at
`least one first
`portion at
`a
`lower level
`than the error protection
`providetl to the respective iirst portion.
`2. An error resilient method of encoding data according to
`claim 1 wherein said step of generating a plurality of code
`words comprises the step of entropy coding the data to
`thereby reduce the size of the resulting code words.
`3. An error resilient method of encoding data according to
`claim 1 wherein said step of generating the second portion
`ol'each code word comprises the step of generating second
`portions having predetermined numbers of characters, and
`wherein said step of generating the first portion ofeach code
`word comprises the step of generating first portions which
`include information representative of the predetermined
`number oi" characters which comprise the associated Second
`portion.
`4. An error resilient method of encoding data according to
`claim 3 further comprising the step of determining the
`probability with which respective ones of the plurality of
`code words are generated, wherein said step of generating
`second portions having predetermined numbers of charac-
`ters comprises the step of generating a plurality of second
`portions having the same predetermined number of
`characters, and wherein the plurality of second portions
`which have the same predetermined number oi‘ characters
`comprise portions of respective code words which have
`corresponding probabilities of generation within a predeter-
`mined range of probabilities.
`5. An error resilient method of encoding data according to
`claim 1 wherein said step of providing error protection
`comprises the steps of:
`storing the at least one first portion ofthe plurality of code
`words in a first data block of a storage medium,
`wherein the lirst data block is error protected; and
`storing the respective second portion associated with the
`at least one first portion in ft second data block of the
`storage medium, wherein any error protection provided
`by the second data block is at a lower level than the
`error protection provided by the first data block.
`6. An error resilient method of encoding data according to
`claim 1 wherein said step of providing error protection
`comprises the steps oi":
`
`‘Jl
`
`l0
`
`Zl}
`
`I~.a-A
`
`30
`
`40
`
`45
`
`{st}
`
`65
`
`Page 24 of 437
`
`Page 24 of 437
`
`

`
`5,850,482
`
`19
`transmitting the at least one first portion ofthe plurality of
`code words via a firs: data link, wherein the First data
`link is error protected; and
`transmitting the respective second portion associated with
`the at
`least one first portion via a second data link,
`wherein any error protection provided by the second
`data link is at a lower level than the error protection
`provided by the first data link.
`7. A data eneorliirg apparatus comprising:
`code word generating means for generating a plurality ol"
`code words representative of respective portions of the
`data, wherein each code word comprises a first portion
`and an associated second portion, and wherein said
`code word generating means comprises:
`first generating means for generating the first portion of
`each code word, said first generating means com-
`prising means for including information within the
`first portion that is representative ofa predetermined
`characteristic of the associated second portion; and
`Second generating means for generating the second
`portion of each code word, said second generating
`means comprising means for including information
`within the second portion that is representative of the
`respective portion of the data; and
`error protection means for providing error protection to at
`least one of the first portions of the plurality of code
`words while maintaining any error protection provided
`to the respective second portion arisociated with the at
`least one First portion at a lower level than the error
`protection provided to the respective first portion.
`3. Adata encoding apparatus according to claim 7 wherein
`said code word generating means comprises entropy coding
`means for entropy coding the data to thereby reduce the size
`of the resulting code words.
`9. A data encoding apparatus according to claim 7 wherein
`said second generating means generates second portions
`having predetermined numbers of characters, and wherein
`said first generating means generates first portions which
`include information representative of the predetermined
`number of characters which comprise the associated second
`portion.
`10. A data encoding apparatus according to claim 7
`wherein said error protection means comprises a storage
`medium for storing the plurality of code words, said storage
`medium being partitioned into a first data block which is
`error protected and a second data block, wherein any error
`protection provided by the second data hioek is at a lower
`level than the error protection provided by the first data
`block, wherein the at least first portion of the plurality of
`code words is stored in the tirst data block of the storage
`medium, and wherein the respective second portion associ-
`ated with the at least one first portion is stored in the second
`data block of the storage rnediumi
`ll. A data encoding apparatus according to claim 7
`wherein said error protection means comprises:
`first data link transmitting means for transmitting the at
`least one iirstportion of the plurality of code words via
`a first data Link, wherein the first data link is error
`protected; and
`second data link transmitting means for transmitting the
`respective second portion associated with the at least
`one first portion via a second data link, wherein any
`error protection provided by said second data link is at
`a lower level than the error protection provided by said
`first data link.
`I2. An error resilient method of compressing data com-
`prising the steps of:
`
`ll)
`
`2t}
`
`tou
`
`40
`
`45
`
`50
`
`{st}
`
`20
`transforming the data based upon a predetermined trans-
`formation function;
`quaritizing the transforrned data such that the quantized
`data has fewer unique coeilicierils than the lransforrnecl
`data; and
`encoding the quantized data, said encoding step compris-
`ing the steps of:
`generating a plurality of code words, representative of
`respective portions of the data. which have respec-
`tive tlrst and second portions, wherein said code
`word generating step comprises the steps of includ-
`ing information within the first portion that is rop-
`resentative of a predetermined characteristic of the
`associated second portion, and including information
`within the second portion that is representative of a
`respective portion of the data; and
`providing error protection to at least one of the first
`portions of the plurality of code words while main-
`taining any error protection provided to the respec-
`tive second portion associated with the at least one
`first portion at a lower level than the error protection
`provided to the respective first portion.
`13. An error resilient method of compressing data accord-
`ing to claim 12 wherein said step of encoding the quantized
`data comprises the step of entropy coding the quantized data
`to thereby reduce the size of the resulting code words.
`l4.An error resilient method of compressing data accord-
`ing to claim 12 wherein said step of generating a plurality of
`code words comprises the steps of:
`generating second portions having predetermined num-
`bers of characters; and
`generating lirst portions which include information rep-
`resentative of the predetermined number of characters
`which comprise the associated second portion.
`15. An error resilient method of oompressing data accord-
`ing to claim [4 further comprising the step of determining
`the probability of occurrence of respective ones of the
`quantized data values, wherein said step of generating
`second portions having predetermined numbers 0|‘ charac-
`ters comprises the step of generating a plurality of second
`portions having the same predetermined number of
`characters, and wherein the plurality of second portions
`which have the same predetermined number of characters
`comprise portions of respective code words which represent
`quantined data values having Corresponding probabilities of
`generation within a predetermined range of probabilities.
`16. An error resilient method of compressing data accord-
`ing to claim 12 wherein said step of providing error protec-
`tion comprises the steps of:
`storing the at least one first portion ofthe plurality ofcode
`words in a lirst data block of a storage medium,
`wherein the tirst data block is error protected; and
`storing the respective second portion associated with the
`at least one first portion in a second data block of the
`storage mcdiu m, wherein any error protection provided
`by the second data block is at a lower level than the
`error protection provided by the firs: data block.
`17. An error resilient method of compressing data accord-
`ing to claim 12 wherein said step of providing error protec-
`tion eomprises the steps of:
`transmitting the at least one first portion of the plurality of
`code words via a first data link, wherein the first data
`link is error protected; and
`transmitting the respective second portion associated with
`the at least one first portion via a second data link,
`wherein any error protection provided by the second
`
`Page 25 of 437
`
`Page 25 of 437
`
`

`
`5,850,482
`
`21
`data link is at it lower level than the error protection
`provided by the first data link.
`18. An error resilient method of compressing data accord-
`ing to claim 12 wherein said transforrning step comprises the
`step of lransforrriing the data based upon a wavelet trans-
`form.
`I9. An error resilient method of compressing data accord-
`ing to claim 18 wherein said transforming step comprises the
`step of transforming the data based upon a biorthogonal
`wavelet transform.
`20. An error resilient method of compressing data accord-
`ing to claim 12 wherein the transformed data includes a
`plurality of transformed eoefiicients, and wherein said quan-
`tizing step Comprises the step of detecting translbrmcd
`cocllicienLs below a predctennincd clipping threshold.
`21. An error resilient method of compressing data accord-
`ing to claim 20 further comprising the step of establishing a
`clipping threshold such that
`the ratio of the number of
`detected coeilicients to the number of transformed welli-
`cients which are not detected is at
`least as great as at
`predetermined clipping ratio.
`22. An error resilient data compression apparatus com-
`prising:
`a data transformer for transitioning the data based upon a
`predetermined transformation function;
`a data quantizer for quantizing the transformed data such
`that the quantized data has fewer unique coellicients
`than the transformed data; and
`El data encoder for encoding the quantized data, said data
`encoder comprising:
`code word generating means for generating a plurality
`of code words, representative of respective portions
`of the data, which have respective llrst and second
`portions, wherein said code word generating means
`comprises means for including information within
`the first portion that is representative of a predeter-
`mined characteristic of the associated second
`portion, and means for including information within
`the second portion that is representative of a respec-
`tive portion of the data; and
`error protection means for providing error protection to
`at least one of the first portions of the plurality of
`code words while maintaining any error protection
`provided to the respective second portion associated
`with the at least one first portion at a lower level than
`the error protection provided to the respective first
`portion.
`23. An error res.-iilicnt data compressittn apparatus accord-
`ing to claim 22 wherein said data encoder comprises entropy
`coding means for entropy coding the quantized data to
`thereby reduce the size of the resulting code words.
`24. An error resiiient data compression apparatus accord-
`ing to claim 22 wherein said code word generating means
`C01’Tlp1'tSL‘.SI
`second generating means for generating second portions
`hav