des brevets
`
`Europalsches
`Patentamt
`European
`Patent Office
`Office européen
`
`TATA
`EP 2 866 443 At
`
`(11)
`
`(12}
`
`EUROPEAN PATENT APPLICATION
`published in accordance with Art. 153(4) EPC
`
`(43} Date of publication:
`29.04.2015 Bulletin 2015/18
`
`(51) Int Ch:
`HOAN 7126 (9000.00)
`
`(21) Application number: 13807680.7
`
`(86) International application number:
`PCT/JP2013/065423
`
`(22) Date offiling: 04.06.2013
`
`(87) International publication number:
`WO 2013/190990 (27.12.2013 Gazette 2013/52)
`
`(84} Designated Contracting States:
`AL AT BE BG CH CY CZ DE DK EE ESFIFRGB
`GR HR HU IE IS IT LILT LULY MC MK MT NL NO
`PL PT RO RS SE $1 SK SMTR
`Designated Extension States:
`BA ME
`
`«
`
`IKAI, Tomohiro
`Osaka-shi
`
`Osaka 545-8522 (JP)
`¢ YAMAMOTO, Tomoyuki
`Osaka-shi
`
`Osaka 545-8522 (JP)
`
`(30} Priority: 22.06.2012 JP 2012141416
`
`{74} Applicant: Sharp Kabushiki Kaisha
`Osaka-shi, Osaka 545-8522 (JP)
`
`(74) Representative: Miller-Boré & Partner
`Patentanwalte PartG mbB
`Friedenheimer Briicke 21
`
`80639 Miinchen (DE)
`
`(72) inventors:
`* TSUKUBA, Takeshi
`Osaka-shi
`
`Osaka 545-8522 (JP)
`
`
`(54}
`
`ARITHMETIC DECODING DEVICE, ARITHMETIC CODING DEVICE, IMAGE DECODING DEVICE
`AND IMAGE CODING DEVICE
`
`with respect to each transform coefficient in a pixel do-
`(57)~~A coefficient presence/absence flag decading
`unit (124), included in a quantized residual information
`main, obtained through a transform skip or a transform/
`decoding unit (111) which reduces a processing amount
`quantization bypass, and each transform coefficient
`related to coding and decoding of a transform coefficient,
`which is obtained for each frequency camponent through
`splits a target frequency domain or a target pixel domain
`frequency transform.
`into sub-blocks, and derives different context indexes
`
`a4
` epeenn.
`
`EP2866443At
`
`
`
`
`SYPASS FLAG
` JASTIZATION
`sqeant_byoess.fag)
`
`
`!
`
`
`.
`iy
`
`~S*
`jast_significantcost»: ;
`lest_significant_cosff_y
`
` xOG, y06
`
`
`COSFACIENT DECODING
`
`se
`UNIT
`CONTROL
`
`
`LI
` [once
`PRESENOEABSENCE
`
`J
`FLAG DECODING
`signdicant_costt_flagixC]fyC}
`
`
`
`
`
`
`
`Printed by Jouve, 75001 PARIS (FR)
`
`

`

`EP 2 866 443 A
`
`Description
`
`Technical Field
`
`{0001} The present invention relates to an arithmetic decoding device for decoding coded data whichis arithmetically
`coded, and an image decoding apparatus including the arithmetic decading device.
`In addition, the present invention
`relates to an arithmetic coding device for generating coded data which is arithmeticaily coded, and an image coding
`apparatus including the arithmetic coding device.
`
`10
`
`Background Art
`
`15
`
`In order to efficiently transmit or record moving images, a moving image coding apparatus (image coding
`[0002]
`apparatus) for generating coded data by coding a moving image, and a moving image decoding apparatus (image
`decoding apparatus) for generating a decoded image by decading the coded data, are used.
`[0003] Asa specific moving image coding method, for example, there are methods {NPL 1} proposed in H. 264/MPEG-
`4, AVC, and High-Efficiency Video Cading (HEVC) which is a succeeding codec thereof.
`[0004]
`In such moving image coding methods, an image (picture} forming a moving image is managedin a layer
`structure which is constituted by a slice obtained by dividing the image, a coding unit obtained by dividing the slice, and
`a block and a partition obtained dy dividing the coding unit, and the image is commonly coded and decoded for each block.
`20
`{0005}=In addition, in these coding methods, typically, a predicted image is generated on the basis of a lacal decoded
`image obtained by coding and decoding an input image, and coding is performed on a transform coefficient which is
`obtained by performing frequencytransform such as discrate cosine transform (DCT) ona difference image (also referred
`to as a "residual image” or "prediction residual" in some cases} between the predicted image and the input image for
`eachblock.
`
`25
`
`{0006} Asaspecificcoding method ofa transform coefficient, context-based adaptive binary arithmetic coding (CABAC)
`is known.
`[0007]
`In the CABAC, a binarization process is performed on various syntaxes indicating a transform coefficient, and
`binary data obtained thraugh the binarization processis arithmetically coded. Here, the various syntaxes include a flag
`indicating whether or not a transform coefficient is 0, that is, a flag significant_coeff_fiag (also referred to as coefficient
`presence/absenceflag) indicating whether or net a non-zero coefficient is present, a flag coeff_abs_level_greater1_flag
`(also referred to as a GR1 flag} indicating whether or not an absolute value of a transform coefficient exceeds 4, a flag
`coeff_abs_level_greater2_flag (also referred to as a GR2 flag) indicating whether or not an absolute value of a transform
`coefficient exceeds 2, syniaxes jast_significant_coeff_x and last_significant_coeff_y indicating a position of the last non-
`zero coefficient in a process order, and the like.
`{0008}
`In addition, in the CABAC, when a single symbol (also referred to as 1 bit of binary data, or a Bin) is coded, a
`context index is referred to, and arithmetic coding corresponding to an occurrence probability indicated by a probability
`state index included in a context variable designated by the context index is performed. For example, in a case where
`a target syntax of arithmetic coding is a coefficient presence/absenceflag, the context index is derived in accordance
`with a position in a frequency component of a transform coefficient. In addition, the occurrence probability designated
`by the probability state index is updated whenever a single symbol is coded. Further, in the following description, a
`process of deriving a context index which is referred to when a specific Bin is coded is also referred to as a context
`deriving process.
`{0009}
`In NPL 1, as @ method of coding signs indicating positive and negative transform coefficients, sign hiding is
`employed. The sign hiding is a method in which a sign of a transform coefficient is not explicitly caced but is calculated
`thraugh computation.
`{0010}=In addition, in NPL 1, two-layer coding is employed as a method of coding a non-zero coefficient. In the two-
`layer coding, the transform unit is split into a plurality of sub-blocks, a flag (significant_coeff_fiag) indicating whether or
`not a transform coefficient is nen-zere is coded for each transform coefficient included in each sub-block, and a flag
`(significant_coeff_group_flag) indicating whether or not a non-zero coefficient is included in each sub-block is coded in
`the sub-block units.
`
`30
`
`a5
`
`40
`
`45
`
`ao
`
`Further, in NPL 1, the following coding is performed as follows in accordance with a size of the transform unit
`{0011}
`(TU), In other words, in the small TU (4x4 or 8x8), 4x4 or 8x2 is used as a sub-blocksize, and a context which is assigned
`to a frequency component on the basis of a position is derived. In the context derivation based on a position, a context
`index (also referred to as a position context) which is defined in accordance with a position of a frequency component
`in a frequency domain is assigned to the frequency component.
`{0012}
`In addition, in the large TU (16x16, 32x32, 16x4, 4x16, 32x8, or 8x32), 4x4 is used as a sub-block size, anda
`context which is assigned to a frequency component on the basis of periphery reference is derived.
`In the context
`derivation based on periphery reference, a context index (also referred to as a periphery reference context} which is
`
`

`

`EP 2 866 443 A1
`
`defined in accordance with the number of non-zero coefficients (that is, significant_caeff_flag is referred to) in peripheral
`frequency companents of a corresponding frequency component is assigned to the frequency component.
`{0013}
`In addition, NPL 2 discloses a technique for controlling whether or not a frequency transform processis skipped
`for each transform unit. Hereinafter, skipping of the frequency transform process is referred to as a transform skip.
`{0014}
`Further, NPL 3 discloses a technique for controlling whether or not a frequency transform process or a quan-
`tization and loop filter process are skipped for each coding unit. Hereinafter, skipping of the frequency transform and
`quantization processesis referred to as a transform/quantization bypass. The transform/quantization bypass is mainly
`usedfor jossless coding. For this reason, the skipping of the frequency transform and quantization processes may not
`be referred to as a transform/quantization bypass but is simply referred to as a jossless mode.
`
`Citation List
`
`Non Patent Literature
`
`[0015] NPL 1: "Suggested bug-fixes for HEVC text specification draft6 (JCTVC-10030)", Joint Collaborative Team on
`Video Cading (JCT-VC) of ITU-T SG16 WP3 and ISO/IEC JTC1/SC2SAWG11 9th Meeting: Geneva, CH, 27 April to 7
`May 2012 (published in April, 2042)
`{0016} NPL 2: #Transform Skipping (JCTVC-i0408} “Intra transform skipping (JCTVC-10408)", Joint Collaborative
`Team on Video Coding (JCT-VC} of ITU-T SG 16 WP 3 and ISO/IEC JTC 1/SC 29AVG 11 9th Meeting: Geneva, CH,
`27 April to 7 May 2012 (published in April, 2072)
`{0017} NPL 3:#cu_transquant_bypass_flag (JCTVC-l0529) "AHG 13: Praposed bugfix for tickets 410 and 470 related
`fo lossiess coding (JCTVC-10529)"", Joint Collaborative Team on Video Coding (JCT-VC} of ITU-T SG 16 WP 3 and
`ISO/MEC JTC 1/S8C 29/WG 11 9th Meeting: Geneva, CH, 27 April to 7 May 2012 (published in April, 2042)
`
`Summary of Invention
`
`Technical Problem
`
`{0018} However, in the above-described related art, since contexts of a flag indicating whether or not a non-zero
`coefficient is present or flags {a GR1 flag and a GR2 flag} indicating whether or net a level value is a specific value are
`not the same in TU or a sub-block in a case where the transform skip or the transform/quantization bypass is applied,
`there is a problem in that a reduction in a processing amount reiated ta coding and decoding of transform coefficients
`is not sufficient.
`In addition, there is a problem in that a process fer performing sign hiding in which signs of some
`coefficients are not decoded in the transform skip is necessary.
`{0019}
`Further, in a case where the transform skip and the transform/quantization bypass are applied, there is a
`tendency for an appearance probability of coefficients in a TU or a sub-block to be the same, but, in the related art, a
`context is changed depending on a position of a transform coefficient and thus there is a problem in that coding efficiency
`is not sufficient.
`
`10
`
`15
`
`20
`
`25
`
`30
`
`a5
`
`40
`
`Furthermore, in a case where the same context is used in a sub-blockin the lassiess mode, there is a problem
`{0020}
`in that the number of context indexes increases.
`
`45
`
`ao
`
`In addition, in a case where sign hiding is performed in the transform skip, there is a problem in that in coding
`[0021]
`efficiency is reduced due to the sign hiding. There is a problem in that a reduction in a processing amount of coding and
`decoding of transform coefficients is not sufficient.
`{0022} The present invention has been madein consideration of the above-described problems, and an object thereof
`is to implement an arithmetic decoding device, an arithmetic coding device, and the like capabie of reducing a processing
`amount reiated to coding and decoding of a transform coefficient and improving coding efficiency when compared with
`the configuration of the related art.
`
`Solution ta Problem
`
`In order to solve the above-described problems, according to an aspect of the present invention, there is
`[0023]
`provided an arithmetic decoding device which decodes coded data which is obtained by arithmetically coding various
`syntaxes indicating a transform coefficient with respect to each transform coefficient which is obtained for each frequency
`component by performing frequency transform on a target image for each unit domain and each transform coefficient
`in a pixel domain, obtained through a transform skip, the device including sub-block splitting means for splitting a target
`frequency domain or a target pixel domain corresponding to a process target unit domain target into sub-biocks each
`having a predetermined size; sub-biock coefficient presence/absence flag decoding means for decoding a sub-block
`coefficient presence/absenceflag indicating whether or not at least one non-zero coefficientis included in the sub-biock
`
`

`

`EP 2 866 443 A1
`
`with respect to the respective sub-blocks into which the frequency domain or the pixel domain is split by the sub-biock
`splitting means; and non-zero coefficient presence/absence tlag decoding means for decoding a non-zero coefficient
`presence/absence flag indicating whether or not a transform coefficient in each of the sub-blocksis 0, in which the non-
`zero coefficient presence/absenceflag decoding means derives separate context indexes for each of a non-zera coef-
`ficient presence/absence flag corresponding to each transform coefficient in the pixel domain, obtained through the
`transform skip and a non-zero coefficient presence/absence flag correspanding to each transform coefficient which is
`obtained for each frequency component through the frequency transform.
`{0024}
`Ina case wherethe transform skip is perfarmed, a decoding target coefficient presence/absenceflag is a non-
`zero coefficient in a pixel domain. Since an appearance probability of a non-zero coefficient at each coefficient position
`in a pixel domain and an appearance probability of a non-zero coefficient at each coefficient position in a frequency
`damain are different from each other, it is preferable to differentiate contexts regarding coefficient presence/absence
`flags from each otherin the pixel domain and the frequency domain.
`[0025} According to the configuration, in a case where at least the transform skip is performed,it is possible to derive
`a context index which is suitable for a non-zero coefficient in a pixel domain. For this reason, since a code amount of a
`non-zero coefficient can be reduced in a TU which is a coding and decoding target,it is possible to reduce a processing
`amount related to decoding of the non-zero coefficient and also to improve coding efficiency.
`[0026]
`In the arithmetic decoding device according to the aspect of the present invention, the non-zero coefficient
`presence/absence flag decoding means may derive a fixed context index in a transform block in relation to a non-zero
`coefficient presence/absence flag corresponding to each transform coefficient in the pixel domain, obtained through the
`transform skip.
`{0027}
`In the arithmetic decoding device according to the aspect of the present invention, the non-zero coefficient
`presence/absence flag decoding means may derive a value based on a transform block size as a context index,
`in
`relation to a non-zero coefficient presence/absence flag corresponding to each transform coefficient in the pixel domain,
`obtained through the transform skip.
`{0028} Aiso in a pixel damain, since appearance probabilities of a non-zero coefficient are different depending on a
`TU size, it is preferable to differentiate contexts regarding coefficient presence/absence flags from each other for re-
`spective TU sizes. According to the configuration, itis possible to derive a context index which is suitable for a non-zero
`coefficient in the pixel domain for each TU size. For this reason, since a code amount of a non-zero coefficient can be
`further reduced in a TU which is a coding and decoding target, it is possible to reduce a processing amount related to
`decoding of the non-zero coefficient and also to improve coding efficiency.
`{0029}
`In the arithmetic decoding device according to the aspect of the present invention, the non-zero coefficient
`presence/absenceflag decading means may derive a predetermined context index on the basis of a prediction type and
`a transform block size,
`in relation to a non-zers coefficient presence/absence flag corresponding to each transform
`coefficient in the pixel domain, obtained through the transform skip.
`{0030}
`In a case where the transform skip is performed, since appearance probabilities of a non-zero coefficient in a
`decoding target pixel domain are different from each other in intra-prediction and inter-predicticn,
`it is preferable to
`differentiate contexts regarding coefficient presence/absence flags from each other for respective prediction types.
`According to the configuration, itis possible to derive a context index suitable for a non-zero coefficient in a pixel domain
`far each prediction type and each TU size. For this reason, since a cade amountof a non-zero coefficient can be reduced
`in a TU which is a coding and decoding target, it is possible to reduce a processing amount related to decoding of the
`non-zero coefficient and also to improve coding efficiency.
`{0031}
`In the arithmetic decoding device according to the aspect of the present invention, the non-zero coefficient
`presence/absence flag decoding means may derive separate context indexes for each of a non-zero coefficient pres-
`ence/absence flag during the transform skip and a non-zero coefficient presence/absence flag during a transform/quan-
`tization bypass, in relation to the non-zero coefficient presence/absenceflag corresponding to each transform coefficient
`in the pixel domain, obtained through the transform skip or the transform/quantization bypass.
`{0032} Generally, since appearance probabilities of a non-zero coefficient in a pixel domain are different from each
`other in the transform skip and the transform/quantization bypass, it is preferable to differentiate contexts regarding
`coefficient presence/absence flags fram each other. According to the configuration, it is possible to derive a context
`index whichis suitable for a non-zero coefficient in a pixel domain for each of the transform skip and the transform/quan-
`tization bypass. For this reason, since a code amount of a non-zero coefficient can be reduced in a TU which is a coding
`and decoding target, it is possible to reduce a processing amount related to decoding of the non-zero coefficient and
`also to improve coding efficiency.
`[0033] The arithmetic decoding device according to the aspect of the present invention may further include GR1 flag
`decoding means for decoding a GR1 flag indicating whether cr not an absolute value of a transform coefficient in each
`of the sub-blocks exceeds 1, and the GR1 flag decoding means may derive separate context indexes for each of a GR1
`flag corresponding to each transform coefficient in the pixel domain, obtained through the transform skip and a GR7 flag
`corresponding to each transform coefficient which is obtained for each frequency component through the frequency
`
`10
`
`15
`
`20
`
`25
`
`30
`
`a5
`
`40
`
`45
`
`ao
`
`

`

`EP 2 866 443 A1
`
`transform.
`
`In a case where the transform skip is performed, a decoding target GR? flag is a GR1 flag in a pixel domain.
`[0034]
`Since an appearance probability ofa GR1 flag at each coefficient position in a pixel damain and an appearance probability
`ofa GR? flag at each coefficient pasition in a frequency domain are different fram each other,it is preferable to differentiate
`contexts regarding GR1 flags from each other in the pixel domain and the frequency domain,
`[0035} According to the configuration, in a case where at least the transform skip is performed,it is possible to derive
`a context index which is suitable for a GR1 flag in a pixel domain. Forthis reason, since a code amount af a GR1 flag
`can be reduced in a TU which is a coding and decoding target, it is possible to reduce a pracessing amountrelated to
`decading of the GR1 flag and also to improve cading efficiency.
`[0036]
`In the arithmetic decoding device according to the aspect of the present invention, the GR1 flag decoding
`means may derive a value based on a transform block size as a context index, in relation to a GR1 flag corresponding
`to each transform coefficient in the pixel domain, obtained through the transform skip.
`[0037] Also in a pixel domain, since appearance probabilities of a GR1 flag are different depending on a TU size, it is
`preferable to differentiate contexts regarding GR1 flags fram each other for respective TU sizes. According to the
`configuration, it is possible to derive a context index which is suitable for a GR1 flag in the pixel domain for each TU
`size. For this reason, since a code amount of a GR1 flag can be further reduced in a TU whichis a coding and decoding
`target, it is possible to reduce a processing amount related to decoding of the GR1 flag and also to improve coding
`efficiency.
`[0038]
`In the arithmetic decoding device according to the aspect of the present invention, the GR1 flag decoding
`means may derive a predetermined context index on the basis of a prediction type and a transform biock size, in relation
`to a GR1 flag corresponding to each transform coefficient in the pixel domain, obtained through the transform skip.
`[0039]
`In a case where the transform skip is performed, since appearance probabilities of a decoding target GR‘ in
`a pixel domain are different from each other in intra-prediction and inter-prediction, it is preferabie to differentiate contexts
`regarding GR1 flags from each other for respective prediction types. According to the configuration, it is passible to
`derive a context index suitable for a GR1 flag in a pixel domain for each prediction type and each TU size. For this
`reason, since a code amount of a GR1 flag can be reduced in a FU which is a coding and decading target, it is possible
`{ao reduce a processing amount related to decoding of the GR? flag and also to improve coding efficiency.
`[0040]
`In the arithmetic decoding device according to the aspect of the present invention, the GR1 flag decoding
`means may derive separate context indexes for each of a GR1 flag during the transform skip and a GR? flag during a
`transform/quantization bypass,in relation to the GR1 flag corresponding to each transform coefficient in the pixel domain,
`obtained through the transform skip or the transfarm/quantization bypass.
`[0041] Generally, since appearance probabilities of a decoding target GR1 flag in a pixel domain are different from
`each otherin the transform skip and the transform/quantization bypass,it is preferable to differentiate contexts regarding
`GR1flags from each other for respective prediction types. According to the configuration, itis possible to derive a context
`index which is suitable for a GR1 flag in a pixel domain for each of the transform skip and the transform/quantization
`bypass. For this reason, since a code amount of a GR1 flag can be reduced in a TU which is a coding and decoding
`target, itis possible to reduce a processing amount related to decoding of the GR1 flag and also to improve coding
`efficiency.
`{0042} The arithmetic decoding device according to the aspect of the present invention may further include GR2 flag
`decoding means for decoding a GR?flag indicating whether or not an absolute value of a transform coefficient in each
`of the sub-blocks exceeds 2, and the GR2 flag decading means may derive separate context indexes for at feast each
`transform coefficient in the pixel domain, obtained through the transform skip and each transform coefficient which is
`obtained for each frequency component through the frequency transform.
`{0043}
`In acase where the transform skip is performed, a decoding target GR2 flag is a GR2 flag in a pixel damain.
`Since an appearance probability of a GR2 flag at each coefficient position in a pixel domain and an appearanceprobability
`ofa GR2flag at each coefficient position in a frequency domain are different from each other,itis preferable to differentiate
`contexts regarding GR? flags from each other in the pixel domain and the frequency domain.
`[0044] According to the configuration, in a case where the transform skip is performed, itis possible to derive a context
`index which is suitable for a GR2 flag in a pixel domain. Far this reason, since a code amount of a GR2 flag can be
`reduced in a TU which js a coding and decoding target, it is possible to reduce a processing amountrelated to decoding
`of the GR2 flag and also to improve coding efficiency.
`{0045}
`In the arithmetic decoding device according to the aspect of the present invention, the GR2 flag decoding
`means may derive a value based an a transform block size as a context index, in relation to a GR2 flag corresponding
`to each transform coefficient in the pixel domain, obtained through the transform skip.
`[0046] Also in a pixel domain, since appearanceprobabilities of a GR2 flag are different depending on a TU size, itis
`preferable to differentiate contexts regarding GR2 flags fram each other for respective TU sizes. According to the
`configuration, it is possible to derive a context index which is suitable for a GR2 flag in the pixel domain for each TU
`size. For this reason, since a code amount of a GR2 flag can be further reduced in a TU which is a coding and decoding
`
`10
`
`15
`
`20
`
`25
`
`30
`
`a5
`
`40
`
`45
`
`ao
`
`

`

`EP 2 866 443 A1
`
`10
`
`15
`
`20
`
`25
`
`30
`
`target, itis possible to reduce a processing amouni related to decoding of the GR2 flag and also to improve coding
`efficiency.
`[0047]
`In the arithmetic decoding device according to the aspect of the present invention, the GR2 flag decoding
`means may derive a predetermined context index on the basis of a prediction type and a transform block size, in relation
`to a GR2 flag corresponding to each transform coefficient in the pixel domain, obtained through the transform skip.
`{0048}
`In acase wherethe transform skip is performed, since appearance probabilities of a decoding target GR2 in
`a pixel domain are different from eachother in intra-prediction and inter-prediction, it is preferable to differentiate contexts
`regarding GR? flags from each other for respective prediction types. According to the configuration, it is possible to
`derive a context index suitable for a GR2 flag in a pixel domain for each prediction type and each TU size. For this
`reason, since a code amount of a GR2 flag can be reduced in a FU whichis a coding and decoding target,it is possible
`fo reduce a processing amount related to decoding of the GR2 flag and alsa to improve coding efficiency.
`[0049]
`In the arithmetic decoding device according to the aspect of the present invention, the GR2 flag decoding
`means may derive separate context indexes for each of a GR2 flag during the transform skip and a GR2 flag during a
`transform/quantization bypass,in relation to the GR2 flag corresponding to each transform coefficient in the pixel domain,
`obtained through the transform skip or the transform/quantization bypass.
`{0050} Generally, since appearance probabilities of a GR2 flag in a pixel domain are different from each other in the
`transform skip and the transform/quantization bypass, itis preferable to differentiate contexts regarding GR2 flags from
`each other. According ta the configuration, it is possible to derive a context index which is suitable for a GR2 flag in a
`pixel domain for each of the transform skip and the transform/quantization bypass. For this reason, since a code amount
`of a GR2 flag can be reduced in a TU which is a coding and decoding target, itis possible to reduce a processing amount
`related to decoding of the GR? flag and also fa improve coding efficiency.
`{0051}=In order ta solve the above-described problems, according to another aspectof the present invention, there is
`provided an arithmetic decoding device which decodes coded data which is obtained by arithmetically coding various
`syntaxesindicating a transform coefficient with respect to each transform coefficient whichis obtained for each frequency
`component by performing frequency transform on a target image for each unit damain and each transform coefficient
`in a pixel domain, obtained through a transform skip or a transform/quantization bypass, the device including coefficient
`sign decoding means for deriving a sign of a transform coefficient on which sign hiding is performed, by using an absolute
`value of a transform coefficient whose sign has been decoded cr a sum of absolute values of transform coefficients
`whose signs have been decoded, and for deriving a sign of a transform coefficient on which the sign hicing is not
`performed, by decoding a syntax fram coded data; and sign hiding flag deriving means for deriving a sign hiding flag
`used to determine whether or not the sign hiding is to be performed, on the basis of a transform skip flag, a trans-
`form/quantization bypassflag, and a difference between pasitions of decaded non-zero coefficients in a target sub-block,
`or through comparison between ike numberof decoded non-zero coefficients in the target sub-block and a predetermined
`threshold value.
`{0052}
`Inacase where the transform skip is performed, or the transform/quantization bypass is performed, performing
`sign hiding which is accompanied by correction of a coefficient level value on a coefficient of a pixel domain may possibly
`cause deterioration in subjective image quality, and thus the sign hiding is preferably prohibited. As mentioned above,
`in this configuration, by prohibiting sign hiding in the transform skip or the transform/quantization bypass, it is possible
`fo improve subjective image quality.
`{0053}
`Inorder to solve the above-described problems, according to still ancther aspectof the presentinvention, there
`is provided an image decoding apparatus including the arithmetic decoding device; inverse frequency transform means
`for generating a residual image by perfarming inverse frequency transform an a transform coefficient which is decaded
`by the arithmetic decoding device; and decoded image generating means for generating a decoded image by adding
`the residual image generated by the inverse frequency transform means to a predicted image which }s predicted from
`a generated decaded image.
`[0054]
`In order to solve the above-described problems, according to still ancther aspectof the present invention, there
`is provided an arithmetic coding device which generates coded data by arithmetically coding various syntaxes indicating
`a transform coefficient with respect to each transform coefficient which is obtained for each frequency component by
`performing frequency transform on a target image for each unit domain and each transform coefficient in a pixel domain,
`obtained through a transform skip, the device including sub-block splitting means forsplitting a target frequency domain
`corresponding to a process target unit domain into sub-blocks each having a predetermined size; sub-block coefficient
`presence/absence flag coding means for coding a sub-block coefficient presence/absence flag indicating whether or
`not al least one non-zero coefficient is included in the sub-block with respect to the respective sub-blocks into which the
`frequency domainis split by the sub-block splitting means; and non-zero coefficient presence/absence flag coding means
`for ceding a non-zero coefficient presence/absence flag indicating whether or not a transform coefficient in each of the
`sub-blocks is 0, in which the non-zero coefficient presence/absence flag coding rmeans uses separate context indexes
`for each of a non-zero coefficient presence/absence flag corresponding to each transform coefficient in the pixel domain,
`obtained through the transform skip and a non-zero coefficient presence/absenceflag corresponding to each transform
`
`a5
`
`40
`
`45
`
`ao
`
`

`

`EP 2 866 443 A1
`
`coefficient which is obtained for each frequency component through the frequency transform.
`[0055]
`In order to solve the above-described prablems, accarding to stilt another aspectof the present invention, there
`is provided an image coding apparatus including transform coefficient generating means for generating a transform
`coefficient by performing frequency transform on a residual image between a coding target image and a predicted image
`for each unit domain; and the arithmetic coding device, in which the arithmetic coding device generates coded data by
`arithmetically coding various syntaxes indicating the transform coefficient which is generated by the transform coefficient
`generating means.
`
`Advantageous Effects of Invention
`
`{0056} As described above, the arithmetic decoding device according to the present invention incfudes the sub-block
`splitting means for splitting a target frequency domain or a target pixel domain corresponding ta a processtarget unit
`domain into sub-blocks each having a predetermined size; the sub-block coefficient presence/absence flag decoding
`means for decoding a sub-block coefficient presence/absence flag indicating whether or not at least one non-zero
`coefficient is included in the sub-block with respect to the respective sub-blocks into which the frequency domain or the
`pixel domai