KR20220134309A - A method and an apparatus for processing a video signal - Google Patents

Abstract

The present invention provides a method for encoding/decoding a residual signal and an apparatus therefor. The purpose of the present invention is to improve the coding efficiency of a video signal. A method for encoding a residual signal in a block, an RC method, a TSRC method, a truncated rice (TR) binarization method, and a limited k-th order exp-golomb binarization method are provided.

Description

Video signal processing method and apparatus {A METHOD AND AN APPARATUS FOR PROCESSING A VIDEO SIGNAL}

The present invention relates to a video signal processing method and apparatus.

The video image is compression-encoded by removing temporal and spatial redundancy and inter-view redundancy, which may be transmitted through a communication line or stored in a form suitable for a storage medium.

An object of the present invention is to improve the coding efficiency of a video signal.

In order to solve the above problems, a method for encoding/decoding a residual signal and an apparatus therefor are provided.

A video signal processing method and apparatus according to the present invention can improve image encoding/decoding efficiency.

Recently, ultra-high-resolution video is the core of not only digital broadcasting but also streaming services such as Netflix and YouTube. In addition to the existing 2D images, VR and 3D image services are being commercialized, and the above image services can be used not only on digital TVs but also on mobile devices such as smartphones. What these video services have in common is that the service is impossible without the application of video compression. In the case of 1080p@60Hz, which can be called Full-HD, a 1920x1080 screen must be transmitted 60 times per second. Like 3D video, twice as much data is required to deliver information to both eyes, and 4K (4096x2048) and 8K (8192x4096) ultra-high-resolution video services require one screen to be transmitted more than 120 times per second. Huge amount of data is generated compared to HD. In order to handle such data, technologies in various fields such as communication bandwidth and image compression technology are required. First, predictive techniques are used on the data to generate residual values. The more accurately the prediction is performed, the closer the residual signals with the original data are to zero. After that, energy is concentrated through transformation, and the quantized coefficient values are encoded with reference to context information.

In particular, in the case of encoding using a low QP, or in the case of a 12-bit or more medical image or an ultra-high-definition image, the residual signal to be encoded may become extremely large. If an encoding method optimized for these values is used, encoding efficiency can be greatly increased.

One. In-block residual signal encoding method

When encoding a block, a residual signal is generated, and entropy encoding is performed on the generated residual signal. This residual signal may be generated in various ways. For example, a residual block including residual values is generated by differentiating prediction samples (pixels) generated through prediction such as inter prediction or intra prediction from samples (pixels) existing in the original block. A quantized transform coefficient obtained by transforming and quantizing the residual blocks may be set as a residual signal. Alternatively, the residual signal may be generated while at least one of prediction, transformation, and quantization is skipped.

The generated residual signal is expressed in various syntaxes, and entropy encoding is performed on them. After being changed into a plurality of bins through a binarization process, it is encoded using entropy encoding. In this case, after the block is additionally divided into sub-block units, entropy encoding may be performed on the residual signal in sub-block units.

Entropy encoding may include encoding using context information and encoding not using context information.

In order to increase the throughput of the entropy coder, the number of bins for performing entropy encoding may be limited by using context information for each block to be encoded. For this, a threshold is set. Encoding using context information is performed for a predetermined threshold value among a plurality of generated bins. The number of encoded bins is counted using context information, and when the counted value exceeds a threshold value, the remaining bins are encoded without using context information.

The threshold may be adaptively determined. For example, the threshold value may include the size and shape of a block, the number of non-zero residual signals, whether a transform (or transform skip) is applied, a type of a transform kernel, whether quantization is applied, a prediction mode, and a color component (YCbCr). ), a quantization parameter, and a bit depth may be determined based on at least one of (BitDepth).

For example, the threshold value may be determined based on the number of non-zero quantized coefficients present in the block. For example, a real multiple of the number or a value obtained by adding/subtracting an offset to the number may be defined as the threshold value.

For example, the threshold value may be determined based on the number of pixels (ie, block size) existing in the block. For example, a real multiple of the number or a value obtained by adding/subtracting an offset to the number may be defined as the threshold value.

As another example, information for determining a threshold may be encoded and signaled. For example, the information may be encoded through a higher header or transmitted to a decoder.

As another example, a threshold value may be predefined for each block size/type. Alternatively, after defining threshold value candidates for each block size/type, one of a plurality of threshold value candidates may be encoded and signaled.

Alternatively, a fixed value in the encoder and the decoder may be defined as the threshold value.

As a method for encoding the residual signal, there are a residual codong (RC) method and a transform skip residual coding (TSRC) method. For example, if the RC method is applied to a block, all subblocks existing in the block are performed by RC. Here, the sub-block means a unit in which TSRC or RC is used before entropy encoding is performed.

1-1: RC method

Figure 1 below is an example of the RC method. For convenience of description, it is assumed that the size of the sub-block is 4x4. Also, it is assumed that the number of bins to be encoded using context information (ie, a threshold value, coded context bin (CCB)) is 52.

Figure 1

In Figure 1 above, C ₁₅ ~ C ₀ means residual signals in the sub-block. The residual signal in the sub-block is encoded from the lower-right direction to the upper-left direction in the sub-block. In the above figure, C ₁₅ is a signal that exists at the lower right of the sub-block, and C ₀ means a signal that exists at the upper left of the sub-block. In addition, a portion indicated by a black arrow (context coding) indicates bins encoded using context information, and a portion indicated by a blue arrow (bypass coding) indicates bins encoded without using context information. Also, encoding may be performed on each syntax along the arrow direction.

In addition, sig_coeff_flag, abs_level_gtx_flag[0], par_level_flag, abs_level_gtx_flag[1], abs_remainder, dec_abs_level, and coeff_sign_flag may be defined as in the following equations.

Pass syntax name Equation pass 1 sig_coeff_flag C _N != 0 abs_level_gtx_flag[0] !! ( |C _N | - 1 ) par_level_flag ( |C _N | - 2 ) & 1 abs_level_gtx_flag[1] ( |C _N | - 2 ) >> 1 pass 2-1 abs_remainder ( |C _N | - 4 ) >> 1 pass 2-2 dec_abs_level pass 3 coeff_sign_flag C _N < 0 ? 1: 0

If the value of C ₁₅ is -21, sig_coeff_flag corresponding to C ₁₅ is 1, abs_level_gtx_flag[0] is 1, par_level_flag is 1, abs_level_gtx_flag[1] is 1, abs_remainder is 8, and the value of coeff_sign_flag is It is expressed as 1, and these syntaxes are entropy-encoded and signaled to the decoder.

The decoder restores the residual signal using the following table.

Equation TmpC _N sig_coeff_flag + abs_level_gtx_flag[0] + par_level_flag + (abs_level_gtx_flag[1]<<1) |C _N | TmpC _N + (abs_remainder<<1)

As described above, when |C ₁₅ | is restored, the value of TmpC _N becomes 5 through 1+1+1+(1<<1), and is restored to 21 by adding the result of (abs_remainder<<1). After that, the value of C ₁₅ is finally restored to -21 through the coeff_sign_flag value.

1-2: TSRC method

Figure 2 below is an example of the TSRC method. For convenience of description, it is assumed that the size of the sub-block is 4x4. Also, it is assumed that the number of bins (ie, a threshold value, coded context bin (CCB)) encoded using context information is 112.

Figure 2

In Figure 2 above, C ₀ ~ C ₁₅ means residual signals existing in the sub-block. When the TSRC method is applied, the encoding of the residual signal existing in the sub-block is performed from the upper-left direction to the lower-right direction in the sub-block. In the above figure, C ₁₅ is a signal that exists at the lower right of the sub-block, and C ₀ means a signal that exists at the upper left of the sub-block. In addition, a portion indicated by a black arrow (context coding) indicates bins on which encoding is performed using context information. A portion indicated by a blue arrow (bypass coding) indicates a bin that is coded without using context information. Also, in the direction of the arrow, encoding may be performed for each syntax.

In addition, sig_coeff_flag, coeff_sign_flag, abs_level_gtx_flag[0], par_level_flag, abs_level_gtx_flag[1], abs_level_gtx_flag[2], abs_level_gtx_flag[3], abs_level_gtx_flag[4] are defined as in the following table,

pass syntax name Equation pass 1 sig_coeff_flag C _N != 0 coeff_sign_flag C _N < 0 ? 1: 0 abs_level_gtx_flag[0] !! ( |C _N | - 1 ) par_level_flag ( |C _N | - 2 ) & 1 pass 2 abs_level_gtx_flag[1] ( |C _N | - 2 ) >= 4 abs_level_gtx_flag[2] ( |C _N | - 2 ) >= 6 abs_level_gtx_flag[3] ( |C _N | - 2 ) >= 8 abs_level_gtx_flag[4] ( |C _N | - 2 ) >= 10 pass 3 abs_remainder ( |C _N | - 10 ) >> 1

If the value of C ₀ is -21, sig_coeff_flag corresponding to C ₀ is 1, coeff_sign_flag is 1, abs_level_gtx_flag[0] is 1, par_level_flag is 1, abs_level_gtx_flag[1] is 1, abs_level_gtx_flag[2] is 1, abs_level_gtx_flag[2] is 1 [3] is 1, abs_level_gtx_flag[4] is 1, and the value of abs_remainder is 5. These syntaxes are entropy-encoded and signaled to the decoder.

The decoder restores the residual signal using the following table.

Equation TmpC _N sig_coeff_flag + abs_level_gtx_flag[0] + par_level_flag C _N TmpC _N + (abs_level_gtx_flag[1]<<1) + (abs_level_gtx_flag[2]<<1) + (abs_level_gtx_flag[3]<<1) + (abs_level_gtx_flag[4]<<1) +
(abs_remainder<<1)

As described above, when |C ₀ | is restored, the value of TmpC _N becomes 3 through 1+1+1, (abs_level_gtx_flag[1]<<1) + (abs_level_gtx_flag[2]<<1) + (abs_level_gtx_flag 8 is restored through [3]<<1) + (abs_level_gtx_flag[4]<<1), and 10 is restored through (abs_remainder<<1). After that, the value of C ₀ is finally restored to -21 through the coeff_sign_flag value.

1-3: Truncated Rice (TR) binarization method

In order to perform entropy encoding on a certain value, binarization must be preceded. Among various binarization methods, two parameters are required to perform the truncated rice (TR) binarization method. Specifically, in order to perform the truncated rice binarization method, the cMax parameter and the rice parameter (cRiceParam) must be determined.

When the TR binarization method is applied, a value to be encoded may be classified into a prefix part and a suffix part based on the parameter. Thereafter, each part can be binarized according to a set method. The prefix part is classified by the following equation.

prefixVal = symbolVal >> cRiceParam (1)

In the above formula, symbolVal means the value to be encoded, and prefixVal means the prefix. After that, it is binarized using the following table.

prefixVal Bin string 0 0 One One 0 2 One One 0 3 One One One 0 4 One One One One 0 5 One One One One One 0 ... bin index 0 One 2 3 4 5 ...

At this time, if the value of prefixVal is smaller than the result of (cMax >> cRiceParam), it can be binarized as shown in the table above. Otherwise, the length of the bin string is not increased further, and the last bin is replaced with 1. For example, when the value of (cMax >> cRiceParam) is 6 and the value of prefixVal is 6, an empty string may be set as follows.

prefixVal Bin string 0 0 One One 0 2 One One 0 3 One One One 0 4 One One One One 0 5 One One One One One 0 6 One One One One One One bin index 0 One 2 3 4 5

The suffix can be used only when cMax is greater than symbolVal and cRiceParam is greater than 0. Otherwise, the suffix is not used. When a suffix is used, the suffix value is set by the following equation.

suffixVal = symbolVal - (prefixVal << cRiceParam) (2)

Thereafter, when suffixVal is binarized, a fixed-length (FL) binarization method may be applied. The cMax value for FL binarization is set to a value of (1 << cRiceParam)-1.

Under the fixed-length binarization method, the fixed length can be set as follows.

fixedLength = Ceil ( Log2 (cMax+1) ) (3)

fixedLength means a fixed length used in the FL binarization method, and Ceil( ) means round-up operation. Also, (2 ^fixedLength ) values can be binarized. For example, if fixedLength is set to 2, 2 ² = 4 values can be binarized. The following table is an example accordingly.

Val Bin string 0 0 0 One 0 One 2 One 0 3 One One bin index 0 One

1-4: Limited k-th order Exp-Golomb binarization method

As inputs of this binarization method, symbolVal, which is a value to be binarized, and variables k, maxPreExtLen, and truncSuffixLen exist.

A method for constrained k-order exponential Gollum binarization using these input values is described through the following table.

codeValue = symbolVal >> k
preExtLen = 0
while( ( preExtLen < maxPreExtLen ) && ( codeValue > ( ( 2 << preExtLen ) - 2 ) ) ) {
preExtLen++
put( 1 )
}
if( preExtLen = = maxPreExtLen )
escapeLength = truncSuffixLen
else {
escapeLength = preExtLen + k
put( 0 )
}
symbolVal = symbolVal - ( ( ( 1 << preExtLen ) - 1 ) << k )
while( ( escapeLength-- ) > 0 )
put( ( symbolVal >> escapeLength ) & 1 )

1-5: The binarization method of abs_remainder and dec_abs_level

For the binarization of abs_remainder, abs_remainder is classified into a prefix part and a suffix part. Thereafter, the prefix part is binarized in the TR method, and the suffix part is binarized in the limited k-th order Exp-Golomb method. At this time, only when the result of binarization by the TR method is 111111 (that is, when the value of prefixVal is (cMax >> cRiceParam)), a suffix of abs_remainder is generated and binarized.

First, in order to binarize the prefix of abs_remainder in the TR method, the value of cRiceParam and the value of cMax must be set. In this case, the value of cRiceParam may be determined according to the encoding method of the residual signal. For example, when the TSRC method is applied for encoding the residual signal, cRiceParam may be set to a predefined constant. Here, the predefined constant may be 1. If RC is applied for encoding the residual signal, cRiceParam is set according to the following description.

First, as shown in Figure 3, the absolute value sum (locSumAbs) of the residual signals existing at each of the surrounding reference positions is derived based on the current encoding position.

Figure 3

At this time, the absolute value of each residual signal existing at the reference position is generated using Table 2. After that, clipping is performed through the following Equation (4). Here, Clip3 (A,B,C) means that if the value of C is less than or equal to A, output A, otherwise, if the value of C is greater than B, output B, otherwise output C.

locSumAbs = Clip3(0, 31, locSumAbs - baseLevel*5 ) (4)

In the above formula, baseLevel is fixed to 4. Then, according to the locSumAbs calculated by the equation, cRiceParam is derived as shown in the following table.

locSumAbs 0 One 2 3 4 5 6 7 8 9 10 11 12 13 14 15 cRiceParam 0 0 0 0 0 0 0 One One One One One One One 2 2 locSumAbs 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 cRiceParam 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3

The parameter cMax is set using the following equation.

cMax = 6<< cRiceParam (5)

Using the above-described method, if the value of cRiceParam and the value of cMax are set to binarize the prefix of abs_remainder to the TR method, the prefix of abs_remainder is set using the following Equation (6).

prefixVal = Min(cMax, abs_remainder) (6)

The prefixVal set in this way is binarized in the TR method using the parameters cMax and cRiceParam.

The following example shows an example of binarizing abs_remainder. For convenience of explanation, it is assumed that cRiceParam is 2 and the value of abs_remainder is 23.

First, since cRiceParam is 2, cMax is set to 24 by Equation (5). After that, the prefix value of abs_remainder is set to 23 by Equation (6). TR binarization is performed using this prefix value and cMax and cRiceParam as input values. At this time. This prefix value is input as symbolVal of TR binarization.

To perform TR binarization, symbolVal is classified into prefix and suffix. symbolVal is 23 and prefixVal is set to 5 through Equation (1). After that, since this prefix value is less than 6 which is the result of (cMax >> cRiceParam), it is binarized to 111110 by Table 5.

And, since the cMax value is greater than the symbolVal value of 23 and the cRiceParam is 2, the suffix of symbolVal must be additionally set. When set as in Equation (2), the value of suffixVal is set to 3, which is the result of 23 - (5<<2), and the value of (1 << cRiceParam)-1 is set to cMax for the FL binarization method. Thereafter, binarization is performed by applying the suffixVal value of 3 and the cMax value of 3 as inputs of a fixed-length (FL) binarization method. At this time, the value of fixedLength becomes 2 by Equation (3), and is binarized to 11 according to Table 7.

In the same way as above, when TR binarization is performed on 23, which is the prefix value of abs_remainder, it is expressed as 111110 11 . Therefore, since it is not 111111, the suffix of abs_remainder is not generated and the binarization process of abs_remainder is terminated.

The following example shows another example of binarizing abs_remainder. For convenience of explanation, it is assumed that cRiceParam is 2 and the value of abs_remainder is 25.

First, since cRiceParam is 2, cMax is set to 24 by Equation (5). After that, the prefix value of abs_remainder is set to 24 by Equation (6). TR binarization is performed using this prefix value and cMax and cRiceParam as input values. At this time. This prefix value is input as symbolVal of TR binarization.

To perform TR binarization, symbolVal is classified into prefix and suffix. symbolVal is 24 and prefixVal is set to 6 through Equation (1). Then, since this prefix value is not less than 6 which is the result of (cMax >> cRiceParam), it is binarized to 111111 by Table 6.

And, since the cMax value is not greater than the symbolVal value of 24, the TR binarization suffix is not generated, and the result of binarizing the abs_remainder prefix ends with 111111.

After that, since the result of binarizing the prefix of abs_remainder by the TR method is 111111, the suffix of abs_remainder is generated by the following equation.

SuffixVal = abs_remainder - cMax (7)

Therefore, if Equation (7) is applied, 1, which is the result of 25-24, is generated as a suffix value. After that, the value of k is set as the result of cRiceParam+1, and maxPreExtLen and truncSuffixLen are set to 11 and 15, respectively. Finally, the suffix value, k, maxPreExtLen, and truncSuffixLen are used as input values and binarized in a limited k-th order Exp-Golomb method.

The binarization of dec_abs_level is the same as that of abs_remainder. However, the baseLevel used in Equation (4) is fixed to 0.

2. How to set the conversion factor range

The bitstream is restored through CABAC, inverse quantization, and inverse transformation processes. At this time, after each process, the range of values must exist in -(1<<15) ~ (1<<15)-1. For this purpose, clipping may be performed. The following figure shows an example related to this.

Figure 4

For example, a value decoded through CABAC in a transform block is stored in a TransCoeffLevel buffer having a size equal to that of the transform block. Values in this buffer must always exist between -(1<<15) ~ (1<<15)-1, that is, -32768 ~ 32767. After that, the values present in this buffer are copied to the dz buffer and stored in the dnc buffer through inverse quantization, and then stored in the d buffer through clipping. The following formula represents this.

bdShift1= BitDepth　+　rectNonTsFlag　+ ((　Log2(W)　+　Log2(H)　)　/　2　)　-　5　+　DQflag (8)

dnc[　x　][　y　] = ( 　dz[　x　][　y　]　*ls[　x　][　y　]　+　(1<<(bdShift1 - 1)) )>>

d[　x　][　y　] = Clip3(　CoeffMin,　CoeffMax,　dnc[　x　][　y　]　) (10)

In the above equations, BitDepth means the bit depth of the signal, and rectNonTsFlag is a flag indicating whether the product of the horizontal and vertical lengths of a block is an exponential power of 4. If the product of the width and height of the block is a power of 4, the value of rectNonTsFlag is set to 1, otherwise it is set to 0. W and H denote the horizontal and vertical length of the block, respectively, and DQflag is information indicating whether Dependent Quantization (DQ) is used. If DQ is used, the value of DQflag is set to 1, otherwise it is set to 0. Also, the ls buffer means an inverse quantization matrix. CoeffMin and CoeffMax are set to -32768 and 32767 respectively, as shown in Figure 4. The Clip3 function outputs CoeffMin as a result if the value of dnc[x][y] is less than CoeffMin, and outputs CoeffMax as a result if the value of dnc[x][y] is greater than CoeffMax. If the value of dnc[x][y] exists between CoeffMin and CoeffMax, dnc[x][y] is output.

If LFNST is performed in the encoder, the d buffer is rearranged to 1D, the inverse transform (LFNST) is performed, and the result is stored in the d buffer again. The following formula represents this process.

(10)

(10)

In the above formula, nonZeroSize is set to 8 or 16 depending on the size of the block, and lowFreqTransMatrix means the LFNST transformation matrix. The x buffer is a buffer in which the d buffer, which is the result of Equation (10), is rearranged in 1D. The y buffer represents the output value of LFNST. This y-buffer is reordered in 2D and stored in the d-buffer.

After that, if transform (MTS, etc.) is performed in the encoder, an inverse transform (Ver) is performed on the d buffer and clipping is performed. The following formula represents this.

g[　x　][　y　] = Clip3( CoeffMin, CoeffMax, ( e[　x　][　y　] + 64 ) >> 7 ) (11)

In the above equation, buffer e represents the result of performing inverse transformation (Ver) on buffer d. After clipping the e-buffer and substituting it into the g-buffer, an inverse transform (Hor) is performed on the g-buffer to perform clipping to generate a restored value. The following formula represents this.

bdShift2 = (　H　>　1　　&&　　W　>　1　)　? (　20　-　BitDepth　)　:　(　21　- BitDepth　) (12)

res[　x　][　y　] = ( r[　x　][　y　] + (1<<(bdShift2 - 1)) ) >> bdShift2 (13)

In the above equation, r denotes the result of inverse transformation (Hor) on the g buffer, and W and H denote the horizontal and vertical lengths of the block, respectively. Through this formula, the r buffer is clipped and assigned to the restoration buffer res.

The above description is based on the decoder. In the encoder, the transformation can be performed in the reverse process of that described.

3. How to Set the Extended Transform Coefficient Range

When the sample exceeds 10 bits or a very low QP is used, the range of values and bit precision may be adjusted for efficient encoding/decoding. The following figure shows an example of this.

Figure 5

Unlike Figure 4, in Figure 5, the range of values can be adjusted by adding a variable called Range. The range may be variably determined based on the bit depth. Alternatively, the Range value can be fixed/defined as an arbitrary constant in the encoder and decoder.

As an example, the Range may be determined through the following equation.

Range = MAX (BitDepth+5, 15) (14)

In the above formula, MAX(A,B) outputs the larger value among A and B.

Also, Equation (8) may be changed to the following expression according to the range.

bdShift1= BitDepth　+　rectNonTsFlag　+ ((　Log2(W)　+　Log2(H)　)　/　2　)　+ 10 - Range　+　DQflag (15)

For example, if the bit depth of the image is 10, the value of bdShift1 is set to be the same as in Equation (8).

Also, Equation (12) may be changed to the following expression according to the range.

bdShift2 = (　H>1　　&&　　W>1　)　? (5+Range - BitDepth　)　:　(　6+Range　- BitDepth　) (16)

For example, if the bit depth of the image is 10, the value of bdShift2 is set to be the same as in Equation (12).

As described above, abs_remainder and dec_abs_level perform binarization by classifying them into a prefix part and a suffix part when performing binarization. When the prefix part is binarized, the TR method is applied, and when the suffix part is binarized, the limited k-order exponential Gollum (hereinafter, referred to as EG(k)) method is applied.

When binarization is performed on abs_remainder and dec_abs_level, the sum of the number of bits for the prefix and the suffix may be set to a preset threshold for convenience of hardware implementation. The following table is an example showing the maximum number of bits of a prefix and a suffix according to a preset threshold. Hereinafter, it is assumed that the threshold value is 32 for convenience of description.

TR EG(k) Range Maximum length of prefixVal maxPreExtLen truncSuffixLen threshold 15 6 11 15 32 16 6 10 16 32 17 6 9 17 32 18 6 8 18 32 19 6 7 19 32 20 6 6 20 32 21 6 5 21 32 22 6 4 22 32 ... ... ... ... 32

In Table 10 above, TR denotes a truncated rice (TR) binarization method, and EG(k) denotes a limited k-th order Exp-Golomb binarization method.

The prefixVal expressed in the above table is the value defined in Table 6. Accordingly, a suffix value may exist only when prefixVal exceeds 6. In addition, maxPreExtLen and truncSuffixLen are used in Table 8 for suffix binarization.

The values of maxPreExtLen and truncSuffixLen of Table 10 may be set using the following equation.

truncSuffixLen = Range (16)

maxPreExtLen = threshold - 6 - truncSuffixLen (17)

In this case, as the Range increases, maxPreExtLen decreases, which may be insufficient to sufficiently express the value. Therefore, the truncSuffixLen value may be fixedly used regardless of the Range without using Equation (16). As an example, the value of truncSuffixLen may be fixed to 20. In this case, the value of maxPreExtLen is always fixed to 6 by Equation (17).

Alternatively, by using the relationship of Equation (17), information for determining either maxPreExtLen or truncSuffixLen may be encoded through upper headers such as slice headers and PPSs to be signaled from the encoder to the decoder. Alternatively, a method of encoding the information for each arbitrary region divided for parallel processing such as a tile and signaling the information from the encoder to the decoder is also possible.

Alternatively, the formula (14) can be changed to the following formula (18).

Range = MAX( 15, Min(20, BitDepth+5) ) (18)

In case of using Equation (18), when encoding an image of 10 bits or less, the value of Range is set to 15 as in the conventional method (Figure 4), and only when encoding an image exceeding 10 bits, the value of Range is 15 is changed to a value that exceeds

Alternatively, regardless of the bit depth, information indicating whether or not the bit precision is extended may be signaled from the encoder to the decoder, and the range may be changed according to this information. For example, assuming that the information indicating whether or not the bit precision is extended is ExtPrecFlag, a method of setting a range according to the following equation is also possible.

Range = (ExtPrecFlag == 1) ? R: 15) (19)

In the above equation, it is possible to use a value preset by the encoder and the decoder for the variable R, or a method for signaling from the encoder to the decoder is also possible. In this case, R may be signaled as it is, and the difference between R and 15 is signaled, and the decoder adds 15 to this difference to restore R and use it as a range. After that, the values of maxPreExtLen and truncSuffixLen are set using Equations (16) and (17).

Alternatively, a method of determining the range using both ExtPrecFlag and bit depth is also possible. The following formula shows an example related to this.

Range = (ExtPrecFlag == 1) ? MAX ( 15, MIN( 20, BitDepth+6 )) : 15 (20)

In the above formula, MIN(A,B) outputs the smaller of A and B.

Claims (1)

A method of encoding/decoding a residual signal.