Patent Application
20190014321
The invention concerns a decoder for decoding a sequence of portions of media from a data stream using an unary code, an encoder for encoding a sequence of portions of media into a data stream using an unary code, a method for decoding a sequence of portions of media from a data stream using an unary code, a method for encoding a sequence of portions of media into a data stream using an unary code, wherein in some embodiments asymmetric unary alphabets for most significant bit plane position entropy coding of coefficient groups are used.
Compression of an image or a video sequence typically consists of a frequency transform, followed by quantization and entropy coding. In the latter, coefficients are grouped into coefficient groups, and their most significant non-null bitplane (GCLI) is computed. GCLIs are transmitted to the decoder along with the raw data.
In order to encode the GCLIs, the number of remaining bits is calculated according to a truncation level (GTLI), and predicted from either a vertical or horizontal coefficient group. Finally, a variable-length unary code may be used to construct the final bitstream.
The employed alphabet for this unary code is symmetric with respect to 0, investing the same bits for coding a value X and −X. However, the commonly used unary codes for coding require a large number of bits.
Thus, it is an object of the invention to provide more efficient coders and coding schemes which allow for a bit saving compared to the prior art.
A first aspect of the invention concerns a decoder for decoding a sequence of portions of media from a data stream, wherein the decoder is configured to decode, for a current portion, a signed integer variable from the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs from a second length of a second codeword of the unary code assigned to a second possible value by exactly one. Said second possible value has a second sign and the predetermined absolute value, or said second possible value is zero and the predetermined absolute value is one. The decoder is further configured to derive a bit plane number predictor for the current portion, to derive a bit plane number indicating coded bit-planes by computing a sum of the bit plane number predictor and the signed integer variable, and to decode the current portion from the data stream by use of the bit plane number.
A second aspect of the invention concerns an encoder for encoding a sequence of portions of media into a data stream, wherein the encoder is configured to derive a bit plane number predictor for a current portion. The inventive encoder is further configured to derive a bit plane number indicating coded bit-planes for the current portion and setting a signed integer variable so that the bit plane number is derivable from a sum between the bit plane number predictor and the signed integer variable. The inventive encoder is further configured to encode, for the current portion, the signed integer variable into the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value. Said second possible values has a second sign and the predetermined absolute value, or said second possible value is zero and the predetermined absolute value is one. The inventive encoder is further configured to encode the current portion into the data stream by use of the bit plane number.
A third aspect of the invention concerns a method for decoding a sequence of portions of media from a data stream, wherein the method comprises a step of decoding, for a current portion, a signed integer variable from the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value. Said second possible value has a second sign and the predetermined absolute value, or said second possible value is zero and the predetermined absolute value is one. The method further comprises a step of deriving a bit plane number predictor for the current portion, deriving a bit plane number indicating coded bit-planes by computing a sum of the bit plane number predictor and the signed integer variable, and decoding the current portion from the data stream by use of the bit plane number.
A fourth aspect of the invention concerns a method for encoding a sequence of portions of media into a data stream, wherein the method comprises a step of deriving a bit plane number predictor for a current portion. The method further comprises a step of deriving a bit plane number indicating coded bit-planes for the current portion and setting a signed integer variable so that the bit plane number is derivable from a sum between the bit plane number predictor and the signed integer variable. The method further comprises a step of encoding, for the current portion, the signed integer variable into the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value. Said second possible values has a second sign and the predetermined absolute value, or said second possible value is zero and the predetermined absolute value is one. The method further comprises a step of encoding the current portion into the data stream by use of the bit plane number.
A fifth aspect of the invention concerns a computer readable digital storage medium having stored thereon a computer program having a program code for performing, when running on a computer, a method for decoding a sequence of portions of media from a data stream, wherein the method comprises a step of decoding, for a current portion, a signed integer variable from the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value. Said second possible value has a second sign and the predetermined absolute value, or said second possible value is zero and the predetermined absolute value is one. The method further comprises a step of deriving a bit plane number predictor for the current portion, deriving a bit plane number indicating coded bit-planes by computing a sum of the bit plane number predictor and the signed integer variable, and decoding the current portion from the data stream by use of the bit plane number.
A sixth aspect of the invention concerns a computer readable digital storage medium having stored thereon a computer program having a program code for performing, when running on a computer, a method for encoding a sequence of portions of media into a data stream, wherein the method comprises a step of deriving a bit plane number predictor for a current portion. The method further comprises a step of deriving a bit plane number indicating coded bit-planes for the current portion and setting a signed integer variable so that the bit plane number is derivable from a sum between the bit plane number predictor and the signed integer variable. The method further comprises a step of encoding, for the current portion, the signed integer variable into the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value. Said second possible values has a second sign and the predetermined absolute value, or said second possible value is zero and the predetermined absolute value is one. The method further comprises a step of encoding the current portion into the data stream by use of the bit plane number.
Further advantageous embodiments are defined in the dependent claims.
Embodiments of the present invention will be discussed subsequently referring to the enclosed drawings, wherein:
In the following, embodiments of the invention will be described in further detail. It is to be pointed out that the same or functionally equal elements are given the same reference numbers in the figures and that a repeated description for elements provided with the same reference numbers is omitted. Hence, descriptions provided for elements having the same reference numbers are mutually exchangeable. Furthermore, when referring to both, encoding and decoding, the abbreviation coding is used. Thus, coding refers to both encoding and decoding.
Before the inventive principle will be described in more detail, the following passages provide an introduction to coding schemes and data structures that may be used by the inventive coders and methods for coding.
Image Transform
Image and video compression typically applies a transform before running entropy coding. Reference [5], for instance, uses a block based prediction, while references [1][2][3][4] advocate for wavelet transforms.
Such a wavelet transform is depicted in
The wavelet transform decomposes an image 1000 into a number of subbands 1001, 1002. As depicted in
After the frequency transform, the coefficients of the subbands 1001, 1002 are entropy coded. In other words, g>1 coefficients of a subband ABm, with A,B∈{L,H}, m∈N, are arranged into a coefficient group. Then the most significant non-zero bit plane of the coefficient group is signaled, followed by the raw data bits. More details on the coding technique are explained in the following section.
These coefficients 2000, 2001, 2002, 2003 are represented in sign-magnitude representation. The coefficients 2000, 2001, 2002, 2003 comprise bit planes 2020. The largest coefficient in the group 2010 determines the number of active bit planes 2021.
A bit plane 2020 is called active, if at least one coefficient bit of the bit plane 2020 itself or any higher bit plane (bit plane representing a larger number) is unequal to zero. The number of active bit planes 2021 is given by the so-called GCLI value (Greatest Coded Line Index) 2030. A GCLI value 2030 of zero means that no bit planes 2020 are active, and hence the complete coefficient group 2010 is zero. In order to achieve compression, only the active bit planes 2021 are placed into the bit stream.
For lossy encoding, some of the bit planes 2020 need possibly to be truncated, such that the number of bit planes 2020 transmitted for a coefficient group 2010 is smaller than the GCLI value 2030. This truncation is specified by the so-called GTLI (Greatest Trimmed Line Index) 2040.
A GTLI 2040 of zero corresponds to no truncation. A GTLI value 2040 of one means that the number of transmitted bit planes for a coefficient group 2010 is one less than the GCLI value 2030. In other words, the GTLI 2040 defines the smallest bit plane position that is included in the bit stream.
In case of a simple deadzone quantization scheme, the transmitted bit planes equal the bit planes of the coefficient group 2010 without the truncated bit planes 2022. In case of more advanced quantization schemes, some information of the truncated bit planes 2022 can be “pushed” into the transmitted bit planes by modifying the quantization bins. More details can be found in [6].
Since for each coefficient 2000, 2001, 2002, 2003 the number of remaining bit planes 2023 equals the difference between the GCLI 2030 and the GTLI values 2040, it gets obvious that coefficient groups 2010 whose GCLI 2030 is smaller or equal to the GTLI value 2040 may not be contained in the bit stream.
The active bit planes remaining after truncation and quantization are called remaining bit planes 2023 in the following. Moreover, the GTLI 2040 is also called truncation point in the following.
These remaining bit planes 2023 are then transmitted as raw bits to the decoder. In order to enable correct decoding, the decoder needs to know the GCLI value 2030 of every coefficient group 2010. Together with the GTLI value 2040, which is also signaled to the decoder, the decoder can infer the number of raw data bit planes that are in the bit stream.
The GCLI values 2030′ of the second group 2010′ may be signaled by a variable length code (VLC) that represents the difference to the GCLI value 2030 of the previous first coefficient group 2010. This previous coefficient group 2010 can in principle be any coefficient group that the encoder has already encoded before. Hence, it can for instance be a horizontal or a vertical neighbor group.
The output from the prediction is the difference in the number of remaining bit planes 2023, 2023′ between two coefficient groups 2010, 2010′ leading to a delta remaining bit planes 2060. More details on this aspect can be found in the section below called “Prediction Schemes”.
GCLI values 2030 being below the GTLI value 2040 may be of less or even no interest, since the coefficients may not be contained in the bit stream. Consequently, the prediction may be performed in such a way that the decoder can infer whether the GCLI 2030 is greater than the GTLI 2040, and if so, what is the value of the GCLI 2030.
The method described herein is agnostic to the transmission order of the different bit stream parts. For instance, it is possible to first place the GCLI coefficients of all subbands into the bit stream, followed by the data bits of all subbands. Alternatively, GCLI and data bits might be interleaved.
The coefficients of the frequency transform as previously described with reference to
In order to enable the decoder to recover the signal, it needs to know the GCLI value 2030 for every coefficient group 2010. Different methods are available in the state of the art [2] to signal them efficiently.
In the RAW mode, the GCLI value 2030 is transmitted without any prediction. Hence, let F1 be the coefficient group 2010′ to be encoded next. Then the GCLI value 2030′ can be encoded by a fixed length codeword representing the value:
max(GCLI(F1)−GTLI(F1),0)
In a horizontal prediction, the symbol coded is the difference between the GCLI value 2030′ and the value of the GCLI 2030 previously coded belonging to the same line and the same wavelet subband, and considering the GTLI 2040. This difference value is called residual or 5 value in the following.
Let F1 and F2 be two horizontally neighbored coefficient groups 2010, 2010′, consisting of g<1 coefficients. Let F2 be the coefficient group 2010′ to be currently coded. Then GCLI(F2) can be signaled to the decoder by transmitting a residual calculated as follows:
The decoder recovers GCLI(F2) by computing
For instance in horizontal prediction, typically GCLI(F1)=GCLI(F2). Furthermore, δ may be transmitted as a variable length code, as described in the section below called “Unary Coding”.
In a vertical prediction between two subband lines, the result is the difference between the current GCLI value and the GCLI value of the same subset of coefficients in the previously coded line.
Let F1 and F2 be two vertically neighbored coefficient groups 2010, 2010′, consisting of g>1 coefficients. Let F2 be the coefficient group 2010′ to be currently coded. Then, GCLI(F2) can be encoded in the same way than in the horizontal prediction described above.
In addition to the prediction modes, different coding modes can be used. In the following, coding modes are simply considered as additional prediction modes for reasons of simplicity.
Reference [6] proposes a method to compress zero GCLIs 2030 more efficiently. To this end, consecutive coefficient groups 2010, 2010′ may be arranged into GCLI groups. All GCLI groups are of the same size (e.g. eight). Then, a single bit flag indicates if the corresponding deltas remaining bit planes 2060 of the complete GCLI group are all zero. Zero GCLI groups may not further be encoded, while non-zero GCLI groups may be encoded as described in the section above called “Prediction Schemes”.
Reference [7] proposes an intrasubband alternative to efficiently compress zero GCLIs 2030. In this method, the decoder is informed that a group (run) of (zero) GCLI values is not contained in the bit stream by signaling a GCLI value that is not used for ordinary coding. For instance, GCLI values 2030 below the GTLI value 2040 are not used for normal coding, since setting GCLI=GTLI already defines a zero coefficient. Hence, signaling a GCLI value 2030 equaling GTLI−1 can also be used to inform the decoder about skipped GCLI coefficients. The GCLI values 2030 used to signal GCLI skipping are called escape GCLI codes in the following.
GCLI zero-run coding uses this concept to skip runs of consecutive zero remaining bit planes values within a subband line. A run happens when (see
Assumed that the encoding of the GCLI coefficients 4001 to 4008 is done sequentially (e.g. in
Reference [8] proposes an intersubband alternative to efficiently compress zero GCLIs. Similarly to spatial zero-run coding (see preceding section above), an escape GCLI code is used to signal the absence of groups of GCLIs in the codestream.
GCLI zero-tree coding, as shown in
As explained above,
Within the precinct, a zero-tree 5000 can be established. The root of this zero-tree is one coefficient group 5001 of the LL subband. For every other subband, the zero-tree contains a well-defined number of coefficient groups, whose number depend on the decomposition level of the subband. For the lowest decomposition level (Level 0), only one GCLI coefficient group is contained, while for higher subbands (e.g. Levels 1 to 9), more coefficient groups are contained. Let
Then the number of nodes from subband s equals 2D
Those coefficient groups are then brought into a parent-child relation as depicted in
The LL coefficient 5001 forms the root of the tree 5000. Since the codec usually uses more horizontal than vertical decompositions, this root node 5001 only has one child node 5002. All further nodes have two children nodes, when only a horizontal decomposition is performed. For the first vertical and horizontal decomposition, the node 5004 has six children nodes. For all consecutive decompositions, a node 5005 has four children nodes belonging to the same subband.
Let's suppose now that the GCLI value of all shaded nodes 5010, 5011, 5012, 5013 in
It is noteworthy that, when working with image dimensions not power of two, the zero trees constructed in the boundaries can be incomplete (e.g. nodes in subband six can have only two children from subband 9, instead of four). In such cases, the zero trees are considered only when they incur a rate saving for signaling the tree, against signaling the delta remaining bits.
As mentioned before, the delta values 2060 calculated with the selected prediction mode (horizontal or vertical) may be transmitted using a variable length unary code, as described in Table 1 of [2].
It can be seen that the unary code is composed of the following types of bits:
The state of the art unary code as shown in
In this invention, four alternative alphabets for the unary code in GCLI coding are proposed. They aim at improving coding efficiency by reducing the average required budget in GCLI coding. The forthcoming subsections give a description of these prefix-free alphabets.
Note that the descriptions are given for values from −15 to 15, and using at most 16 bits. However, they can be easily extended to higher (lower) values by going beyond 16 bits. This can be achieved by replacing those codes of length 16 by others that continue adding l's in the left.
Moreover, there are different scenarios in which these proposed alphabets can be used. On the one hand, they can be used as an option depending on the targeted bits per pixel (bpp), specifically for low bpp. On the other hand, the choice of whether to choose an alphabet for GCLI coding can be also made per precinct, or even per subband. This choice is done based on which alphabet requires the smaller budget to encode the precinct or subband.
In particular, this invention describes a set of asymmetric alphabets for the unary code of GCLI residuals. This method leads to increased compression efficiency. These alphabets can be categorized as follows:
These categories of alphabets will be described in detail in the following. Since these alphabets may be used for efficient coding, and in particular efficient GCLI coding, the inventive coders (i.e. encoder and decoder) shall be described first. The coders may be described for GCLI coding by way of example, wherein other coding schemes may also be used. Thus, reference will be made to
The encoder 70 may be configured to derive a bit plane number predictor for a current portion 71. The encoder 70 may do so by, e.g. predictive coding. For example, if e.g. GCLI coding may be used, the bit plane number predictor may be a GCLI value 2030 of a previously coded coefficient group 2010, assumed that the coefficient group to be currently coded may be the second coefficient group 2010′ when taking
The encoder 70 may further be configured to derive a bit plane number indicating coded bit-planes for the current portion 71. The encoder 70 may do so by, e.g. predictive coding. For example, if GCLI coding is used, a bit plane number may, for instance, be represented by a GCLI value 2030′ indicating the coded, i.e. active, bit planes.
The encoder 70 may further be configured to set a signed integer variable 73 so that the bit plane number 2030′ is derivable from a sum between the bit plane number predictor 2030 and the signed integer variable 73. Said signed integer variable 73 may represent in an encoded representation the delta remaining bit planes 2060, also referred to as δ-value, if e.g. GCLI coding is used.
The encoder 70 may further be configured to encode, for the current portion 71, the signed integer variable 73 into the data stream 72 by use of an inventive unary code, which will be described in detail below. Said unary code may be used at least for the values around zero.
A first example for an inventive unary code is shown in the table depicted in
The signed integer variable 90 may correspond to the integer valued variable 73 which has been previously described with reference to the inventive encoder 70 depicted in
The codewords 91 are assigned to the possible values in a manner so that a first possible value 92 having a first sign and a predetermined absolute value (e.g. ‘+1’) has assigned a first codeword (e.g. ‘10’) of the unary code of a first length (e.g. two bit) which differs from a second length (e.g. three bit) of a second codeword of the unary code assigned to a second possible value 93a by exactly one, wherein the second possible value 93a has a second sign and the predetermined absolute value (e.g. ‘−1’).
Alternatively, the first possible value 92 having the first sign and the predetermined absolute value (e.g. ‘+1’) having assigned the first codeword (e.g. ‘10’) of the unary code of the first length (e.g. two bit) differs from a second length (e.g. one bit) of a second codeword of the unary code assigned to a second possible value 93b by exactly one, wherein the second possible value 93b is zero and the predetermined absolute value of the first value 92 is one.
In other words, the unary code codewords may be assigned in non-magnitude-plus-sign manner sequentially by traversing the possible values alternatingly 0→+/−→−/+1→4+/−2→−/+2 and so on.
Referring back to
The inventive encoder 70 may do so in that merely the relevant bits of the coefficients of the respective portion (group) are read from the bitstream or, alternatively speaking, the bit plane number may be used to determine the coded bit planes and the bits of coded bit planes may be read from the data stream (with setting the other bits to default values such as zero with respect to bits of more significant bit planes, and, optionally, zero or random values with respect to bits of lower significant bit planes).
The invention also concerns a decoder for decoding the above mentioned data stream.
The decoder 80 may be configured to decode, for a current portion 71, a signed integer variable 73 from the data stream 72 by use of a unary code comprising a series of codewords of increasing length. As explained above with respect to the encoder 70, said signed integer variable may be a delta value indicating the delta remaining bit planes if, e.g. GCLI coding may be used. Furthermore, the unary code may be used at least for the values around zero.
One exemplary unary code shall again be described with reference to
Alternatively, the first length (e.g. two bits) of the first codeword (e.g. ‘10’) of the unary code differs by exactly one from a second length (e.g. one bit) of a second codeword (e.g. ‘0’) of the unary code assigned to a second possible value 93b by exactly one, wherein the second possible value 93b is zero and the predetermined absolute value is one.
In other words, the unary code codewords may be assigned in non-magnitude-plus-sign manner sequentially by traversing the possible values alternatingly 0→+/−1→−/+1→+/−2→−/+2 and so on.
The inventive decoder 80 may further be configured to derive a bit plane number predictor for the current portion 71, and to derive a bit plane number (e.g. a GCLI value) indicating coded bit-planes by computing a sum of the bit plane number predictor and the signed integer variable.
Furthermore, the inventive decoder 80 may be configured to decode the current portion 71 from the data stream 72 by use of the bit plane number (e.g. GCLI value).
The inventive decoder 80 may do so in that merely the relevant bits of the coefficients of the respective portion (group) are read from the bitstream or, alternatively speaking, the bit plane number may be used to determine the coded bit planes and the bits of coded bit planes may be read from the data stream (with setting the other bits to default values such as zero with respect to bits of more significant bit planes, and, optionally, zero or random values with respect to bits of lower significant bit planes).
According to a further embodiment, the sequence of portions 71 of the media may be blocks of coefficients of a spectral decomposition of a picture, as described with reference to
According to yet a further embodiment, the spectral decomposition may be a wavelet transform or a block-wise spectral transform based on trigonometric basis functions.
According to yet a further embodiment, the portions 71 may be blocks of spatially neighboring coefficients within subbands into which the picture is decomposed by the spectral transform.
Referring back to
Furthermore, the inventive decoder 80 may be configured to derive a bit plane number (e.g. a GCLI value) indicating coded bit-planes by computing a sum of the bit plane number predictor (e.g. GCLI value of previously decoded coefficient group) and the signed integer variable, i.e. the above mentioned delta value 90.
Furthermore, the inventive decoder 80 may be configured to decode the current portion 71 from the data stream 72 by use of the bit plane number (e.g. GCLI value).
In the following, several examples of alphabets comprising unary codes that may be used by the inventive coders 70, 80 in the above described manner shall be introduced and discussed below.
A first set of alphabets comprising unary codes will be exemplarily described below with reference to the following four different types of full-range alphabets for unary coding of GCLI values.
The aim of the alphabet shown in
The aim of this second alphabet shown in
The aim of this third alphabet is to save one bit to code the value ‘+1’ with respect to the state of the art (
The aim of this alphabet is to save one bit to code the value ‘−1’ with respect to the state of the art (
Similarly to the alphabets defined in previous Section A), which are defined for a whole range of values for GCLIs between −15 to 15, it is also possible to define alphabets with a clipped (smaller) range. The clipped variants of each alphabet defined above are depicted in
However, clipped alphabets may work as an option to be chosen per precinct or per subband, being as an alternative the alphabet defined in the state of the art (
Moreover, since the choice is made per precinct/subband, there is no need to implement a global switch depending on the targeted bpp.
Such clipped alphabtes, thus, enable a compression gain in case of being offered in addition to a full-range alphabet as a fallback alphabet. Another advantage of a clipped alphabet is that it doesn't require an special set of codewords to complete the range, what is hard for parallel decoders.
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘+1’ is coded with two bits and ‘−1’ is coded with three bits. Remember, the unary codes of the state of the art, as shown in
Furthermore, the values ‘+15’ and ‘−15’ are clipped.
The aim of this first clipped signed alphabet #1 is to save one bit to code the value 1 with respect to the state of the art (
This unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘−1’ is coded with two bits and ‘+1’ is coded with three bits. Furthermore, the values ‘+15’ and ‘−15’ are clipped.
The aim of this second clipped alphabet #2 is to save one bit to code the value ‘−1’ with respect to the state of the art (
This unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘+1’ is coded with two bits, ‘−1’ is coded with three bits, ‘+2’ is coded with four bits and ‘−2’ is coded with five bits. Furthermore, the values ‘+14’, ‘−14’, ‘+15’ and ‘−15’ are clipped.
The aim of this third clipped alphabet #3 is to save one bit to code the value ‘+’1 with respect to the state of the art (
This unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘−1’ is coded with two bits, ‘+1’ is coded with three bits, ‘−2’ is coded with four bits and ‘+2’ is coded with five bits. Furthermore, the values ‘+14’, ‘−14’, ‘+15’ and ‘−15’ are clipped.
The aim of this fourth clipped alphabet #4 is to save one bit to code the value ‘−1’ with respect to the state of the art (
Similarly to the alphabets defined in previous Section A), which are defined for a whole range of values for GCLIs between −15 to 15, it is also possible to define alphabets with a clipped (smaller) range, as shown in Section B).
However, in some scenarios, the sign bits corresponding to the unary codes are signaled separately in the code stream. Hence, the unary code for the clipped alphabet can harness this situation in order to expand the range of supported values.
In this way, the alphabets #1 (c.f.
Therefore, similarly to the clipped alphabets introduced in Section B), the clipped unsigned alphabets #3 (c.f.
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘+1’ is coded with two bits and ‘−1’ is coded with three bits. Remember, the unary codes of the state of the art, as shown in
The aim of this first unsigned full-range alphabet is to save one bit to code the value ‘+1’ with respect to the state of the art (
The sign bit must be signaled separately at least for the values such that ‘value’>1.
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘−1’ is coded with two bits and ‘+1’ is coded with three bits. Remember, the unary codes of the state of the art, as shown in
The aim of this second unsigned full-range alphabet is to save one bit to code the value ‘−1’ with respect to the state of the art (
The sign bit must be signaled separately at least for the values such that ‘value’>1.
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘+1’ is coded with two bits, ‘−1’ is coded with three bits, ‘+2’ is coded with four bits and ‘−2’ is coded with five bits. Remember, the unary codes of the state of the art, as shown in
Furthermore, the values ‘+15’ and ‘−15’ are clipped.
The aim of this third alphabet is to save one bit to code the value ‘+1’ with respect to the state of the art (
The sign bit must be signaled separately at least for the values such that ‘value’>2.
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘−1’ is coded with two bits, ‘+1’ is coded with three bits, ‘−2’ is coded with four bits and ‘+2’ is coded with five bits. Remember, the unary codes of the state of the art, as shown in
Furthermore, the values ‘+15’ and ‘−15’ are clipped.
The aim of this fourth alphabet is to save one bit to code the value ‘−1’ with respect to the state of the art (
The sign bit must be signaled separately at least for the values such that ‘value’>2.
Similarly to the alphabets defined in previous Section C), which are defined for a whole range of values for GCLIs between −15 to 15, and employing at maximum 16 bits for a unary code, it is also possible to define alphabets with an extended amount of bits for the same range.
The extended variants of each alphabet defined above are depicted in
As with the full-range alphabets defined in section A) above, these extended alphabets of this section D) can be employed as alternatives to the state of the art (
It can be also considered as a choice depending on the targeted bpp, by the same means explained before.
Boundaries can end in different ways, now that the number of bits is not a restriction. In
Alphabets with such an enforced stop bit at boundaries are easier to decode, in general, and can be easily extended to larger ranges beyond 15/−15. Moreover, the state of the art alphabet shown in
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘+1’ is coded with two bits and ‘−1’ is coded with three bits. Remember, the unary codes of the state of the art, as shown in
Furthermore, compared with the regular (i.e. non-extended) full range alphabet #1 in section A), the extended full-range alphabet #1 of
The aim of this first extended full-range alphabet #1 is to save one bit to code the value ‘+1’ with respect to the state of the art (
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘−1’ is coded with two bits and ‘+1’ is coded with three bits. Remember, the unary codes of the state of the art, as shown in
Furthermore, compared with the regular (i.e. non-extended) full range alphabet #2 in section A), the extended full-range alphabet #2 of
The aim of this second extended full-range alphabet #2 is to save one bit to code the value ‘-1’ with respect to the state of the art (
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘+1’ is coded with two bits, ‘−1’ is coded with three bits, ‘+2’ is coded with four bits and ‘−2’ is coded with five bits. Remember, the unary codes of the state of the art, as shown in
The aim of this third extended full-range alphabet #3 is to save one bit to code the value ‘+1’ with respect to the state of the art (
The unary code may be an asymmetric unary code. That is, ‘0’ is coded with one bit, ‘−1’ is coded with two bits, ‘+1’ is coded with three bits, ‘−2’ is coded with four bits and ‘+2’ is coded with five bits. Remember, the unary codes of the state of the art, as shown in
The aim of this fourth extended full-range alphabet #4 is to save one bit to code the value ‘−1’ with respect to the state of the art (
Alphabets for Unary Coding (e.g. of GCLI Values) with Different Ranges or Different Length Restrictions
The introduced alphabets above (full-range, clipped (signed and unsigned) and extended full-range alphabets #1, #2, #3 and #4) are defined for a range of values for GCLIs between −15 and 15 and a maximum of 16 bits as a target. Note that the same principles can be applied to the alphabets when:
The usage of alphabets where negative values require less bits than positive values (
Specifically in subband boundaries, zero-runs and also zero-trees (when image dimensions are not power of 2) have a more relaxed condition for usage so that smaller groups can be also considered for the runs. To this end,
Depending on the alphabet and the prediction reference, it varies between zero and three bits. However, by selecting the appropriate alphabet, the number of additional bits can be minimized for those prediction reference values that are occurring more frequently.
Moreover, obviously an escape GCLI code should only be emitted when the savings from the omitted coefficients exceeds the number of additional bits required for the escape GCLI.
Thus, a lower number of additional bits for reference values with higher probability of occurrence allow more frequent use of zero runs with smaller size. Such small-sized zero runs occur at subband boundaries as described in the section above called “spatial zero-run coding”.
Similar reason is valid for a zero-tree, when the number of children nodes is smaller than usual as explained in the section above called “zero-tree coding” for image dimensions not being powers of two. Then, by making the number of additional bits for an escape GCLI code as small as one for the frequently occurring prediction reference values, usage of a zero-tree either improves coding efficiency or does not change it. But, it cannot get worse for these frequently occurring reference values.
Now that examples of alphabets with unary codes have been introduced, which alphabets may be used for coding by the inventive coders 70, 80, further embodiments of the coders 70, 80 will be described in the following.
The embodiments will be exemplarily described for the decoder 80. However, everything that will be described with respect to the decoder 80 also holds true for the inventive encoder 70, except where explicitly stated otherwise. Furthermore, whenever the term coding may be used, this term may refer to both encoding and decoding.
According to an embodiment, the inventive coders 70, 80 may be configured to code the signed integer variable 90 from/into the data stream 72 by use of one of the above described unary codes if the integer valued variable 90 falls within a first subinterval 94 of a value range of the signed integer variable 90. As exemplarily described with reference to the unary code depicted in
Furthermore, the inventive coders 70, 80 of this embodiment may be configured to code the signed integer variable 90, if the signed integer variable 90 falls within a second subinterval 95 of the value range of the signed integer variable 90, from the data stream by coding magnitude and sign of the signed integer variable 90 separately, wherein the magnitude is coded by use of a further unary code, and separately coding a sign bit, wherein none of the further unary code's codewords used in the second interval 95 is a prefix of any codeword of the unary code used in the first interval 94, and vice versa.
As can be seen, the first interval 94 may comprise an asymmetric unary code (as described in sections A to D above) while the second interval 95 may comprise a symmetric unary code.
According to yet a further embodiment, the inventive coders 70, 80 may be configured to code the magnitude and sign of the signed integer variable 90 from/into contiguous bits of the data stream 72 or to code the magnitude and sign of the signed integer variable 90 from/into different fragments of the data stream 72 a first one of which is related to the magnitude, and a second one of which is related to the sign.
In other words, the magnitude bits may be coded from/into a first data stream, while the sign bit(s) may be coded from/into a second data stream. That is, the sign may be either coded in the codeword or coded separately.
Referring back to
According to yet another embodiment, the inventive coders 70, 80 may be configured to code the signed integer variable 90, if the signed integer variable 90 falls within a third subinterval 96 of the value range of the signed integer variable 90, from/into the data stream by decoding magnitude and sign of the signed integer variable 90 separately, wherein the magnitude is coded by use of a fixed length code prefixed with a codeword of which neither one of the codewords of the unary code 94 and the further unary code 95 is a prefix, followed by decoding a sign bit, wherein the first subinterval 94 is contiguous and comprises zero, the second subinterval 95 flanks the first subinterval 94 on both sides of the first subinterval 94, and the third subinterval 96 flanks the second subinterval 95 on both outer sides of the second subinterval 95.
In the example shown in
Of course, the variable length code (VLC) alphabet of
According to embodiments, the vlc alphabets may differ in cardinality of a set of possible values of the signed integer variable 90 and/or in maximum codeword length. The cardinality is the number of elements. For example, the unary code shown in
According to embodiments, the inventive decoder 80 may be configured to derive from the data stream 72 for each of different sections of the spectral decomposition a hint as to which of a set of variable length code (VLC) alphabets to use for the decoding the signed integer variable 90 of portions within the respective section, with at least one of the VLC alphabets involving the unary code.
According to corresponding embodiments, the inventive encoder 70 may be configured to provide in the data stream 72 for each of different sections of the spectral decomposition a hint which of a set of variable length code (VLC) alphabets is used for encoding the signed integer variable 90 of portions within the respective section, wherein at least one of the VLC alphabets involves the unary code.
According to embodiments, the above mentioned different sections may correspond to different subbands into which the picture is subdivided by the spectral decomposition, as explained further above.
In other words, the inventive coders 70, 80 may be configured to switch between alphabets in order to limit a maximum codeword length, e.g. compare
According to yet further embodiments, the unary code may define a subset of a VLC alphabet used for coding the signed integer variable 90 from/into the data stream 72, wherein all VLC alphabet's codewords other than the codewords of the unary code may comprise a concatenation of a unary codeword ending with zero, followed by a sign bit. This may correspond to an enforcement of a stop-bit for easier decoding.
As described above, spatial prediction may be used for coding GCLI residuals. Thus, according to embodiments, the inventive coders 70, 80 may be configured to derive the bit plane number predictor for the current portion 71 by using spatial prediction.
According to embodiments, the inventive coders 70, 80 may be configured to initiate a so-called escape coding mechanism with respect to subsequent portions (i.e. consecutive with respect to the coding order) of the media depending on the signed integer value 90.
An example of an escape coding mechanism may be a zero run coding, wherein coefficients of predetermined portions (e.g. coefficient groups) are set to zero (or some noise is synthesized therein) and coding a signed integer variable 90 is skipped for such portions.
For performing escape encoding, the inventive encoder 70 may comprise a determination unit 74 (c.f.
For performing escape decoding, the inventive decoder 80 may comprise a decoding unit 84 (c.f.
According to embodiments, the inventive coders 70, 80 may be configured to initiate the escape coding mechanism with respect to subsequent portions of the media if the bit plane number derived is negative, and the series of codewords of increasing length are sequentially assigned to a set of possible values {0, . . . ,±a} of the signed integer variable 90 such that the codewords' length assigned to the set increases from possible value 0 to possible value ‘a’ according to 0, −1, 1, . . . , −a, +a with ‘a’ being an integer number greater than 0.
In other words, shorter codewords may be used for spatial zero-run coding.
According to yet further such embodiments, the inventive coders 70, 80 may be configured to skip coding the signed integer variable 90 from/into the data stream 72 for those portions for which the escape coding mechanism is initiated.
According to yet further such embodiments, the inventive coders 70, 80 may be configured to, in coding the current portion 71 from/into the data stream 72 by use of the bit plane number, code bits of bit planes with respect to the current portion from/into the data stream 72, which are indicated by the bit plane number, and skip coding the bits of the bit planes from/into the data stream 72 for those portions for which the escape coding mechanism is initiated.
Generally, the prediction may be performed as a horizontal prediction or as a vertical prediction.
According to embodiments, the inventive coders 70, 80 may be configured to derive the bit plane number predictor for the current portion using spatial prediction along a prediction direction coinciding with a coding order along which the sequence of portions are ordered. Thus, this corresponds to a horizontal prediction.
According to a further such embodiment, the coding order traverses the portions row by row.
According to further embodiments, the inventive coders 70, 80 may, additionally or alternatively, be configured to derive the bit plane number predictor for the current portion 71 by using spatial prediction along a prediction direction, wherein the prediction direction is a column direction and a coding order along which the sequence of portions are ordered traverses the portions row by row. This corresponds to a vertical prediction.
The inventive coders 70, 80 may also be configured to use both the horizontal and the vertical predictions selectively. That is, the coders 70, 80 may be configured to selectively choose which one of the horizontal and the vertical prediction may be best suited for coding the current portions.
According to such embodiments, the inventive coders 70, 80 may be configured to derive the bit plane number predictor for the portions using spatial prediction along a first prediction direction (horizontal) coinciding with a coding order along which the sequence of portions are ordered or along a second prediction direction (vertical), wherein the second prediction direction is a column direction and the coding order traverses the portions row by row, wherein the coder 70, 80 decides, on a subband by subband basis, depending on the data stream 72, to use the first prediction direction or the second prediction direction for portions within the subbands.
According to yet further embodiments, the inventive coders 70, 80 may be configured to restrict the initiating the escape coding mechanism with respect to subsequent portions of the media depending on the signed integer value 90 to a proper subset of the portions.
In other words, the signed integer value 90 may represent a so-called escape GCLI code indicating towards the coders 70, 80 to use the escape coding mechanism, e.g. to skip coding of certain subsequent portions of the media (i.e. a predetermined number of subsequent coefficient groups) from/into the data stream 72. Said proper subset of portions of media may, for instance be, a predetermined number of subsequent coefficient groups that have no active bit planes.
According to yet further embodiments, the inventive coders 70, 80 may be configured so that said proper subset of the portions is distanced along a coding order along which the sequence of portions are ordered, by a predetermined number of portions coinciding, in number, with the subsequent portions with respect to which the escape coding mechanism is initiated. In other words, the coders 70, 80 may use the escape coding mechanism for coefficient groups of a predetermined run-length. That is, if a number of consecutive coefficient groups having no active bit planes is equal or larger than the predetermined run-length, then these coefficient groups are part of a run of the predetermined run-length, and coding of these coefficient groups may be skipped. A run may start at a selected one coefficient group which may also be referred to as a run head.
According to yet further embodiments, the inventive coder 70, 80 may be configured to derive the bit plane number predictor for the current portion using spatial prediction along a prediction direction (e.g. vertical prediction direction), wherein the prediction direction is a column direction and a coding order along which the sequence of portions are ordered traverses the portions row by row, and wherein the proper subsets of the portions form mutually vertically aligned portions, wherein the coders 70, 80 may be configured to perform the coding of the signed integer variable 90, deriving the bit plane number predictor and deriving the bit plane number with respect to the proper subset of portions at a first fragment of the data stream which precedes a second fragment of the data stream at which the coder performs the above mentioned actions of decoding and deriving, in particular for portions external to and subsequent (in terms of the coding order) to said proper subset of portions.
In other words, the GCLIs of run heads are arranged first in the data stream, followed by GCLIs of non-run heads. This is exemplarily depicted in
According to yet further embodiments, the inventive coders 70, 80 may be configured to perform the coding of the signed integer variable 90, deriving the bit plane number predictor and deriving the bit plane number with respect to the sequence of portions at a bit plane number indication fragment 262 of the data stream 72 which precedes a bit plane indication fragment 263 of the data stream 72 at which the coder 70, 80 performs the coding of the portions of media.
In other words, the GCLIs of the coefficient groups are arranged first in the data stream 72, followed by data contained in the bit planes of the respective coefficient groups. This is exemplarily depicted in
Further embodiments of the invention concern a method for encoding and a method for decoding a sequence of portions of media into/from a data stream.
In block 271a it is decoded, for a current portion, a signed integer variable from the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value having a second sign and the predetermined absolute value.
Alternatively, in block 271b, it is decoded, for a current portion, a signed integer variable from the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value being zero with the predetermined absolute value being one.
In block 272, a bit plane number predictor for the current portion is derived.
In block 273, a bit plane number is derived, the bit plane number indicating coded bit-planes by computing a sum of the bit plane number predictor and the signed integer variable.
In block 274, the current portion is decoded from the data stream by use of the bit plane number.
In block 281, a bit plane number predictor for a current portion is derived.
In block 282, a bit plane number is derived, the bit plane number indicating coded bit-planes for the current portion and setting a signed integer variable so that the bit plane number is derivable from a sum between the bit plane number predictor and the signed integer variable.
In block 283a it is encoded, for the current portion, the signed integer variable into the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value having a second sign and the predetermined absolute value.
In block 283b, alternatively, it is encoded, for the current portion, the signed integer variable into the data stream by use of a unary code comprising a series of codewords of increasing length which are sequentially assigned to possible values of the signed integer variable in a manner so that a first possible value having a first sign and a predetermined absolute value has assigned a first codeword of the unary code of a first length which differs by exactly one from a second length of a second codeword of the unary code assigned to a second possible value being zero with the predetermined absolute value being one.
In block 284, the current portion is encoded into the data stream by use of the bit plane number.
Although some aspects have been described in the context of an apparatus, it is clear that these aspects also represent a description of the corresponding method, where a block or device corresponds to a method step or a feature of a method step. Analogously, aspects described in the context of a method step also represent a description of a corresponding block or item or feature of a corresponding apparatus. Some or all of the method steps may be executed by (or using) a hardware apparatus, like for example, a microprocessor, a programmable computer or an electronic circuit. In some embodiments, one or more of the most important method steps may be executed by such an apparatus.
Depending on certain implementation requirements, embodiments of the invention can be implemented in hardware or in software or at least partially in hardware or at least partially in software. The implementation can be performed using a digital storage medium, for example a floppy disk, a DVD, a Blu-Ray, a CD, a ROM, a PROM, an EPROM, an EEPROM or a FLASH memory, having electronically readable control signals stored thereon, which cooperate (or are capable of cooperating) with a programmable computer system such that the respective method is performed. Therefore, the digital storage medium may be computer readable.
Some embodiments according to the invention comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.
Generally, embodiments of the present invention can be implemented as a computer program product with a program code, the program code being operative for performing one of the methods when the computer program product runs on a computer. The program code may for example be stored on a machine readable carrier.
Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine readable carrier.
In other words, an embodiment of the inventive method is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.
A further embodiment of the inventive methods is, therefore, a data carrier (or a digital storage medium, or a computer-readable medium) comprising, recorded thereon, the computer program for performing one of the methods described herein. The data carrier, the digital storage medium or the recorded medium are typically tangible and/or non-transitory.
A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for performing one of the methods described herein. The data stream or the sequence of signals may for example be configured to be transferred via a data communication connection, for example via the Internet.
A further embodiment comprises a processing means, for example a computer, or a programmable logic device, configured to or adapted to perform one of the methods described herein.
A further embodiment comprises a computer having installed thereon the computer program for performing one of the methods described herein.
A further embodiment according to the invention comprises an apparatus or a system configured to transfer (for example, electronically or optically) a computer program for performing one of the methods described herein to a receiver. The receiver may, for example, be a computer, a mobile device, a memory device or the like. The apparatus or system may, for example, comprise a file server for transferring the computer program to the receiver.
In some embodiments, a programmable logic device (for example a field programmable gate array) may be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor in order to perform one of the methods described herein. Generally, the methods are preferably performed by any hardware apparatus.
The apparatus described herein may be implemented using a hardware apparatus, or using a computer, or using a combination of a hardware apparatus and a computer.
The methods described herein may be performed using a hardware apparatus, or using a computer, or using a combination of a hardware apparatus and a computer.
The above described embodiments are merely illustrative for the principles of the present invention. It is understood that modifications and variations of the arrangements and the details described herein will be apparent to others skilled in the art. It is the intent, therefore, to be limited only by the scope of the impending patent claims and not by the specific details presented by way of description and explanation of the embodiments herein.
For further background knowledge, in particular with respect to the above mentioned escape coding mechanism, the entire content including the description, the claims, the figures and the abstract of European Patent Application No. 17162866.2 is incorporated herein by reference.
