JP3919038B2 - Variable length coding method and apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a variable length coding method and apparatus used in a signal transmission system using a binary code.
[0002]
[Prior art]
In order to efficiently compress video information, various video encoding methods have been internationally standardized and used in various applications. Among them, in the information source encoding unit that performs predictive coding such as quantization and motion compensation prediction, transform coding such as orthogonal transform, etc., the encoding efficiency is improved by an adaptive method according to the image properties such as the prediction method. However, in a transmission path coding unit that performs variable length coding, error correction coding, etc., a variable length code (hereinafter referred to as VLC) is used in a fixed manner in order to omit generation processing. ing.
[0003]
On the other hand, it has been reported that the coding efficiency can be improved by adaptively updating the VLC (Uekura et al., “Study on Adaptive Variable Length Coding Method for Video”, 1997 Video Information Media Annual Conference 21-4, p.262, 19977.7).
[0004]
[Problems to be solved by the invention]
However, if the VLC is arbitrarily designed to have an arbitrary bit pattern as in the proposed method, a unique word (hereinafter referred to as UW) that is periodically inserted to ensure synchronization may not be used. There is.
[0005]
Accordingly, an object of the present invention is to provide a variable-length coding method and apparatus capable of improving coding efficiency while ensuring synchronization.
[0006]
[Means for Solving the Problems]
According to the present invention, there is provided a variable-length coding method for generating and using an emulation-less VLC having a bit pattern that does not emulate UW, and measuring and measuring the occurrence probability of an event (event) every predetermined period. Variable length coding method using emulationless VLC obtained by adaptively generating emulationless VLC assigned to an event for each predetermined period according to the occurrence probability of the event for each of a plurality of different events Is provided.
[0007]
According to the present invention, there is further provided a variable-length encoding device that generates and uses an emulation-less VLC having a bit pattern that does not emulate UW, and includes a measuring unit that measures the occurrence probability of an event for each predetermined period, A generation unit adapted to adaptively generate an emulation-less VLC assigned to an event according to the occurrence probability of the event measured by the measurement unit for each predetermined period, and a plurality of different measurement and generation by the measurement unit There is provided a variable length coding device using emulationless VLC obtained by performing each of the above events.
[0008]
As shown in FIG. 1, UW is inserted into an encoded bit stream at a constant period (for example, one slice or one frame) corresponding to an image frame structure in order to ensure synchronization during decoding. VLC and fixed-length codes (hereinafter referred to as FLC) in the encoded payload require that no combination of codewords emulate UW. Conventionally, emulation has been solved by an empirical method. However, a systematic generation method of a code that prevents emulation while substantially maintaining coding efficiency has already been proposed by the present applicant (Japanese Patent Laid-Open No. 9-232968). Publication). This method generates a unique VLC from the UW bit length (N) consisting only of “0” bits and the continuous “0” bit length restrictions (s, t) allowed for the prefix and suffix of the VLC. Thus, it can be easily applied to video coding.
[0009]
In the present invention, the occurrence probability of an event is measured for each predetermined period, and a VLC assigned to each event is adaptively generated for each predetermined period according to the measured occurrence probability of the event, and the generated VLC As described above, an emulation-less VLC having a bit pattern that does not emulate UW as described above is generated. For this reason, encoding efficiency can be improved while ensuring synchronization reliably. Moreover, even when switching between the old VLC and the new VLC, UW emulation does not occur even if the old and new VLC are combined, and synchronization can be ensured.
[0010]
It is preferable to use a newly generated emulationless VLC within a predetermined period during which the measurement is performed. Since the VLC is created based on the probability of occurrence occurring in the predetermined period, naturally the coding efficiency of the VLC is optimized, and there is no delay from measurement to application.
[0011]
It is also preferable to use the newly generated emulationless VLC within a predetermined period following the predetermined period during which the measurement is performed. Since the VLC table can be created on the receiving side as well as the transmitting side from the measured information for a predetermined period, transmission of the VLC table (overhead information) is unnecessary, and no buffering delay is required for the data for the predetermined period. It becomes.
[0012]
It is also highly preferable to generate an emulation-less VLC for each event obtained by concentrating the events that occur to a specific event.
[0018]
The predetermined period described above may be a T (natural number) image frame.
[0019]
In the embodiment of the present invention, each event is a DCT transform coefficient “0” run length and level, a motion vector used for motion compensation prediction, or a coded block pattern.
[0020]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[0021]
FIG. 2 is a block diagram schematically showing a part of the configuration of the encoding apparatus as an embodiment of the present invention.
[0022]
In the figure, 20 is a motion compensated prediction encoding unit, 21 is a DCT (Discrete Cosine Transform) conversion unit, 22 is a quantization unit, 23 is a zigzag scan or other method event scanning unit, and 24 is a variable related to the present invention. Each of the long encoding units is shown. Since each part except the variable length encoding part 24 is a known structure, description is abbreviate | omitted in this specification.
[0023]
FIG. 3 is a block diagram schematically showing the configuration of the variable length coding unit 24 in the present embodiment.
[0024]
In the figure, reference numeral 240 denotes, for example, a coding event (event) after a zigzag scan, for example, “0” run length and level of a DCT transform coefficient, a coordinate value of a motion vector used for motion compensation prediction, and a significant coding block pattern. An occurrence probability measurement unit that measures occurrence probabilities of blocks and non-significant blocks every predetermined time, 241 receives a set of pairs of events and measured occurrence probabilities, and concentrates events to reduce the number of events The VLC generator 242 receives a set of the centralized event and its occurrence probability from the centralizer 241 and generates an emulationless VLC having a bit pattern that does not emulate UW. For example, a buffer unit for delaying a coding event after zigzag scanning, 2 4 shows the VLC allocation unit for allocating a receive VLC the VLC table generated from VLC generating unit 242, respectively.
[0025]
In the present embodiment, it is assumed that every predetermined time, which is the timing for updating the VLC, is every T (natural number) image frame. The occurrence probability measurement unit 240 measures the occurrence probability of each event for each T frame and outputs a set of sets of these events and the measured occurrence probabilities.
[0026]
FIG. 4 is a flowchart for explaining the operation of the event concentration unit 241 shown in FIG.
[0027]
As shown in the figure, the centralizing unit 241 first classifies the event by knowing the type of the event from the set of the set of the received event and the occurrence probability (step S1).
[0028]
When the event is a motion vector (step S2), when the inter-frame difference error value is equal to or less than a threshold value, the event is concentrated on a vector having a specific coordinate value. Specifically, for example, when X> Xth and X is an odd number, X = [X / 2] × 2 (where [X] is equal to X or closer to “0” than X and The vector coordinates are converted such that it is an integer closest to X (step S3), and the set of the event and the occurrence probability is corrected (step S4).
[0029]
If the event is a coded block pattern (step S5), the block is treated as a non-coded block when the activity of the DCT transform coefficient of the block is less than a threshold value. Specifically, for example, the activity A of the significant block is measured by measuring the absolute value sum of the signals of the significant blocks whose conversion coefficients are not “0” (step S6). If the activity A is A ≧ Ath, (Step S7) No centralization is performed, and if A <Ath (Step S7), the block is treated as an insignificant block (Step S8), and the combination of the event and the occurrence probability is corrected and concentrated (Step S7). Step S4).
[0030]
If the event is a DCT transform coefficient (step S9), the DCT transform coefficient is concentrated to a specific level at the quantization level. Specifically, the level value Level of the DCT conversion coefficient is not converted when, for example, | Level | ≦ Th, and Level = [Level / 2] × 2 (takes only an even number) otherwise. Level conversion is performed (step S10), and a set of events and occurrence probabilities is corrected and concentrated (step S4).
[0031]
The VLC generation unit 242 receives a set of pairs of events and occurrence probabilities centralized in this way, and newly generates (updates) an emulation-less VLC having a bit pattern that does not emulate UW for each T frame. .
[0032]
Emulation-less VLC creates a UW consisting of consecutive bits “0” of length N (N is a plurality of natural numbers), and the continuous length of bits “0” in the prefix of the codeword is s (s is a natural number) or less. Each code word including at least one bit “1” is created, and the continuous length of the bit “0” in the suffix of each created code word is less than or equal to t (t is a natural number), It is generated by correcting the bit pattern of each codeword so that the continuous length of bit “0” is less than N and s + t <N.
[0033]
The requirements for the characteristics of the UW and the structure of the VLC to prevent the VLC from emulating the UW are as follows.
[0034]
As UW characteristics, it is assumed that this UW satisfies the following conditions.
(Assumption 1) The UW used in the handled code is of a single type.
(Assumption 2) It is assumed that UW is composed of only bit “0”.
[0035]
Assumption 1 is an important condition to facilitate the generation of VLC, but this does not limit the usage of UW. When it is necessary to use a plurality of types of UWs, a single type of UW defined here may be used as an escape symbol, and an extension code assigned for each function may be placed behind it. Assumption 2 is an important condition for facilitating the generation of VLC as well as a widely used setting because of the ease of recognition in an actual system. The bit length of UW is represented as N.
[0036]
The requirement for VLC that does not emulate UW depends on the characteristics of UW based on the above assumption and the appearance pattern of VLC sets (not the appearance pattern of codewords in each code). Regarding the appearance pattern of the VLC set, first, the simplest single code infinite loop transmission system is assumed. In this system, each code word appears based only on the occurrence probability of a symbol in the code. The requirements for each codeword in this system are as follows.
(Condition 1) Code words composed only of bit “0” are prohibited.
(Condition 2) The maximum continuous length of bit “0” in the prefix of each codeword is set to s or less.
(Condition 3) The maximum continuous length of bit “0” in the suffix of each code word is set to t or less.
(Condition 4) s + t <N.
(Condition 5) The maximum continuous length of the bit “0” in the middle of each codeword is less than N.
[0037]
(Condition 1) is a condition for preventing UW from being emulated when the same codeword appears continuously. (Condition 2), (Condition 3), and (Condition 4) are conditions that prevent a combination of two codewords from emulating UW. (Condition 5) is a condition for preventing emulation due to the inclusion of UW in the code word. Satisfying (Conditions 1 to 5) is a necessary and sufficient condition not to cause UW emulation.
[0038]
As described above, the VLC generation unit 242 updates the VLC for each T frame based on the measured occurrence probability of the event. For events having a large number of event types such as “0” run length and level of DCT conversion coefficient, a plurality of events having low occurrence probabilities are collectively treated as escape events, and identified by subsequent FLC.
[0039]
FIG. 5 shows an example of the VLC table obtained by actually updating in this way when the event is the DCT transform coefficient “0” run length and level. FIG. 6 shows a conventional MPEG-2 VLC table. In these figures, RUN is a run length of a coefficient of “0”, LEVEL is a level of a coefficient other than “0”, and VLC is set corresponding to each of these events.
[0040]
The VLC allocation unit 244 allocates a VLC corresponding to the updated VLC table from the VLC generation unit 242 for the input encoding event. Regarding the application of the updated VLC, the method used in the T frame period in which measurement is performed as shown in FIG. 7A and the T frame period in which measurement is performed as shown in FIG. Schemes used within a T frame period. In the present embodiment, a method of applying VLC within the T frame period in which the former measurement is performed is used. Since the VLC is generated based on the occurrence probability generated during the T frame period, naturally the encoding efficiency of the VLC is optimized, and there is no delay from measurement to application. However, in this case, it is necessary to buffer the encoding events for T frames in the buffer unit 243. Further, transmission (overhead information) of the VLC table is performed from the VLC generation unit 242 to the reception side.
[0041]
As described above, according to the present embodiment, the occurrence probability of an event is measured for each T frame period, and the VLC assigned to each event according to the measured event occurrence probability is adaptive for each T frame period. An emulation-less VLC having a bit pattern that does not emulate UW is generated as a VLC that is newly generated. For this reason, encoding efficiency can be improved while ensuring synchronization reliably. Moreover, even when switching between the old VLC and the new VLC, UW emulation does not occur even if the old and new VLC are combined, and synchronization can be ensured. In addition, since the VLC is generated based on the occurrence probability generated in the T frame period, naturally, the VLC encoding efficiency is optimized, and there is no delay from measurement to application.
[0042]
By using adaptive VLC in this way, an event set can be selected to shorten the average code length. That is, by reducing the event types by concentrating the occurrence of specific events, there is a trade-off with an increase in inter-frame difference error and quantization error, but the VLC code length is reduced by reducing the VLC average code length. Efficiency can be further improved. For example, in the “0” run length and level of the DCT transform coefficient, selectable levels are discretely set and rounded to the nearest level. Further, regarding the selection of a motion vector, vector coordinate values are set discretely and the closest vector is selected.
[0043]
In the embodiment described above, the event centralization is performed. However, as long as the emulationless VLC is configured to be updated, the event centralization may not be performed. However, the encoding efficiency is somewhat lowered. In addition, when performing event centralization, a normal VLC may be updated. However, in that case, synchronization may be uncertain because of the possibility of emulating UW.
[0044]
FIG. 8 is a block diagram schematically showing a configuration of a variable length coding unit in another embodiment of the present invention.
[0045]
In the figure, reference numeral 840 denotes an encoding event after the zigzag scan, for example, “0” run length and level of the DCT transform coefficient, coordinate value of the motion vector used for motion compensation prediction, and significant encoding block pattern An occurrence probability measuring unit that measures the occurrence probabilities of blocks and non-significant blocks every predetermined time, 841 is a concentration that receives a set of pairs of events and measured occurrence probabilities and centralizes the events to reduce the number of events 842 is a VLC generation unit that receives a set of a set of events centralized from the centralization unit 841 and their occurrence probabilities and generates an emulation-less VLC having a bit pattern that does not emulate UW, and 844 is a VLC generation unit A VLC allocation unit that receives the VLC table generated from 842 and allocates the VLC. Re respective shows.
[0046]
In the present embodiment, a method is used in which the updated VLC is used within a T frame period subsequent to a T frame period in which measurement is performed as illustrated in FIG. Since the VLC table can be created on the receiving side as well as the transmitting side from the information of the measured T frame period, transmission of the VLC table (overhead information) is not necessary, and buffering delay for data for the T frame period Is no longer necessary. Other configurations in the present embodiment are the same as those in the embodiment of FIG.
[0047]
According to this embodiment, the occurrence probability of an event is measured every T frame period, and a VLC assigned to each event is adaptively generated for each T frame period according to the measured event occurrence probability. An emulation-less VLC having a bit pattern that does not emulate UW is generated as the generated VLC. For this reason, encoding efficiency can be improved while ensuring synchronization reliably. Moreover, even when switching between the old VLC and the new VLC, UW emulation does not occur even if the old and new VLC are combined, and synchronization can be ensured. In addition, since VLC is applied within the next T frame period, transmission of VLC table (overhead information) is unnecessary, and no buffering delay is required for data for the T frame period.
[0048]
By using adaptive VLC in this way, an event set can be selected to shorten the average code length. That is, by reducing the event types by concentrating the occurrence of specific events, there is a trade-off with an increase in inter-frame difference error and quantization error, but the VLC code length is reduced by reducing the VLC average code length. Efficiency can be further improved. For example, in the “0” run length and level of the DCT transform coefficient, selectable levels are discretely set and rounded to the nearest level. Further, regarding the selection of a motion vector, vector coordinate values are set discretely and the closest vector is selected.
[0049]
In this embodiment, the event centralization is performed. However, as long as the emulationless VLC is configured to be updated, the event centralization may not be performed. However, the encoding efficiency is somewhat lowered. In addition, when performing event centralization, a normal VLC may be updated. However, in that case, synchronization may be uncertain because of the possibility of emulating UW.
[0050]
The encoding efficiency of the method of each of the above embodiments was measured by computer simulation. In the MPEG-2 encoding, as an event, the VLC is adaptively updated with respect to the “0” run length and level of the DCT transform coefficient. Mobile & Calendar and Flower Garden were used as test images. In MPEG-2, since various start codes are composed of 23 consecutive “0” bits followed by an identifier, N = 23, s = 19, and t = 3. Since the maximum codeword length is a few bits shorter than 23 bits, the efficiency degradation compared to a normal Huffman code is less than 1%. In event measurement, events that occur once in the T frame period are integrated into an escape event, which is identified by subsequent 18-bit FLC. Further, the transmission of the VLC table in the embodiment of FIG. 3 adopting the method of FIG. 7A requires the event name (FLC) 18 bits and the codeword 32 bits in total 50 bits / event overhead information. And For example, when 200 events exist in the VLC and VLC update is performed for each frame (T = 1), 300 kbit / s is required.
[0051]
(1) Effect of adaptive VLC update Measurement was performed at T = 1, 3, 6, and 15. The coding efficiency for T = 1, 3 is shown in FIG. Note that overhead information is taken into consideration in the method of the embodiment of FIG. 3 (method (A)). In the method (A), the efficiency is maximum at T = 1, and a rate reduction of 10.3% (Q = 2) is confirmed for both images. In the method of the embodiment of FIG. 8 (method (B)), the efficiency is maximum at T = 3, and the gain is 6.6% (Q = 2). In the method (B), the number of events is limited at a small T, and the usage rate of escape events in the VLC usage frame period is high. Therefore, it is considered that the efficiency is maximized at T = 3. Comparing both methods shows that the method (A) can obtain higher efficiency even if overhead information is included. Note that the difference from the standard VLC decreases as the encoding rate decreases in both systems. This is probably because standard VLC is optimized at a low coding rate.
[0052]
(2) Effect of event centralization In the methods (A) and (B), coding efficiency was measured when the DCT coefficient level values were limited as follows.
if | level | ≦ 10, level = level
else level = [level / 2] × 2
However, [x] is an integer that is equal to x or closer to 0 than x and closest to x.
FIG. 10 shows a comparison of coding efficiency according to the presence / absence of event concentration with respect to method T = 1 in (A) and method T = 3 in (B). There is a slight improvement in efficiency. Since the number of events is reduced by about 13%, the processing load in VLC / VLD (variable length coding / variable length decoding) can be reduced.
[0053]
The above-described embodiments are all illustrative of the present invention and are not limited to the present invention, and the present invention can be implemented in various other variations and modifications. Therefore, the scope of the present invention is defined only by the claims and their equivalents.
[0054]
【The invention's effect】
As described above in detail, according to the present invention, there is provided a variable-length encoding method for generating and using an emulation-less VLC having a bit pattern that does not emulate UW, wherein the occurrence probability of an event is determined for each predetermined period. The emulation-less VLC obtained by performing each of a plurality of different events adaptively generating the emulation-less VLC assigned to the event in accordance with the occurrence probability of the measured event for each predetermined period. And a variable-length encoding apparatus that generates and uses an emulation-less VLC having a bit pattern that does not emulate UW, and includes a measurement unit that measures the occurrence probability of an event at predetermined intervals. Depending on the event occurrence probability measured by this measuring means, Generation means adapted to adaptively generate an emulation-less VLC assigned to each of the predetermined periods, and emulation obtained by performing measurement by the measurement means and generation by the generation means for each of a plurality of different events. A variable length coding device using less VLC is provided. Thus, the occurrence probability of an event is measured every predetermined period, and a VLC assigned to each event is adaptively generated for each predetermined period according to the measured occurrence probability of the event, and the generated VLC As described above, an emulation-less VLC having a bit pattern that does not emulate UW as described above is generated. For this reason, encoding efficiency can be improved while ensuring synchronization reliably. Moreover, even when switching between the old VLC and the new VLC, UW emulation does not occur even if the old and new VLC are combined, and synchronization can be ensured.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating UWs inserted into an encoded bitstream.
FIG. 2 is a block diagram schematically showing a partial configuration of an encoding apparatus as an embodiment of the present invention.
FIG. 3 is a block diagram schematically showing the configuration of a variable length coding unit in the embodiment of FIG. 2;
4 is a flowchart for explaining an operation of an event centralization unit in FIG. 3; FIG.
FIG. 5 is a diagram illustrating an example of a VLC table updated according to the embodiment of FIG. 2;
FIG. 6 is a diagram illustrating a conventional MPEG-2 VLC table.
FIG. 7 is a diagram for explaining an application period of an updated VLC.
FIG. 8 is a block diagram schematically showing a configuration of a variable length coding unit according to another embodiment of the present invention.
FIG. 9 is a diagram illustrating a comparison of coding efficiency in adaptive VLC.
FIG. 10 is a diagram illustrating coding efficiency of VLC by a centralized event set.
[Explanation of symbols]
20 motion compensated prediction encoding unit 21 DCT conversion unit 22 quantization unit 23 scanning unit 24 variable length encoding unit 240, 840 occurrence probability measurement unit 241 841 concentration unit 242, 842 VLC generation unit 243 buffer unit 244, 844 VLC allocation Part

Claims (16)

  A variable-length encoding method for generating and using an emulation-less variable-length code having a bit pattern that does not emulate a unique word, and measuring the occurrence probability of an event for each predetermined period and measuring the occurrence probability of the event The emulation-less variable-length code assigned to the event is adaptively generated for each predetermined period, and the emulation-less variable-length code obtained by performing each for a plurality of different events is used. A variable length encoding method.   The method according to claim 1, wherein the generated emulationless variable length code is used within a predetermined period during which the measurement is performed.   The method according to claim 1, wherein the generated emulationless variable length code is used within a predetermined period following the predetermined period during which the measurement is performed.   The method according to any one of claims 1 to 3, wherein an emulationless variable length code is generated for each event obtained by concentrating the generated events on a specific event.   5. The method according to claim 1, wherein the predetermined period is a T (natural number) image frame.   The method according to claim 1, wherein the event includes a DCT transform coefficient “0” run length and level.   The method according to claim 1, wherein the event includes a motion vector used for motion compensation prediction.   The method according to claim 1, wherein the event includes an encoded block pattern.   A variable-length encoding device that generates and uses an emulation-less variable-length code having a bit pattern that does not emulate a unique word, the measurement means for measuring the occurrence probability of an event every predetermined period, and the measurement means A generation unit adapted to adaptively generate an emulationless variable length code assigned to the event according to the occurrence probability of the event for each predetermined period, and a plurality of different measurement and generation by the measurement unit A variable-length coding apparatus using emulation-less variable-length codes obtained by performing each of the above events.   10. The apparatus according to claim 9, further comprising means for using the emulationless variable length code generated by the generating means within a predetermined period during which the measurement is performed.   10. The apparatus according to claim 9, further comprising means for using the emulationless variable length code generated by the generating means within a predetermined period following the predetermined period during which the measurement is performed.   Concentrating means for concentrating the generated events to specific events is further provided, and the generating means is configured to generate an emulationless variable length code for each event obtained by the centralizing means. The device according to any one of claims 9 to 11.   13. The apparatus according to claim 9, wherein the predetermined period is a T (natural number) image frame.   14. An apparatus according to any one of claims 9 to 13, wherein the event comprises a "0" run length and level of DCT transform coefficients.   The apparatus according to claim 9, wherein the event includes a motion vector used for motion compensation prediction.   The apparatus according to any one of claims 9 to 15, wherein the event includes a coded block pattern.

Images