JP4211780B2 - Digital signal encoding apparatus, digital signal decoding apparatus, digital signal arithmetic encoding method, and digital signal arithmetic decoding method - Google Patents

Description

  The present invention relates to a digital signal encoding device, a digital signal decoding device, a digital signal arithmetic encoding method, and a digital signal arithmetic decoding method used for video compression encoding technology, compressed video data transmission technology, and the like.

  Conventionally, Huffman coding is adopted as entropy coding in international standard video coding schemes such as MPEG and ITU-T H.26x. Huffman coding gives optimum coding performance when individual source symbols need to be expressed as individual codewords, but on the other hand, the signal behavior varies locally like video signals There is a problem that the optimality is not guaranteed when the occurrence probability of the source symbol varies. In such a case, arithmetic coding has been proposed as a method of dynamically adapting to the occurrence probability of each information source symbol and expressing a plurality of symbols together by one code word. A brief explanation of the concept of arithmetic coding is given by citing Mark Nelson, "Arithmetic Coding + Statistical Modeling = Data Compression Part 1-Arithmetic Coding", Dr. Dobb's Journal, February 1991. Here, consider an information source that uses alphabetic characters as an information source symbol, and consider arithmetically encoding the message “BILL GATES”. At this time, the occurrence probability of each character is defined as shown in FIG. Also, as shown in the range of values in the figure, the area on the probability number straight line defined by the interval [0, 1] is uniquely determined. Next, the encoding process is started. First, the character “B” is encoded, which corresponds to specifying the range [0.2, 0.3] on the probability number line. Therefore, the letter “B” corresponds to a set of upper limit (High) and lower limit (Low) values in the range [0.2, 0.3]. Next, when encoding "I", the range [0.2, 0.3] specified by the encoding of "B" is again considered as the [0, 1] section, and the section of [0.5, 0.6] is specified. To do. That is, the arithmetic encoding process corresponds to narrowing down the range of probability line. When this process is repeated for each character, as shown in FIG. 21, the arithmetic encoding result of “BILL GATES” is expressed as a low value “0.2572167752” when encoding of the character “S” is completed. The

  The reverse of the decoding process may be considered. First, the encoding result “0.2572167752” is checked to which value range on the probability number line is assigned, and “B” is obtained. After that, after subtracting the low value of “B”, division is performed in the range to obtain “0.572167752”. As a result, the character “I” corresponding to the interval [0.5, 0.6] can be decoded. Thereafter, this process can be repeated to decode “BILL GATES”. From the above processing, according to arithmetic coding, even endless message encoding is finally mapped to one codeword. Due to the fact that infinite decimal point precision cannot be handled in actual implementation, multiplication and division are necessary for the encoding / decoding process, and the calculation load is high, for example, floating using an integer register as a codeword representation Decimal points are calculated, and measures such as replacing the multiplication and division with shift operations by approximating the occurrence probability of the low value by a power of 2 have been made. According to the arithmetic coding, ideally, entropy coding that is well suited to the occurrence probability of the information source symbol is possible by the above-described process. In particular, when the occurrence probability changes dynamically, it is possible to obtain higher encoding efficiency than Huffman encoding by tracing the state of occurrence probability change and appropriately updating the table of FIG. .

Mark Nelson, "Arithmetic Coding + Statistical Modeling = Data Compression Part 1-Arithmetic Coding", Dr. Dobb's Journal, February 1991.

  By the way, when transmitting an entropy-encoded video signal, usually, in order to minimize the disturbance of the video due to transmission errors, each frame of the video is divided into partial areas and resynchronized units, for example, It is often transmitted with MPEG-2 slice structure.

  However, in Huffman coding, each encoding target symbol is mapped to an integer bit length codeword, so it is only necessary to collectively define the corresponding codeword as a transmission unit. In addition to requiring a special code for interrupting the encoding process, when resuming encoding, the learning process of the probability of occurrence of the previous symbol is reset once and the bit for determining the code is set. Since it needs to be discharged, there is a possibility that the encoding efficiency is lowered before and after the interruption. Furthermore, if the arithmetic encoding process is encoded without being reset in one video frame and it must be divided into small units such as packet data at the time of transmission, the decoding process for a certain packet There is a problem that the video quality is significantly deteriorated when packet loss due to transmission errors or delays cannot be performed without data.

  Therefore, as described in the above prior art, the present invention is a digital having a mechanism for determining an optimal trade-off between coding efficiency and error resilience, which is a problem when arithmetic coding is used as entropy coding for video coding. It is an object of the present invention to provide a signal encoding device, a digital signal decoding device, a digital signal arithmetic encoding method, and a digital signal arithmetic decoding method. Note that the present invention is applicable to all devices and methods using arithmetic coding.

  In order to solve the above-described problems, the present invention provides an arithmetic coding unit that compresses a predetermined unit of a digital signal by arithmetic coding in a digital signal encoding apparatus that performs compression encoding by dividing a digital signal into predetermined units. The arithmetic encoding unit is a digital signal encoding apparatus that multiplexes information representing the arithmetic encoding state at the time when encoding of a certain transmission unit is completed as part of data of the next transmission unit. It is characterized by.

  In particular, the arithmetic coding unit determines an occurrence probability of a coded symbol based on a dependency relationship between a digital signal of a predetermined unit and a signal included in one or a plurality of adjacent transmission units. It is a digital signal encoding device to be converted.

  The arithmetic coding unit is a digital signal coding device that learns the occurrence probability by counting the appearance frequency of the symbol to be coded.

  The information representing the arithmetic coding state is a digital signal which is a register reset flag indicating whether or not a register value indicating the arithmetic coding process is reset and an initial register value added only when the register value is not reset. It is an encoding device.

  Further, a digital signal encoding apparatus that performs compression encoding by dividing a digital signal into predetermined units includes an arithmetic encoding unit that compresses a predetermined unit of digital signal by arithmetic encoding, and the arithmetic encoding unit includes a predetermined unit Determines the probability of occurrence of a coded symbol and counts the frequency of occurrence of the symbol to be coded based on the dependency between the digital signal of the unit and a signal included in one or more adjacent transmission units The digital signal encoding device that learns the occurrence probability and multiplexes information representing the occurrence probability learning state at the time when encoding of a certain transmission unit is completed as a part of the next transmission unit data. It is characterized by.

  In particular, the information representing the occurrence probability learning state is a digital signal encoding device that is information indicating a context model state obtained by modeling a dependency relationship with other information that causes variation in the occurrence probability of the encoded symbol. It is characterized by that.

  The digital signal may be a video signal, and the transmission unit may be a digital signal encoding device that is a slice composed of one or more macroblocks in a video frame.

  The digital signal is a video signal, and the transmission unit is a digital signal encoding device that is an encoded data unit reconfigured according to the type of encoded data included in the slice. And

  The digital signal is a video signal, and the transmission unit is a digital signal encoding device which is a video frame.

  In addition, in a digital signal decoding apparatus that receives and decodes a compression-coded digital signal in a predetermined unit, the digital signal decoding apparatus includes an arithmetic decoding unit that decodes the compressed digital signal in a predetermined unit based on an arithmetic coding process. The decoding unit is a digital signal decoding device that initializes a decoding operation based on information representing an arithmetic coding state multiplexed as a part of transmission unit data at the start of decoding of a certain transmission unit. It is characterized by.

  In particular, the arithmetic decoding unit determines a probability of occurrence of a decoded symbol based on a dependency relationship between signals included in one or a plurality of adjacent transmission units when decoding a compressed digital signal of a predetermined unit. It is a digital signal decoding device that performs.

  In particular, the arithmetic decoding unit is a digital signal decoding device that learns the occurrence probability by counting the appearance frequency of decoded symbols.

  In addition, in a digital signal decoding apparatus that receives and decodes a compression-coded digital signal in a predetermined unit, the digital signal decoding apparatus includes an arithmetic decoding unit that decodes the compressed digital signal in a predetermined unit based on an arithmetic coding process. The decoding unit initializes the occurrence probability used for decoding the transmission unit based on information representing the symbol occurrence probability learning state multiplexed as part of the transmission unit data at the start of decoding of the transmission unit. And, when decoding a compressed digital signal of a predetermined unit, determines the probability of occurrence of a decoded symbol based on the dependency between signals included in one or a plurality of adjacent transmission units, and the appearance of a symbol to be decoded It is a digital signal decoding device that learns the occurrence probability by counting the frequency and performs decoding.

  In particular, the digital signal is a video signal, and the transmission unit is a digital signal decoding device which is a slice composed of one or a plurality of macroblocks in a video frame.

  The digital signal may be a video signal, and the transmission unit may be a digital signal decoding device that is an encoded data unit reconfigured according to the type of encoded data included in the slice. To do.

  The digital signal is a video signal, and the transmission unit is a digital signal decoding device which is a video frame.

Also, a digital signal arithmetic encoding method when performing compression encoding by dividing a digital signal into predetermined units,
A digital signal that multiplexes information representing the arithmetic coding state at the time when encoding of a certain transmission unit is completed as part of the data of the next transmission unit when the digital signal of a predetermined unit is compressed by arithmetic coding It is a decoding device.

  In addition, a digital signal arithmetic coding method for compressing and encoding a digital signal by dividing the digital signal into predetermined units. When the digital signal of a predetermined unit is compressed by arithmetic encoding, the digital signal of the predetermined unit is 1 Based on the dependency between signals contained in one or more adjacent transmission units, the occurrence probability of the encoded symbol is determined and the occurrence probability is learned by counting the appearance frequency of the encoded symbol. The digital signal arithmetic coding method multiplexes information expressing the occurrence probability learning state at the time when encoding of a certain transmission unit is completed as a part of the next transmission unit data.

  Also, a digital signal arithmetic decoding method for receiving and decoding a compression-encoded digital signal in a predetermined unit, when decoding a predetermined unit of a compressed digital signal based on an arithmetic encoding process A digital signal arithmetic decoding method for initializing a decoding operation based on information representing an arithmetic coding state multiplexed as part of the transmission unit data at the start of decoding of the transmission unit. .

  Also, a digital signal arithmetic decoding method for receiving and decoding a compression-encoded digital signal in a predetermined unit, when decoding a predetermined unit of a compressed digital signal based on an arithmetic encoding process At the start of decoding of a transmission unit, initialization of the occurrence probability used for decoding of the transmission unit is performed based on information representing a symbol occurrence probability learning state multiplexed as a part of the transmission unit data. Determining the probability of occurrence of a decoded symbol based on the dependency between signals included in one or more adjacent transmission units and counting the frequency of appearance of the decoded symbol In the digital signal arithmetic decoding method, the occurrence probability is learned and decoded.

  According to the digital signal encoding apparatus and the digital signal arithmetic encoding method of the present invention, when a digital signal of a predetermined transmission unit is compressed by arithmetic encoding, the arithmetic encoding state at the time when encoding of a certain transmission unit ends is determined. The information to be expressed may be multiplexed as part of the data of the next transmission unit, or based on the dependency between signals included in one or more adjacent transmission units, The occurrence probability is determined and the occurrence probability is learned by counting the appearance frequency of the symbol to be encoded, and information representing the occurrence probability learning state at the time when encoding of a certain transmission unit is completed is transmitted to the next transmission. Since it is multiplexed as part of the unit data, it is not necessary to reset the previous arithmetic coding state or symbol occurrence probability learning state. Which takes over can continue encoding, while maintaining the error resilience, it is possible to implement the enhanced encoding the coding efficiency of arithmetic coding.

  Further, according to the digital signal decoding apparatus and the digital signal arithmetic decoding method of the present invention, when decoding of a certain transmission unit is started, based on information expressing an arithmetic coding state multiplexed as a part of the transmission unit data. , Decoding of the transmission unit based on information representing a symbol occurrence probability learning state multiplexed as a part of the transmission unit data at the start of decoding operation or decoding start of a certain transmission unit Initializing the occurrence probability to be used, and determining the occurrence probability of a decoded symbol based on the dependency between signals included in one or more adjacent transmission units when decoding a compressed digital signal of a predetermined transmission unit At the same time, the occurrence probability is learned and counted by counting the appearance frequency of the symbol to be decoded. Even when the taken over by coding without resetting the operative coding state or symbol occurrence probability learning status, it is possible to decode correctly.

Embodiment. 1
In the first embodiment, D. Marpe is applied as an example in which arithmetic coding is applied to a video coding method in which a video frame is coded in a unit equally divided into 16 × 16 pixel rectangular areas (hereinafter referred to as macroblocks). The present invention will be described using an example disclosed in “Video Compression Using Context-Based Adaptive Arithmetic Coding”, International Conference on Image Processing 2001.

FIG. 1 shows the configuration of the video encoding apparatus according to the first embodiment.
In FIG. 1, the video coding apparatus according to the first embodiment includes a motion detection unit 2, a frame memory 3a, an arithmetic coding unit 6, a motion compensation unit 7, a spatial prediction unit 10a, a coding mode determination unit 12, an orthogonal transform. Section 15, quantization section 16, inverse quantization section 18, inverse orthogonal transform section 19, encoding control section 22, transmission buffer 24, subtractor 51, switching section 52, and adder 53.

FIG. 2 shows the configuration of the video decoding apparatus according to the first embodiment.
2, the video decoding apparatus according to the first embodiment includes a frame memory 3b, a motion compensation unit 7, a spatial prediction unit 10b, an inverse quantization unit 18, an inverse orthogonal transform unit 19, an arithmetic decoding unit 27, a switching unit 54, An adder 55 is included.

  In these drawings, the operations of the arithmetic encoding unit 6 and the arithmetic decoding unit 27 are the points of the present invention. First, the operation of the encoding device and the entire decoding device will be described based on these drawings.

1. 1. Overview of Operation of Encoding Device In the encoding device of FIG. 1, it is assumed that an input video signal 1 is input in units in which individual video frames are divided into macroblocks. First, the motion detection unit 2 uses a frame memory 3a. The motion vector 5 is detected in units of macroblocks using the reference image 4 stored in. A temporal prediction image 8 is obtained in the motion compensation unit 7 based on the motion vector 5, and a temporal prediction residual signal 9 is obtained by taking a difference from the input signal 1. On the other hand, the input signal 1 is predicted from a spatial vicinity region in the same video frame in the spatial prediction unit 10a, and a spatial prediction residual signal 11 is generated.

  In the encoding mode determination unit 12, a motion prediction mode for encoding the temporal prediction residual signal 9, or, as a special example thereof, a skip mode in which the motion vector is zero and no temporal prediction residual signal component is present, a spatial prediction residual From the intra mode for encoding the difference signal 11, a mode capable of encoding the macroblock most efficiently is selected.

  The encoding mode information 13 is transferred to the arithmetic encoding unit 6 as encoding target information. When the intra mode is selected as the encoding mode information 13, the intra prediction mode 14 used in the spatial prediction unit 10a is transferred to the arithmetic encoding unit 6 as encoding target information. When the motion prediction mode is selected, the motion vector 5 is transferred to the arithmetic encoding unit 6 as encoding target information.

  The encoding target signal selected by the encoding mode determination unit 12 is passed to the arithmetic encoding unit 6 as orthogonal transform coefficient data 17 through the orthogonal transform unit 15 and the quantization unit 16. After the orthogonal transform coefficient data 17 passes through the inverse quantization unit 18 and the inverse orthogonal transform unit 19, the switching unit 52 performs spatial prediction residual in the spatial prediction image 20, that is, the spatial prediction unit 10 a according to the encoding mode information 13. Of the spatial prediction image candidates used to generate the signal 11, a prediction image generated using the spatial prediction mode 14 or a signal that selects either the temporal prediction image 8 is added to the local decoding. It returns to the image 21. Since the locally decoded image 21 is used for motion prediction of subsequent frames, it is stored in the frame memory 3a as reference image data.

  The quantization unit 16 quantizes the orthogonal transform coefficient data with a granularity given by the quantization step parameter 23 determined by the encoding control unit 22. By adjusting the quantization step parameter 23, the output encoding rate and the quality are balanced. In general, after arithmetic coding, the occupation amount of encoded data accumulated in the transmission buffer 24 immediately before transmission is checked at regular intervals, and the quantization step parameter 23 is adjusted according to the remaining amount 25. Is called. For example, when the remaining amount of buffer is large, the rate is reduced, and when there is a margin in the remaining amount of buffer, the rate is increased to improve the quality.

  The arithmetic encoding unit 6 performs entropy encoding of encoding target data such as the motion vector 5, the encoding mode information 13, the spatial prediction mode 14, and the orthogonal transform coefficient data 17 according to the procedure described later, and via the transmission buffer 24, It is transmitted and recorded as compressed video data 26.

2. Operation Overview of Decoding Device In the decoding device shown in FIG. 2, when video compressed data 26 is received, an entropy decoding process, which will be described later, is performed by an arithmetic decoding unit 27, and motion vector 5, encoding mode information 13, spatial prediction is performed. The mode 14, the orthogonal transform coefficient data 17, the quantization step parameter 23, and the like are restored. The orthogonal transform coefficient data 17 and the quantization step parameter 23 are locally decoded by the same inverse quantization unit 18 and inverse orthogonal transform unit 19 as those on the encoding side. Either temporal prediction or spatial prediction is selected by the encoding mode information 13. In the case of temporal prediction, the temporal prediction image 8 is restored using the motion vector 5 decoded by the motion compensation unit 7. . When the prediction in the spatial direction is selected, the spatial prediction image 11 is restored using the spatial prediction unit 10b. Here, the difference between the spatial prediction unit 10a on the encoding device side and the spatial prediction unit 10b on the decoding device side is that the former is most efficient for all types of spatial prediction modes that the former can take. The latter is limited to only the process of generating the spatial prediction image 20 from the given spatial prediction mode 14.

  In the switching unit 54, either the temporal prediction image 8 or the spatial prediction image 11 is determined as a prediction image based on the coding mode information 13, and this is added by the adder 55 as a local output that is the output of the inverse orthogonal transform unit 19. A decoded image 21 is obtained by adding the decoded signal. The decoded image 21 is stored in the frame memory 3b because it is used to generate a predicted image of a subsequent frame. The difference between the frame memories 3a and 3b merely means the distinction between those mounted on the encoding device and the decoding device.

3. Arithmetic Coding / Decoding Processing The arithmetic coding and decoding processing that is the point of the present invention will be described in detail below. The encoding process is performed in the arithmetic encoding unit 6 in FIG. 1, and the decoding process is performed in the arithmetic decoding unit 27 in FIG.

FIG. 3 shows the internal configuration of the arithmetic encoding unit 6.
In FIG. 3, the arithmetic encoding unit 6 according to the first embodiment uses individual data types such as a motion vector 5, encoding mode information 13, a spatial prediction mode 14, and orthogonal transform coefficient data 17 that are encoding target data. A context model determining unit 28 that defines a context model defined later, a binarizing unit 29 that converts multilevel data into binary data in accordance with a binarization rule defined for each encoding target data type, and binary An occurrence probability generation unit 30 that gives the occurrence probability of the value (0 or 1) of each binarized sequence bin after conversion, an encoding unit 31 that performs arithmetic encoding based on the generated occurrence probability, and arithmetic encoding And a transmission unit generation unit 35 that constitutes data serving as a transmission unit at that timing.

FIG. 4 shows an operation flow of the arithmetic encoding unit 6 shown in FIG.
With reference to FIG. 4, the operation of the arithmetic coding unit 6 shown in FIG. 3 will be described in detail.

1) Context model determination process (step S1)
A context model is a model of the dependency relationship with other information that causes fluctuations in the occurrence probability of an information source (encoded) symbol. By switching the state of the occurrence probability according to this dependency relationship, Thus, it is possible to perform encoding adapted to the actual occurrence probability of the symbol.

  FIG. 5 shows the concept of the context model (ctx). In the figure, the information source symbol is binary. The ctx options of 0 to 2 in the figure are defined on the assumption that the state of the occurrence probability of the information source symbol using this ctx will change according to the situation. In the case of video coding in the first embodiment, the value of ctx is switched according to the dependency between coded data in a certain macroblock and coded data of the surrounding macroblocks.

FIG. 6 shows an example of a context model related to a motion vector of a macroblock disclosed in, for example, D. Marpe et al., “Video Compression Using Context-Based Adaptive Arithmetic Coding”, International Conference on Image Processing 2001. In the figure, the motion vector of block C is the encoding target, and more precisely, the prediction difference value mvd _k (C) _obtained by predicting the motion vector of block C from the vicinity is encoded. ctx_mvd (C, k) indicates a context model. mvd _k (A) represents the motion vector prediction difference value in block A, and mvd _k (B) represents the motion vector prediction difference value in block B, and is used to define the context model switching evaluation value e _k (C). The evaluation value e _k (C) indicates the degree of variation of neighboring motion vectors. Generally, when this variation is small, mvd _k (C) is small, and conversely, e _k (C) is large. In the case, mvd _k (C) also tends to increase. Therefore, it is desirable that the symbol occurrence probability of mvd _k (C) is adapted based on e _k (C). This variation set of occurrence probability is a context model, and in this case, it can be said that there are three types of occurrence probability variations.

  In addition, a context model is defined in advance for each encoding target data such as the encoding mode information 13, the spatial prediction mode 14, and the orthogonal transform coefficient data 17, and the arithmetic encoding unit 6 of the encoding device and the arithmetic decoding of the decoding device. Shared with the unit 27. The context model determination unit 28 of the arithmetic encoding unit 6 shown in FIG. 3 performs a process of selecting a model determined in advance based on the type of data to be encoded. Which occurrence probability variation to select in the context model corresponds to the occurrence probability generation process in 3) below, and will be described there.

2) Binarization (Step S2)
The context model is determined in accordance with each bin (binary position) of the binary sequence after the data to be encoded is binarized by the binarization unit 29. According to the binarization rule, conversion into a variable-length binary sequence is performed in accordance with a rough distribution of values that can be taken by each encoded data. Binarization is a streamlined context model that can reduce the number of probabilistic linear divisions and simplify operations by encoding bin target data as it is rather than arithmetic coding as it is, which can originally take multiple values. There are merits such as becoming possible.

3) Occurrence probability generation process (step S3)
In the above processes 1) and 2), the binarization of the multi-value encoding target data and the setting of the context model applied to each bin are completed, and the preparation for encoding is completed. Since each context model includes variations in occurrence probability for each value of 0/1, the occurrence probability generation unit 30 refers to the context model determined in step S1 as shown in FIG. Generate 0/1 occurrence probability in each bin.

FIG. 6 shows an example of the evaluation value e _k (C) for selecting the occurrence probability. The occurrence probability generation unit 30 determines an evaluation value for selecting an occurrence probability as shown in e _k (C) of FIG. 6, and according to this, which occurrence probability variation is selected from the context model options to be referenced. Determine whether to use for encoding.

4) Encoding process (steps S3 to S7)
Since the occurrence probability of each value of 0/1 on the probability number line necessary for the arithmetic encoding process is obtained by 3), the encoding unit 31 performs arithmetic encoding according to the process given in the conventional example (step S4 ). The actual encoded value (0 or 1) 32 is fed back to the occurrence probability generation unit 30 and the occurrence frequency 0/1 is counted to update the occurrence probability variation portion of the context model used (step S1). S5). For example, when 100 bins are encoded using an occurrence probability variation in a specific context model, the occurrence probability of 0/1 in the occurrence probability variation is 0.25 and 0.75, respectively. . Here, when 1 is encoded using the same occurrence probability variation, the appearance frequency of 1 is updated, and the occurrence probability of 0/1 changes to 0.247 and 0.752. This mechanism makes it possible to perform efficient encoding adapted to the actual occurrence probability.

  Also, the arithmetic code 33 of the encoded value (0 or 1) 32 newly generated by the encoding unit 31 is sent to the transmission unit generation unit 35 and constitutes a transmission unit as described in 6) below. Multiplexed as data (step S6). Also, it is determined whether or not the encoding process has been completed for the entire binary sequence bin of one encoding target data (step S7). If not completed (step S7 “N”), the process goes to step S3. The process after the generation process of the occurrence probability in each bin is performed, and if completed (step S7 “Y”), the process proceeds to the transmission unit generation process described below.

5) Transmission unit generation processing (steps S8 to S9)
Arithmetic coding converts a sequence of a plurality of data to be encoded into one codeword, but video signals perform motion prediction between frames or display in units of frames, so that units of frames are used. It is necessary to generate a decoded image and update the frame memory. Therefore, it is necessary to be able to clearly determine the unit breaks of frames on the arithmetically encoded compressed data. Furthermore, for the purpose of multiplexing with other media such as voice / audio, packet transmission, etc., it is also necessary to divide and transmit the compressed data in smaller units within the frame. An example of this is generally a slice structure, that is, a unit in which a plurality of macroblocks are grouped in raster scan order.

  FIG. 7 shows an example of a slice structure. A rectangle surrounded by a dotted line corresponds to a macroblock. In general, the slice structure is handled as a unit of resynchronization at the time of decoding. As a simple example, there is a case where slice data is directly mapped to a packet payload for IP transmission. RTP (Realtime Transport Protocol) is often used for IP transmission of real-time media that does not allow much transmission delay such as video. RTP packets are often transmitted with a time stamp added to the header portion and video slice data mapped to the payload portion. For example, Kikuchi et al., "RTP Payload Format for MPEG-4 Audio / Visual Streams", RFC 3016, specifies a method for mapping MPEG-4 video compressed data to RTP payload in units of MPEG-4 slices (video packets). Has been. Since RTP is transmitted as a UDP packet, there is generally no retransmission control, and when packet loss occurs, slice data may not reach the entire decoding device. If the subsequent slice data is encoded depending on the information of the discarded slice, it cannot be normally decoded even if it normally reaches the decoding device. For this reason, an arbitrary slice needs to be able to be normally decoded without being bound by any dependency from the beginning. For example, in general, when encoding Slice5, encoding is not performed using the information of the macroblock group of Slice3 positioned at the top or Slice4 positioned on the left.

  On the other hand, in order to improve the efficiency of arithmetic coding, it is desirable to adapt the probability of symbol occurrence based on the surrounding situation or keep the process of dividing the probability number line. For example, to encode Slice5 completely independently of Slice4, the probability number straight line division situation at the time when arithmetic coding of the final macroblock of Slice4 is completed, i.e., a register value representing a codeword in arithmetic coding is set. In Slice 5, encoding is restarted after resetting the register to the initial state. As a result, the correlation existing between the end of Slice 4 and the start of Slice 5 cannot be used, which may lead to a decrease in encoding efficiency. In other words, it is common to design to improve resistance to unexpected loss of slice data due to transmission errors or the like at the expense of a decrease in coding efficiency.

  The transmission unit generator 35 of Embodiment 1 according to the present invention provides a method and apparatus for improving the adaptability of this design. That is, in a case where the probability of loss of slice data due to a transmission error or the like is extremely low, the dependency relationship between slices related to arithmetic coding is not always cut off, but can be actively used. On the other hand, when there is a high possibility of loss of slice data, the dependency between slices can be broken so that the coding efficiency in transmission units can be adaptively controlled.

  That is, the transmission unit generation unit 35 according to the first embodiment receives the transmission unit instruction signal 36 as the control signal inside the encoding apparatus, receives the transmission unit instruction signal 36 at the timing for dividing the transmission unit, and based on the timing at which the transmission unit instruction signal 36 is input. Thus, the data of the transmission unit is generated by dividing the code word of the arithmetic code 33 input from the encoding unit 31.

  Specifically, the transmission unit generator 35 in Embodiment 1 sequentially multiplexes the arithmetic code 33 of the encoded value 32 as transmission unit constituent bits (step S6) and is included in the transmission unit. Whether or not the encoding of data is completed by the number of possible macroblocks is determined by the transmission unit instruction signal 36 (step S8), and when it is determined that all the encodings within the transmission unit have not been completed. (Step S8 “N”), the process returns to Step S1 to perform the processing after the context model determination.

  On the other hand, when it is determined that all the encodings within the transmission unit have been completed (step S8 “Y”), the transmission unit generation unit 35 adds the following two pieces of information as header information of the next transmission unit data. (Step S9).

1. “Register reset flag” indicating whether or not to reset the register value indicating the arithmetic coding process for the codeword representation in the next transmission unit. In the first transmission unit to be generated, this register reset flag is set so as to always instruct to “reset”.

2. Above 1. The initial register value, which is the current register value to be used as the register value to be used at the start of arithmetic encoding and decoding of the next transmission unit only when the register reset flag of “1” indicates “not reset” ”. This initial register value is an initial register value 34 input from the encoding unit 31 to the transmission unit generation unit 35 as shown in FIG.

FIG. 8 shows a bit stream syntax of slice video compression data which is video compression data having a slice structure.
As shown in FIG. 8, for each slice video compressed data, the slice header data, which is the header of each slice video compressed data, includes the above-described 1. Register reset flag and 1. above. And an initial register value that is multiplexed only when the register reset flag indicates "no reset".

  According to the above two additional information, even when the previous slice is lost, the register reset flag included in the slice header data of itself and the value for register initialization of the initial register value are used. Even between slices, it is possible to perform coding while maintaining the continuity of arithmetic codewords, and it is possible to maintain coding efficiency.

  In FIG. 8, the slice header data and the slice video compressed data are multiplexed on the same stream. However, as shown in FIG. 9, the slice header data is transmitted offline in the form of another stream, and the slice video is transmitted. You may comprise so that ID information of corresponding slice header data may be attached to compressed data. The figure shows an example of transmitting a stream according to the IP protocol. An example of transmitting the header data portion with reliable TCP / IP and transmitting the video compression data portion with low-latency RTP / UDP / IP Is shown. According to the separated transmission format of the header and transmission unit according to the configuration of FIG. 9, the data transmitted by RTP / UDP / IP does not necessarily have to be divided into data units called slices. In a slice, it is basically necessary to reset all the dependencies (context model) with the video signal in the neighboring area so that decoding can be resumed independently in that slice, but this reduces the video coding efficiency. Invite. As shown in FIG. 11, if the initial register state can be transmitted by TCP / IP, the video signal itself is encoded using all context models in the frame, and the arithmetic code is used at the stage of RTP packetization. Divided data may be divided and transmitted. Therefore, according to this mechanism, since the arithmetic coding process can be stably acquired regardless of the line condition, a bit stream that has been coded without being restricted by the slice structure can be maintained with high error tolerance. It is possible to transmit.

  In addition, as shown in FIG. 10, whether to use the syntax of the register reset flag and the initial register value may be configured to be shown in a higher layer. In FIG. 10, an example in which a register reset flag and a register reset control flag indicating whether or not to use the syntax of an initial register value are multiplexed in header information given in units of a video sequence composed of a plurality of video frames. Show. For example, if the line quality is poor and it is determined that register transmission through the video sequence is more stable, the register reset control flag should be set to “Always reset the register at the beginning of the slice through the video sequence”. Set to a value indicating. At this time, the register reset flag and the initial register value to be multiplexed in units of slices do not need to be multiplexed at the slice level. As a result, when certain specific transmission conditions (such as line error rate) continue, if register reset control is performed in units of video sequences, overhead information transmitted in units of slices is reduced. be able to. Of course, the register reset control flag may be added to header information of an arbitrary video frame in the video sequence indicated by the Nth frame, the (N + 1) th frame, or the like.

  FIG. 11 shows the internal configuration of the arithmetic decoding unit 27. The arithmetic decoding unit 27 of the decoding device according to the first embodiment performs transmission unit decoding initialization for initializing arithmetic decoding processing for each received transmission unit based on additional information regarding the arithmetic coding process included in the header. The encoding unit 37 specifies the types of data to be decoded such as the motion vector 5, the encoding mode information 13, the spatial prediction mode 14, and the orthogonal transform coefficient data 17 based on the process of arithmetic decoding, A context model determination unit 28 that defines a commonly defined context model, a binarization unit 29 that generates a binarization rule determined based on the type of data to be decoded, and bin (0) according to the binarization rule and the context model. or an occurrence probability of 1), an occurrence probability generation unit 30 that performs arithmetic decoding based on the generated occurrence probability, and a binary sequence obtained as a result, And a reduction rule, a motion vector 5, the coding mode information 13, the decoding unit 38 for decoding the data, such as spatial prediction mode 14, the orthogonal transform coefficient data 17, and a city.

FIG. 12 shows an operation flow of the arithmetic decoding unit 27 shown in FIG.
The operation of the arithmetic decoding unit 27 shown in FIG. 11 will be described in detail with reference to FIG.

6) Transmission unit decoding initialization process (S10)
As shown in FIG. 8, based on a register reset flag indicating whether or not a register value indicating an arithmetic encoding process is reset and multiplexed on a transmission unit such as a slice, and an initial register value 34, a decoding unit The arithmetic decoding start state at 38 is initialized (step S10). When the register value is reset, the initial register value 34 is not used.

7) Context model determination process, binarization process, occurrence probability generation process These processes are performed by the context model determination unit 28, the binarization unit 29, and the occurrence probability generation unit 30 shown in FIG. Are the same as the context model determination process S1, the binarization process S2, and the occurrence probability generation process S3 shown in the processes 1) to 3) on the conversion apparatus side. And

8) Arithmetic decoding process (S11)
Since the occurrence probability of bins to be decoded is determined in the process up to 7), the decoding unit 38 restores the bin value according to the arithmetic decoding process shown in the conventional example (step S11), Similar to the processing on the encoding device side, the occurrence frequency of bin is updated by counting the occurrence frequency of 0/1 (step S5), and the bin bin decoded by comparing with the binary sequence pattern defined by the binarization rule It is determined whether or not the value is confirmed (step S12). If the bin value decoded in comparison with the binary sequence pattern defined by the binarization rule is not confirmed (step S12 "N"), again after the 0/1 occurrence probability generation process in each bin in step S3 Processing is performed (steps S3, S11, S5, S12). On the other hand, when the value of each decoded bin is confirmed by confirming the match with the binary sequence pattern determined by the binarization rule (step S12 “Y”), the data value indicated by the matched pattern is the decoded data. Output as a value, and if decoding has not been completed for all transmission units such as slices (step S13 "N"), in order to decode all transmission units, the processing after the context model determination processing in step S1 is repeated. .

  According to the video encoding / decoding device provided with the arithmetic coding and arithmetic decoding processes according to the first embodiment described above, even when video compressed data is transmitted divided into fine transmission units such as slices, as slice header data Since the register reset flag indicating whether or not the register value indicating the arithmetic encoding process is reset and the initial register value 34 are added, encoding can be performed without interrupting the continuity of the encoding process of arithmetic encoding. As a result, it is possible to maintain encoding efficiency while enhancing resistance to transmission errors, and it is possible to perform decoding thereof.

  In the first embodiment, a slice structure is assumed as a transmission unit, but the present invention can also be applied to a video frame as a transmission unit.

Embodiment 2. FIG.
In the second embodiment, another form of the arithmetic encoding unit 6 and the arithmetic decoding unit 27 will be described. In the second embodiment, not only the register value indicating the state of the code word in the arithmetic coding process but also the learning state for the occurrence probability variation in the context model, that is, the context by bin occurrence probability update processing in the occurrence probability generation unit 30. The learning state for the occurrence probability variation in the model is also multiplexed in the slice header.

  That is, for example, in FIG. 6 described in the first embodiment, in order to improve the efficiency of arithmetic coding of block C, for example, information on the motion vector of block B located at the upper part of block C is used as the occurrence probability variation. Use for decision. Therefore, if block C and block B are located in different slices, it is necessary to prohibit the use of block B information in the occurrence probability determination process. This means that the coding efficiency due to the occurrence probability adaptation by the context model is lowered. Therefore, the second embodiment provides a method and apparatus for improving the adaptability of this design. In the case where the probability of loss of slice data due to a transmission error or the like is extremely low, between the slices related to arithmetic coding. While making it possible to actively use without continually severing the dependency relationship, if there is a high possibility of losing slice data, the dependency relationship between slices can be severed in transmission units. The coding efficiency can be adaptively controlled.

  FIG. 13 shows an internal configuration of arithmetic coding unit 6 in the second embodiment. The arithmetic coding unit 6 according to the second embodiment is different from the arithmetic coding unit 6 according to the first embodiment shown in FIG. 3 in that the occurrence probability generation unit 30 is to be multiplexed into a slice header. The only difference is that the context model state 39 is transferred to the transmission unit generator 35.

  FIG. 14 shows an operation flow of the arithmetic coding unit 6 according to the second embodiment. As apparent from comparison with the operation flow of the arithmetic coding unit 6 of the first embodiment shown in FIG. 4, the difference is that the context model state 39 in the 0/1 occurrence probability generation process in each bin in step S3, that is, the occurrence probability. The learning state 39 for the occurrence probability variation in the context model by the bin occurrence probability update process in the generation unit 30 is also the next transmission in the transmission unit generation unit 35 in step S9, similarly to the register value in the binary arithmetic encoding process in step S4. It is only the point which is multiplexed on the slice header in the unit header configuration processing.

The meaning of the state 39 of the context model will be described with reference to FIG.
FIG. 15 shows a case where there are n macroblocks in the k-th transmission unit, and a context model ctx that is used only once for each macroblock is defined. The occurrence of ctx is generated for each macroblock. It shows how the probability varies. The meaning that the state 39 of the context model is succeeded to the next transmission unit means that the final state ctx ^k (n−1) of the kth transmission unit is ctx in the k + 1th transmission unit, as shown in FIG. The occurrence probability po, p1 of the value 0,1 at ctx ^{k + 1} (n-1) = 0,1,2 is the value 0,1 at ctx ^k (n-1) = 0,1,2 It means to be equal to the occurrence probability of po, p. For this purpose, the transmission unit generation unit 35 is configured to transmit data indicating the state of ctx ^k (n−1) as part of the header information in the (k + 1) th transmission unit.

  FIG. 16 shows an example of the bit stream syntax according to the second embodiment. As shown in the figure, in the second embodiment, in addition to the slice start code, the register reset flag, and the initial register value similar to the first embodiment shown in FIG. The information indicating the context model state of the immediately preceding slice is added. However, in the second embodiment, the register reset flag includes not only the presence / absence of multiplexing of the initial register value but also the meaning of whether or not the context model state data is multiplexed. Of course, the information indicating whether or not the context model state data is multiplexed may be configured to have another flag instead of the register reset flag. In addition, as described in the first embodiment, in the same figure, slice header data and slice video compression data are multiplexed on the same stream, but the slice header is transmitted offline in the form of another stream, You may comprise so that ID information of corresponding slice header data may be attached to compressed data.

  FIG. 17 shows the internal configuration of the arithmetic decoding unit 27 in the second embodiment. The arithmetic decoding unit 7 in the second embodiment shown in FIG. 17 is different from the arithmetic decoding unit 27 in the first embodiment shown in FIG. 11 in that a transmission unit decoding initialization unit 37 is multiplexed on a slice header. In other words, the state 39 of the context model of the immediately preceding slice is transferred to the occurrence probability generation unit 30, and the state of the context model is inherited from the immediately preceding slice.

  FIG. 18 shows an operation flow of the arithmetic decoding unit 27 in the second embodiment. As apparent from comparison with the operation flow of the arithmetic decoding unit 27 of the first embodiment shown in FIG. 12, the difference is that the context model state 39 decoded from the slice header in each transmission unit decoding initialization process of step S10 is the step. The process of S3, that is, the process of generating the 0/1 occurrence probability in each bin with reference to the context model determined in step S1, is output to the process of generating the 0/1 occurrence probability in the occurrence probability generation unit 30 It is a point used for.

  Note that the context model state passed in the slice header is the overhead of the slice header when the number of context models is extremely large, so select a context model that contributes significantly to the coding efficiency and change its state. You may comprise so that it may multiplex. For example, since motion vectors and orthogonal transform coefficient data occupy a large proportion of the total code amount, it may be configured to take over the state only for these context models. In addition, it is possible to explicitly multiplex the type of context model that inherits the state into the bitstream, and selectively carry out the state inheritance only for an important context model according to the local situation of the video. .

  According to the video encoding device / decoding device according to the second embodiment having the above-described configuration, when video compressed data is transmitted divided into fine transmission units, reset of a register value indicating an arithmetic encoding process as slice header data is performed. Since the register reset flag indicating presence / absence, initial register value 34, and information indicating the context model state of the immediately preceding slice are added, encoding is performed without interrupting the continuity of the encoding process of arithmetic encoding. Thus, it is possible to maintain the coding efficiency while enhancing the resistance to transmission errors. In this embodiment, a slice structure is assumed as a transmission unit. However, the present invention can also be applied to a video frame as a transmission unit.

  In particular, in Embodiment 2, since information indicating the context model state of the immediately preceding slice is added, for example, even if block C and block B immediately before this block C are positioned in different slices in FIG. By using the context model state of block B in the occurrence probability determination process of block C, it is possible to improve the coding efficiency by adaptation of the occurrence probability by the context model. In other words, in cases where the probability of loss of slice data is extremely low due to a transmission error or the like, it is possible to actively use the context model state of the immediately preceding slice without always severing the dependency relationship between slices related to arithmetic coding. On the other hand, if there is a high possibility of loss of slice data, the dependency between slices can be broken without using the context model state of the previous slice, and coding in transmission units is performed. The efficiency can be controlled adaptively.

  In the case of the second embodiment, as in the bit stream syntax shown in FIG. 16, for each slice data, each data of the immediately preceding slice is parallel to the addition of the register reset flag and the initial register value of the first embodiment. In the above description, information indicating the context model state of the first slice is added as slice header data. However, the register reset flag and the initial register value of the first embodiment are omitted without being added, and the context model state of each data of the immediately preceding slice is omitted. Only the information indicating the above may be added as slice header data, and the context model state reset flag regardless of whether it is provided in parallel with the addition of the register reset flag and the initial register value of the first embodiment. (See Fig. 19), this context model Status reset flag is off, i.e. course may be allowed to use in decoding by adding information indicating the context model status of each data immediately preceding slice only if not reset.

Embodiment 3 FIG.
In the third embodiment, an example will be described in which transmission units are configured in a data partitioning format for grouping by type of data to be encoded. Take, for example, the data partitioning disclosed in Draft Working Draft Number 2, Revision 3, JVT-B118r3, a video coding system specification draft that will be studied in the Joint Video Team (JVT) of ISO / IEC MPEG and ITU-T VCEG. 7 shows a method of transmitting a data unit formed by grouping specific types of data by the number of macroblocks existing in the slice structure as shown in FIG. 7 in the form of slice data. Has been. As data types of slice data, which is a data unit configured by grouping, for example, there are data types 0 to 7 shown below.

0 TYPE_HEADER Picture (frame) or slice header
1 TYPE_MBHEADER Macroblock header information (encoding mode information, etc.)
2 TYPE_MVD motion vector
3 TYPE_CBP CBP (Effective orthogonal transform coefficient distribution in macroblock)
4 TYPE_2x2DC orthogonal transform coefficient data (1)
5 TYPE_COEFF_Y Orthogonal transformation coefficient data (2)
6 TYPE_COEFF_C Orthogonal transformation coefficient data (3)
7 TYPE_EOS Stream end identification information

  For example, in a TYPE_MVD slice of data type 2, data obtained by collecting only motion vector information for the number of macroblocks in the slice is transmitted as slice data. Therefore, when decoding the TYPE_MVD data of the (k + 1) th slice following the TYPE_MVD data of the kth slice, only the state of the context model related to the motion vector at the end of the kth slice is If it is multiplexed in the header of the slice for sending the TYPE_MVD data of the (k + 1) th slice, the context model learning state for arithmetic coding of the motion vector can be taken over.

  FIG. 19 shows an example of the bitstream syntax when grouped by data type in the third embodiment. In FIG. 19, for example, when a motion vector that is a slice of TYPE_MVD of data type 2 is multiplexed as slice data, a slice start code, a data type ID indicating TYPE_MVD, and a context model state are included in the slice header. Information indicating the reset flag and the context model state for the motion vector of the immediately preceding slice is added.

  For example, when only orthogonal transform coefficient data (2) of orthogonal transform coefficient data (2) of TYPE_COEFF_Y of data type 5 is multiplexed as slice data, a slice start code or TYPE_COEFF_Y is indicated in the slice header. Information indicating the data type ID, the context model state reset flag, and the context model state for the orthogonal transformation coefficient data of the immediately preceding slice is added. In the figure, the slice header data and the compressed data are multiplexed on the same stream, but the slice header is transmitted off-line in a separate stream, and the compressed data includes the ID information of the corresponding slice header data. You may comprise so that it may attach.

  In addition, in the configuration of FIG. 13, the arithmetic encoding unit 6 according to the third embodiment is configured so that the transmission unit generation unit 35 reconstructs macroblock data in a slice according to the data partitioning rules, and each data type What is necessary is just to comprise so that the ID information showing the classification of this and the learning state of the context model corresponding to each data type may be multiplexed. In addition, the arithmetic decoding unit 27 in the present embodiment is used when the transmission unit decoding initializing unit 37 notifies the context model determining unit 28 of the data type classification ID multiplexed in the slice header in the configuration of FIG. What is necessary is just to comprise so that the context model learning state 39 may be succeeded between slices by determining the context model to perform, and notifying the occurrence probability generation part 30 of a context model learning state.

  According to the video encoding device and the video decoding device described in the third embodiment, even when compression encoding is performed by dividing a video signal into transmission units grouped by a predetermined data type, When performing arithmetic coding of the video signal to which it belongs, the encoding has been continued without resetting the symbol occurrence probability learning state in the transmission unit grouped with the previous predetermined data type. Even when grouped by type, it is possible to perform coding with improved coding efficiency of arithmetic coding while ensuring error tolerance.

  In the third embodiment, the data type classification for each slice structure is assumed as a transmission unit. However, the present invention can be applied even when transmission for each data type classification in a video frame unit is assumed.

  Further, in the example of the bit stream syntax of the third embodiment shown in FIG. 19, for each slice data for each data type, the context model state reset flag and each data of the immediately preceding slice when the flag is off The information indicating the context model state has been described as being added as slice header data. However, as in the example of the bit stream syntax of the second embodiment illustrated in FIG. 16, a register is provided for each slice data of each data type. In parallel with the addition of the reset flag and initial register value, the context model state reset flag and information indicating the context model state of each data of the immediately preceding slice when the flag is off may be added as slice header data. , Register reset flag and initial Regardless of whether or not it is provided in parallel with the addition of the register value, the context model state reset flag is omitted, and information indicating the context model state of each data of the immediately preceding slice is always added and used at the time of decoding. Of course it is good.

  In the first to third embodiments described above, video data has been described as an example of a digital signal. However, the present invention is not limited to this, and is not limited to this. The present invention can also be applied to a digital signal of a picture, a digital signal of a text, and a digital signal of multimedia data obtained by arbitrarily combining these.

  In the first and second embodiments, the transmission unit of the digital signal is described as a slice, and in the third embodiment, a predetermined transmission unit such as a data type partitioned by data type in the slice is described as an example. In the present invention, the present invention is not limited to this, and one image (picture) composed of a plurality of slices, that is, one video frame may be set as a predetermined transmission unit, or to a storage system other than communication. As a matter of course, a predetermined storage unit may be used instead of a predetermined transmission unit.

1 is a diagram illustrating a configuration of a video encoding device according to Embodiment 1. 1 is a diagram illustrating a configuration of a video decoding device according to Embodiment 1. It is drawing which showed the internal structure of the arithmetic coding part 6 in FIG. It is drawing explaining the operation | movement flow of the arithmetic encoding part 6 of FIG. It is drawing explaining the concept of a context model. 4 is a diagram illustrating an example of a motion vector context model in the first embodiment. It is drawing explaining a slice structure. 6 is a diagram illustrating an example of a bitstream generated by the arithmetic encoding unit 6 according to the first embodiment. 10 is a diagram illustrating an example of another bit stream generated by the arithmetic encoding unit 6 according to the first embodiment. 10 is a diagram illustrating still another example of a bitstream generated by the arithmetic encoding unit 6 according to the first embodiment. 3 is a diagram illustrating an internal configuration of an arithmetic decoding unit 27 in the first embodiment. It is drawing explaining the operation | movement flow of the arithmetic decoding part 27 of FIG. 6 is a diagram illustrating an internal configuration of an arithmetic encoding unit 6 according to Embodiment 2. It is drawing explaining the operation | movement flow of the arithmetic coding part 6 of FIG. It is drawing explaining the learning state of a context model. 10 is a diagram illustrating an example of a bitstream generated by the arithmetic encoding unit 6 according to the second embodiment. 6 is a diagram illustrating an internal configuration of an arithmetic decoding unit 27 in the second embodiment. 18 is a diagram illustrating an operation flow of the arithmetic decoding unit 27 in FIG. 17. 10 is a diagram illustrating an example of a bitstream generated by the arithmetic encoding unit 6 according to the third embodiment. It is a figure which shows the occurrence probability of each character when the character "BILL GATES" is arithmetically encoded. It is a figure which shows the arithmetic coding result at the time of carrying out arithmetic coding of the character "BILL GATES".

Explanation of symbols

  2 motion detection unit, 3a, 3b frame memory, 6 arithmetic coding unit, 7 motion compensation unit, 10a, 10b spatial prediction unit, 12 coding mode determination unit, 15 orthogonal transform unit, 16 quantization unit, 18 inverse quantization 19, an inverse orthogonal transform unit, 22 an encoding control unit, 24 a transmission buffer, 51 a subtracter, 52 a switching unit, 53 an adder, a switching unit 54, and an adder 55.

Claims (2)

In a digital signal decoding apparatus that receives and decodes a compression-coded digital signal in a predetermined unit,
An arithmetic decoding unit configured to decode the predetermined unit of the compressed digital signal based on an arithmetic encoding process by determining a context model to be used for each type of data to be decoded from among a plurality of context models;
The arithmetic decoding unit generates the context model occurrence probability based on the information representing the initial state of the context model occurrence probability that is multiplexed as a part of a predetermined unit of data and is associated with the type of data to be decoded. Initializing the state, decoding the initialization information of the register value indicating the arithmetic decoding process, initializing the register value based on the initialization information of the register value, the initialized occurrence probability state and the A digital signal decoding apparatus, wherein the predetermined unit data is decoded from an initialized register value. A digital signal arithmetic decoding method for receiving and decoding a compression-coded digital signal in a predetermined unit,
The predetermined unit of the compressed digital signal is decoded based on an arithmetic coding process by defining a context model to be used for each type of data to be decoded from among a plurality of context models,
The context model occurrence probability state is initialized based on information representing the initial state of the context model occurrence probability that is multiplexed as part of a predetermined unit of data and is associated with the type of data to be decoded. The register value initialization information indicating the arithmetic decoding process is decoded, the register value is initialized based on the register value initialization information, the initialized occurrence probability state and the initialized register value The digital signal arithmetic decoding method, wherein the predetermined unit data is decoded.

Images