JP5047244B2 - Context adaptive entropy encoding method and apparatus, context adaptive entropy decoding method and apparatus, and program thereof - Google Patents

Description

  The present invention relates to a context adaptive entropy encoding / decoding technique in high-efficiency image signal encoding.

  H. FIG. 12 shows an example of the configuration of a moving picture coding apparatus corresponding to a typical moving picture coding system such as H.264 / AVC.

  In this moving image encoding apparatus, for example, Y, Cb, and Cr each input an YCbCr format image signal of 10 bits / channel. When the encoding target image signal is input, the subtracter 110 calculates the difference between the input image signal (predicted signal) and the prediction signal. The transform / quantization unit 111 performs orthogonal transform such as DCT transform and quantization on the transform coefficient on the calculated difference signal. The quantized transform coefficient is output to the entropy coding unit 119 and also sent to the inverse quantization / inverse transform unit 112.

  In the inverse quantization / inverse transform unit 112, inverse quantization and inverse transform are performed on the quantized transform coefficient, and the output and the prediction signal are added by the adder 113 to obtain a decoded signal. The decoded signal is subjected to filter processing for reducing block noise by the deblocking filter 114, and the resulting signal is stored in the frame memory 115 as a reference signal in later inter prediction (interframe prediction).

  In the intra prediction unit 116, a prediction signal is generated by intra-frame prediction that does not refer to other encoded frames. On the other hand, the inter prediction unit 118 generates a prediction signal using another encoded / decoded signal stored in the frame memory 115 in accordance with the motion vector detected by the motion detection unit 117.

  The entropy encoding unit 119 entropy encodes the quantized transform coefficient, motion vector, and other encoded information, and outputs an encoded stream. The encoding control unit 100 controls the amount of generated code and other overall encoding.

  FIG. 13 shows a configuration example of a conventional typical moving picture decoding apparatus. This moving picture decoding apparatus receives the encoded stream encoded by the moving picture encoding apparatus shown in FIG. 12 and decodes the input signal by the entropy decoding unit 200. The decoded quantized transform coefficient is subjected to inverse transform processing such as inverse quantization and inverse DCT transform by the inverse quantization / inverse transform unit 211, and the prediction residual signal of the transform result is output to the adder 214.

  Further, the prediction mode information decoded by the entropy decoding unit 200 is output to the intra prediction unit 212 and the inter prediction unit 213, and in the case of intra prediction, a prediction signal based on intra-frame prediction is generated by the intra prediction unit 212, and the inter prediction unit 212 generates inter prediction. In the case of prediction, the inter prediction unit 213 generates a prediction signal for inter-frame prediction with reference to the already decoded decoded image stored in the frame memory 216.

  The prediction signal output from the intra prediction unit 212 or the inter prediction unit 213 is output to the adder 214 and is added to the prediction residual signal output from the inverse quantization / inverse conversion unit 211 by the adder 214. The output signal of the adder 214 is subjected to filter processing for reducing block noise by the deblocking filter 215, and the result is output as a decoded image. The decoded image is stored in the frame memory 216 as a reference image in the subsequent inter-frame prediction decoding. The decoding control unit 210 performs overall decoding control.

  In the moving picture encoding apparatus as shown in FIG. 12, various encoding symbols such as quantized DCT coefficients and motion vectors (which are called syntax elements) are entropy encoded by the entropy encoding unit 119. Processing is performed.

  In the moving picture decoding apparatus as shown in FIG. 13, the encoded data encoded by the entropy encoding unit 119 of the moving picture encoding apparatus is decoded by the entropy decoding unit 200 into a syntax element.

H. H.264 / AVC employs a method called CABAC (see Non-Patent Document 1) as the entropy encoding process. CABAC (Context-Adaptive Binary Arithmetic Coding) is a method classified as a context-adaptive arithmetic code, and includes the following three processes.
(1) binarization: a process of converting a multilevel syntax element into a binary signal.
(2) context modeling: a process of selecting a probability model (context table) of a symbol to be encoded for encoding a binary signal of a syntax element in accordance with an encoded binary signal. The context table is updated using the encoded binary signal.
(3) binary arithmetic coding: a process of performing binary arithmetic coding on a binary signal using a selected context table.

  The context table models conditional occurrence probabilities with surrounding symbols for the encoding target symbol. The state determined by the surrounding symbols is called a context, and the context is further subdivided (called context classification), and an appropriate probability model is assigned to each context to improve coding efficiency. Yes.

  FIG. 14 shows a configuration example of CABAC. The CABAC 300 includes a binarization unit 301 that converts a multilevel signal, which is a syntax element, into a binary signal, and a context calculation unit 302 that calculates and updates the occurrence probability of the binary signal to be encoded according to the surrounding situation. And a binary arithmetic encoding unit 303 that arithmetically encodes the binary signal based on the binary signal occurrence probability given from the context calculation unit 302.

  The context calculation unit 302 holds a plurality of binary signal generation probabilities, and switches the generation probabilities according to the current encoding target and the surrounding situation to give to the binary arithmetic encoding unit 303. The occurrence probability of the binary signal is one of symbols 0 and 1 (MPS: Most Probable Symbol) having a high occurrence probability and its occurrence probability table (a table having the occurrence probability of MPS. When a number is specified, the corresponding occurrence probability is Obtained) number pStateIdx (see FIG. 14B).

  The MPS occurrence probability corresponding to the occurrence probability table number pStateIdx corresponds to a higher MPS occurrence probability as the value of pStateIdx increases.

  In CABAC, the context calculation unit 302 adaptively switches the occurrence probability of a binary signal by updating pStateIdx for each encoding operation. FIG. 14C shows a table of update values of occurrence probability table numbers. In this table, the updated pStateIdx value when MPS (Most Probable Symbol: 0, 1 which has a higher probability of occurrence) is encoded is transIdxMPS, LPS (Least Probable Symbol: 0, 1). The value of pStateIdx after updating when the symbol with the lower probability of occurrence is encoded as transIdxLPS is shown. According to this table, the occurrence probability table number pStateIdx is updated and the occurrence probability table is switched for each encoding operation depending on whether the MPS is encoded or the LPS is encoded.

  This occurrence probability table number pStateIdx, current MPS, encoding state, and other information are set as one set and switched for each syntax element to be encoded such as macroblock (MB) type, CBP, BCBP, MAP, MVD, etc. Thus, entropy encoding by CABAC is performed.

  A state determined by its surrounding symbols for a coding target symbol is called a context. Information on a probability model obtained by modeling the conditional occurrence probability of such a coding symbol (in CABAC, occurrence probability table number pStateIdx) And the occurrence probability table, information on the current MPS, etc.) will be described here as a context table.

D. Marpe, H. Schwarz, and T. Wiegand, “Context-Based Adaptive Binary Arithmetic Coding in the H.264 / AVC Video Compression Standard”, IEEE trans. CSVT, Vol. 13, No. 7, pp. 620- 636, July 2003.

  The CABAC initializes the context table in units called slices, which are sets of macroblocks (hereinafter also referred to as MB). In the prior art, the initial value of the context table at the start of slice encoding is fixed regardless of the input image. However, the signal in the slice is not necessarily a signal having similar statistical properties. For example, when a moving object region and a background region coexist, both have statistically different properties.

  Originally, signals with different statistical properties can be further classified into different contexts to further improve coding efficiency. That is, under the condition where signals having different statistical properties are mixed as described above, there is still room for improvement in improving coding efficiency by context classification.

  Therefore, in the present invention, the initial value of the context table at the start of slice encoding is adaptively changed according to the input image. By such adaptive processing, the code length generated by CABAC is shortened, so that the effect of improving the coding efficiency can be expected.

  The present invention has been made in view of such circumstances, and classifies MBs into sets (called classes) having similar properties, and completes encoding of reference MBs from MBs belonging to the same class (called reference MBs). It is an object of the present invention to establish an adaptive entropy encoding method that inherits information in a context table as context table information used for initialization. In particular, the present invention is efficient for inheriting context table information for each class. It proposes a good concrete mechanism.

  In the present invention, macroblocks (MB) are classified into sets (called classes) having similar properties in the space-time domain. This classification may be performed for each encoded symbol of the macroblock. Note that the number M of classes to be classified is a value determined in advance between the encoder and the decoder. When a value other than this value is used, it is transmitted to the decoder as additional information of encoded data. Here, to classify into a set with similar properties means to classify according to some predetermined classification criteria such as the same statistical properties of the image or a specific region where the position in the image is. .

  Information used for updating the context table is inherited from a macro block (reference macro block) belonging to the same class as the target macro block. The reference macroblock is identified based on the frame number, slice number, and MB number. As a method of identifying a reference macroblock, for example, a method of using information specifying a frame number, slice number, and MB number as additional information for encoding, a method of using reference frame information or reference region information in inter-frame prediction, Alternatively, a method of combining these can be used. When the reference region in inter-frame prediction extends over a plurality of macro blocks, the macro block that occupies the maximum area may be set as the reference macro block.

  The context table must be stored so that it can be referenced in subsequent macroblocks. Therefore, in order to perform this storage, in the present invention, a buffer called a context information buffer (CIB) is prepared. When switching the reference in units of macroblocks, the context table in the macroblock is stored in the context information buffer CIB for each macroblock. When switching the reference in units of slices, the context table in the specified macroblock in the slice is stored in the context information buffer CIB for each slice. Here, the position of the designated macroblock can be fixed at a predetermined position, for example, fixed to the last macroblock in the slice.

  H. In H.264 / AVC, a decoded picture buffer (DPB: Decoded Picture Buffer) is prepared as an image buffer for storing a decoded picture used for inter-frame prediction (frame memory 115 in FIG. 12 and frame memory 216 in FIG. 13). A reference image in inter-frame prediction is stored in the decoded image buffer DPB. Data stored in the decoded picture buffer DPB is classified into a long-time reference picture, a short-time reference picture, and a non-reference picture. The long-time reference picture and the short-time reference picture are stored in the reference picture memory in the decoded picture buffer DPB. The non-reference picture is stored in the output picture memory in the decoded image buffer DPB.

Therefore, the CIB memory is managed in units of pictures, and the memory management is performed in synchronization with the decoded image buffer DPB. There are the following two implementation methods.
[CIB Management Method 1] A memory area for CIB is separately secured in the decoded image buffer DPB.
[CIB Management Method 2] In addition to the decoded image buffer DPB, a memory area for the CIB is secured, and the timing of memory management is synchronized.

  In the former case, no new processing is required for synchronization of memory management. However, the decoding of the decoded image buffer DPB is the same as the conventional H.264 standard. H.264 / AVC is not possible, and backward compatibility with respect to the decoded image buffer DPB cannot be maintained.

  In the latter case, the content of the decoded image buffer DPB is unchanged, so H.264 / AVC guarantees backward compatibility for the decoded image buffer DPB. However, new processing is required for the synchronization of memory management of the decoded image buffers DPB and CIB.

  1A and 1B are diagrams for explaining an example of a management method for a context information buffer CIB. FIG. 1A shows an example of a CIB management method 1, and FIG. 1B shows an example of a CIB management method 2. Yes.

[Details of CIB Management Method 1]
In the CIB management method 1, as shown in FIG. 1A, a memory area for CIB is separately secured in the decoded image buffer DPB. When the decoded image data is stored in the long-time reference picture memory / short-time reference picture memory, the CIB information for the same picture is also stored. If the area where the decoded image is stored is a long-time reference picture memory, the CIB data is also stored in the long-time reference CIB memory provided corresponding to the long-time reference picture memory, and the decoded image is stored. If the area is a short-time reference picture memory, the CIB data is also stored in the short-time reference CIB memory provided corresponding to the short-time reference picture memory. In addition, when data is released to the long-time reference picture memory / short-time reference picture memory, the CIB information for the same picture is also released. Note that the output picture memory is not used for CIB. In this case, there is no need to add a new process for the CIB memory management.

[Details of CIB Management Method 2]
In the CIB management method 2, as shown in FIG. 1B, a CIB memory area is secured separately from the decoded image buffer DPB, and the timing of memory reservation / release is synchronized between the decoded image buffers DPB and CIB. Take. Two types of CIB memory areas are prepared: a long-time reference CIB memory and a short-time reference CIB memory.

  The long-term reference CIB memory is subjected to memory reservation / release processing in conjunction with the reservation / release of the long-time reference picture memory in the decoded image buffer DPB. On the other hand, in the short-time reference CIB memory, memory reservation / release processing is performed in conjunction with the reservation / release of the short-time reference picture memory in the decoded image buffer DPB. These memory allocation / release methods are given from the outside. For example, a FIFO (First In, First Out) that opens in the order in which it is stored can be mentioned.

  In this case, since the content of the decoded image buffer DPB is unchanged, the content of the decoded image buffer DPB is the same as the conventional H.264 format. It can be used as it is in the H.264 / AVC decoder.

  Sets the maximum capacity of the context information buffer CIB. For example, H.C. In the case of H.264 / AVC, the capability of the decoder and the complexity of the encoded stream are defined as “level”. Therefore, the maximum capacity of the context information buffer CIB is also H.264. When applied to H.264 / AVC, the maximum capacity is changed according to the level.

  That is, the features of the present invention are as follows. In the present invention, for a symbol to be encoded, a state determined by its peripheral symbols is called a context, and the context is subdivided (referred to as context classification). In entropy coding that performs adaptive coding processing by assigning a probability model that models the conditional occurrence probability of the symbol to be coded based on the symbol information, each block (symbol) is classified into a set called a class Then, the probability model is updated using the encoded information for each class, and the updated probability model is stored in the memory of the context information buffer CIB as information that can be referred to in subsequent processing. The context information buffer CIB is provided in association with a decoded image buffer DPB for storing a decoded image used for inter-frame prediction.

  Memory management in the context information buffer CIB for storing the probability model information, for example, management such as memory allocation, buffer writing, buffer movement, and buffer release, is performed in synchronization with the decoded image buffer DPB.

  When memory management in the context information buffer CIB is performed on a frame basis, information in the decoded image buffer DPB is associated with information in the context information buffer CIB using the frame number. Further, when rearranging the memories in the context information buffer CIB in units of frames, the same frame number is used so that the information in the frames in the context information buffer CIB is in the same order as the information of each frame in the decoded image buffer DPB. Move.

  The context information buffer CIB may be provided in the decoded image buffer DPB. In this case, a memory area for CIB is separately set in the decoded image buffer DPB, and the memory area for CIB is secured / released in accordance with the timing of securing / releasing the memory of the decoded image buffer DPB. In addition, when the context information buffer CIB is provided separately from the decoded image buffer DPB, a memory area for the CIB is secured separately from the decoded image buffer DPB, and the memory of the decoded image buffer DPB is secured / released, The memory area for CIB is secured / released.

  When managing the CIB memory for each picture, the probability model (context table) of the encoding target symbol at the end of encoding of each macroblock is written to the CIB memory area in the short-time reference picture memory, and the macroblock The context table can be referred to in units.

  In addition, when managing the CIB memory area for each picture, the position of the reference macroblock in the slice given from the outside is read, and the probability model of the encoding target symbol at the end of encoding of the reference macroblock in the slice ( It is also possible to write the context table) to the CIB memory area in the short-time reference picture memory and refer to the context table in units of slices.

  Also in the case of entropy decoding in the video decoding device, using the context information buffer CIB as in the case of the entropy encoding, referring to the probability model (context table) of the same class as the block to be decoded (symbol), Perform entropy decoding.

  According to the present invention, since the context table is updated for a class composed of macroblocks having the same statistical properties, adaptive processing specialized for signals within the class becomes possible, and a reduction in code amount can be expected. .

It is a figure explaining the example of the management method of context information buffer CIB. It is a block diagram of the context adaptive entropy encoding apparatus which shows the principal part which concerns on one Embodiment of this invention. It is a flowchart of the encoding process by a context adaptive entropy encoding apparatus. It is a block diagram of the context adaptive entropy decoding apparatus which shows the principal part which concerns on one Embodiment of this invention. It is a flowchart of the decoding process by a context adaptive entropy decoding apparatus. It is a flowchart which shows the process of the writing method 1 to CIB. It is a flowchart which shows the process of the writing method 2 to CIB. It is a flowchart of the CIB management process in the case of FIFO. It is a flowchart of a CIB management process when a memory management command is used. It is a figure which shows the hardware structural example when implement | achieving with a software program. It is a figure for demonstrating the effect by embodiment of this invention. It is a figure which shows the structural example of the conventional moving image encoder. It is a figure which shows the structural example of the conventional moving image decoding apparatus. It is explanatory drawing of the conventional CABAC.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 2 is a configuration diagram of a context adaptive entropy encoding device showing a main part according to an embodiment of the present invention.

  The context adaptive entropy encoding device 1 includes a class classification information storage unit 10, a context table initial value setting unit 11, a context calculation unit 12, a binarization unit 14, a binary arithmetic encoding unit 15, and a context information buffer (CIB) 16. Is provided. The context calculation unit 12 includes a context table selection unit 13 that selects a probability model (context table) used in the binary arithmetic coding unit 15.

  The decoded image buffer (DPB) 17 corresponds to the frame memory 115 described with reference to FIG. 12, and stores a local decoded image in the moving image encoding device. As described in FIG. 1, the context information buffer 16 may be provided separately from the decoded image buffer 17 or may be provided in the decoded image buffer 17.

  The context table management processing unit 18 is processing means for managing the context information buffer 16 in synchronization with the management of the decoded image buffer 17. Here, management refers to reading information, moving and writing in the buffer, discarding information that is no longer needed, and the context table management processing unit 18 links the context information buffer 16 and the decoded image buffer 17 together. Buffer management. In buffer management of the context information buffer 16, the information in the decoded image buffer 17 is used. In this diversion, an element in the decoded image buffer 17 is associated with an element in the context information buffer 16 by using, for example, a frame number. When the data in the context table group corresponding to each frame in the context information buffer 16 is rearranged, the procedure performed for each reference frame in the decoded image buffer 17 is followed.

  The context adaptive entropy encoding device 1 shown in FIG. This is used when entropy-encoding various encoded symbols such as quantized DCT coefficients and motion vectors in a moving image encoding apparatus such as H.264 / AVC. Here, as an example, a processing example in which the present invention is incorporated will be described mainly based on CABAC. However, the object of the present invention is not limited to CABAC, and any other system can be used as long as it is context adaptive coding. Is equally applicable.

  Similarly to the binarization unit 301 in the conventional CABAC 300 shown in FIG. 14, the binarization unit 14 performs processing for converting a multilevel signal of a syntax element to be encoded into a binary signal. If the signal to be encoded is not a multilevel signal, the binarization unit 14 may be omitted.

  The binary arithmetic encoding unit 15 arithmetically encodes the binary signal output from the binarizing unit 14 according to the binary signal occurrence probability given from the context calculation unit 12, and outputs the encoded bit of the encoded stream. To do. The binary arithmetic encoding unit 15 can also be configured by the same one as the conventional binary arithmetic encoding unit 303 shown in FIG.

  The context calculation unit 12 also switches the binary signal generation probability by the context table selection unit 13 according to the current encoding target and surrounding conditions, and gives the binary signal generation probability to the binary arithmetic encoding unit 15 The point of performing is basically the same as in the prior art. However, it differs from the prior art in that the context table initialized by the context table initial value setting unit 11 is used for each coding unit such as at the start of slice coding.

  The class classification information storage unit 10 stores information about which class each encoded macroblock for a plurality of frames or slices is classified into.

  The context table initial value setting unit 11 determines the encoding target macroblock at the start of encoding at the start of encoding of a certain encoding unit, for example, a slice, according to the class classification information stored in the class classification information storage unit 10. Information of the context table of the reference macroblock belonging to the same class is acquired from the context information buffer 16, and the context table used by the context calculation unit 12 is initialized based on the information. For example, in the case of CABAC, the occurrence probability table number pStateIdx is set to the occurrence probability table number pStateIdx updated by the reference macroblock encoding result as an initial value.

  As a result, the binary arithmetic encoding unit 15 performs adaptive encoding specialized for signals in the class, and the encoding efficiency is improved.

  Next, the processing flow of the context adaptive entropy encoding device 1 will be described with reference to the flowchart shown in FIG. Here, the flow of processing when the context table is initialized at the start of slice encoding will be described. However, the context table can be initialized at the start of another encoding unit such as a picture. Also, the context table may be initialized when the class of the encoding target macroblock is changed to another class.

  [Step S10]: The following steps S11 to S110 are repeated for all slices.

  [Step S11]: For all macroblocks in the encoding target slice, a syntax element to be subjected to context adaptive entropy encoding is read. This is, for example, DCT coefficients, macroblock mode information, motion vectors, and the like.

  [Step S12]: For each syntax element, specify the context table information of the syntax element in the encoded slice used for updating the context table. Here, updating the context table means initializing the context table, and specifying the context table information means specifying the context table information of the corresponding class stored in the context information buffer (CIB) 16. Means. The corresponding class is determined with reference to the class classification information storage unit 10. In the conventional CABAC, the context table is initialized by using a fixed table for each syntax element. In the present apparatus, the context table is initialized by using the context table information tuned to the class. .

  [Step S13]: Next, the following steps S14 to S19 are repeated for every macroblock (MB) in the current slice.

  [Step S14]: The binarization unit 14 converts the multi-value syntax element into a binary signal. This binary signal is hereinafter referred to as a binary symbol.

  [Step S15]: For a binary symbol, a context table is selected by using the same type of binary symbol information of the encoded peripheral macroblock. When CABAC is used, the CABAC regulations shown in Non-Patent Document 1 are followed.

  [Step S16]: The context table is updated using encoded binary symbols belonging to the same class. When using CABAC, follow the CABAC regulations. In the case of CABAC, the occurrence probability table number is updated according to the update value of the occurrence probability table number shown in FIG. At the head of the slice, the context table is updated using the context table information specified in step S12.

  [Step S17]: The binary symbol to be encoded is entropy-encoded by the binary arithmetic encoding unit 15 using the updated context table.

  [Step S18]: The updated context table information is written and stored in the working memory.

  [Steps S19, S110]: The above processing is repeated for all MBs in the slice, and the processing is repeated in the same manner for all slices.

  [Step S111]: The information in the context table stored in step S18 is written to the memory corresponding to the corresponding picture in the context information buffer 16. Details of this processing will be described later with reference to FIG. 6 or FIG.

  FIG. 4 is a configuration diagram of a context adaptive entropy decoding device showing the main part according to an embodiment of the present invention.

  The context adaptive entropy decoding device 2 includes a class classification information storage unit 20, a context table initial value setting unit 21, a context calculation unit 22, a binary arithmetic decoding unit 24, a syntax element conversion unit 25, and a context information buffer (CIB) 26. Is provided. The context calculation unit 22 includes a context table selection unit 23 that selects a probability model (context table) used in the binary arithmetic decoding unit 24.

  This context adaptive entropy decoding device 2 is an H.264 standard. It is used when decoding a moving image encoded stream in a moving image decoding apparatus such as H.264 / AVC. Here, as in the case of encoding, a processing example based on CABAC will be described. However, if it is context adaptive decoding, it can be similarly applied to other systems.

  The binary arithmetic decoding unit 24 arithmetically decodes the encoded data to be decoded according to the binary signal occurrence probability given from the context calculation unit 22, and outputs a binary signal (binary symbol) as a result of decoding. The syntax element conversion unit 25 converts a binary signal obtained by arithmetic decoding into a multi-value syntax element. These are the same as the decoding method in the conventional CABAC.

  The context calculation unit 22 also performs a process of switching the generation probability according to the context table selected by the context table selection unit 23 according to the current decoding target and the surrounding situation, and giving the binary arithmetic decoding unit 24 the binary signal generation probability. The points to be performed are basically the same as those of the prior art. However, it differs from the prior art in that the context table initialized by the context table initial value setting unit 21 is used for each decoding unit such as when slice decoding starts.

  The class classification information storage unit 20 stores information about which class each decoded macroblock is classified into, corresponding to the decoded image stored in the decoded image buffer 27.

  The context table initial value setting unit 21 refers to the class classification information storage unit 20 at the start of decoding of a certain decoding unit, for example, a slice, and a context table of reference macroblocks belonging to the same class as the decoding target macroblock at the start of decoding. Is obtained from the context information buffer 26, and the context table used by the context calculation unit 22 is initialized based on the information. For example, in the case of CABAC, the occurrence probability table number pStateIdx is set with the occurrence probability table number pStateIdx updated by the decoding result of the reference macroblock as an initial value.

  The functions of the context information buffer (CIB) 26, the decoded image buffer (DPB) 27, and the context table management processing unit 28 are the context information buffer (CIB) 16, the decoded image buffer (DPB) 17, and the context table described with reference to FIG. The function is the same as that of the management processing unit 18.

  Next, the processing flow of the context adaptive entropy decoding device 2 will be described with reference to the flowchart shown in FIG. Here, the flow of processing when the context table is initialized at the start of slice decoding will be described. However, the context table can be initialized at the start of another decoding unit as well. Also, the context table may be initialized when the class of the macroblock to be decoded is changed to another class. However, this initialization unit needs to match the processing at the time of encoding.

  [Step S20]: The following steps S21 to S210 are repeated for all slices.

  [Step S21]: For all macroblocks in the decoding target slice, the encoded data of the syntax element to be subjected to context adaptive entropy decoding is read. This is, for example, data obtained by entropy encoding DCT coefficients, macroblock mode information, motion vectors, and the like.

  [Step S22]: For each syntax element, specify the context table information of the syntax element in the decoded slice to be used for updating the context table. Here, updating the context table means initializing the context table, and specifying the context table information means specifying the context table information of the corresponding class stored in the context information buffer 26. . The corresponding class is determined with reference to the class classification information storage unit 20. In the conventional CABAC, the context table is initialized by using a fixed table for each syntax element. In the present apparatus, the context table is initialized by using the context table information tuned to the class. .

  [Step S23]: Next, the following steps S24 to S29 are repeated for every macroblock (MB) in the current slice.

  [Step S24]: A context table is selected by using the same kind of binary symbol information of the decoded neighboring macroblocks. When CABAC is used, the CABAC regulations shown in Non-Patent Document 1 are followed.

  [Step S25]: The context table is updated using decoded binary symbols belonging to the same class. When using CABAC, follow the CABAC regulations. At the head of the slice, this context table is updated using the context table information specified in step S22.

  [Step S26]: The binary symbol to be decoded is entropy-decoded by the binary arithmetic decoding unit 24 using the updated context table.

  [Step S27]: The decoded binary symbol is converted into a multi-value syntax element by the syntax element conversion unit 25 and output.

  [Step S28]: The updated context table information is written and stored in the working memory.

  [Steps S29, S210]: The above processing is repeated for all MBs in the slice, and the processing is repeated in the same manner for all slices.

  [Step S211]: The information in the context table stored in step S28 is written to the memory corresponding to the picture in the context information buffer 26. Details of this processing will be described later with reference to FIG. 6 or FIG.

  Next, detailed processing in step S111 shown in FIG. 3 and step S211 shown in FIG. 5 will be described. As an example of a method for writing a context table to the CIB, as shown in FIG. 6, a method of writing in units of macro blocks (CIB writing method 1) and a method of writing in units of slices (to CIB) as shown in FIG. There is a writing method 2).

[CIB export method 1 (export in macroblock units)]
FIG. 6 is a flowchart showing the process of the writing method 1 to the CIB.

  [Step S30]: It is determined whether the storage location of the decoded image of the picture is the short-time reference picture memory. If the determination result is true, the process proceeds to step S31. If the determination result is false, the process proceeds to step S32.

  [Step S31] Repeat the process of writing the probability model (context table) of the encoding target symbol at the end of encoding of each macroblock to the CIB area in the short-time reference picture memory for all macroblocks in the picture. If the process is repeated until the last macroblock, the writing process is terminated.

  [Step S32]: It is determined whether the storage location of the decoded image of the picture is the long-time reference picture memory. If the determination result is true, the process proceeds to step S33. If the determination result is false, that is, if the decoded image is stored in the non-reference picture memory, the process ends without performing the writing process to the CIB.

  [Step S33] Repeat the process of writing the probability model (context table) of the symbol to be encoded at the end of encoding of each macroblock to the CIB area in the long-time reference picture memory for all macroblocks in the picture. If the process is repeated until the last macroblock, the writing process is terminated.

[CIB export method 2 (slice unit)]
FIG. 7 is a flowchart showing the processing of the CIB writing method 2.

  [Step S40]: The reference macroblock position in the slice is read.

  [Step S41]: It is determined whether the storage location of the decoded image of the picture is the short-time reference picture memory. If the determination result is true, the process proceeds to step S42, and if the determination result is false, the process proceeds to step S43.

  [Step S42] Processing for writing the probability model (context table) of the encoding target symbol at the end of encoding of the reference macroblock in the slice to the CIB area in the short-time reference picture memory for all slices in the picture When the process is repeated until the last slice, the writing process is terminated.

  [Step S43]: It is determined whether the storage location of the decoded image of the picture is the long-time reference picture memory. If the determination result is true, the process proceeds to step S44. If the determination result is false, that is, if the decoded image is stored in the non-reference picture memory, the process ends without performing the writing process to the CIB.

  [Step S44] Processing for writing the probability model (context table) of the encoding target symbol at the end of encoding of the reference macroblock in the slice to the CIB area in the long-time reference picture memory for all slices in the picture When the process is repeated until the last slice, the writing process is terminated.

  Next, the flow of management processing performed by the context table management processing unit 18 shown in FIG. 2 and the context management table management processing unit 28 shown in FIG. 4 will be described. As an example of buffer management, an example in which the FIFO is released in the order in which data is stored will be described first, and then an example in which a memory management command is used will be described.

[Flow of CIB management processing (in the case of FIFO)]
FIG. 8 is a flowchart of CIB management processing in the case of FIFO.

  [Step S50]: The decoded image buffer DPB is read.

  [Step S51]: Next, the stored number N of short-time reference pictures is read.

  [Step S52]: Further, the current number n of stored decoded images is read.

  [Step S53]: It is determined whether or not the current stored number n of decoded images is smaller than the stored number N of short-time reference pictures. If the determination result is true, the process proceeds to step S54. Proceed to S57.

  [Step S54]: The index (short-term reference picture index) for identifying each picture data in the short-term reference picture memory in the decoded image buffer DPB is increased by one.

  [Step S55]: The short-time reference picture index is set to 1, and the current decoded image is stored in the short-time reference picture memory.

  [Step S56]: The context table of the same picture is stored in the CIB area (short-time reference CIB memory) in the short-time reference picture memory, and the process ends.

  [Step S57]: When the current stored number n is N or more, the decoded image data in the short-time reference picture memory to which N is given as the short-time reference picture index is released and stored in the non-reference picture memory.

  [Step S58]: The data in the context table of the CIB area (short-time reference CIB memory) in the short-time reference picture memory to which N is given as the short-time reference picture index is released.

  [Step S59]: The index for identifying each picture data in the short-time reference picture memory (short-time reference picture index) is increased by one.

  [Step S510]: The short-time reference picture index is set to 1, and the current decoded image is stored in the short-time reference picture memory.

  [Step S511]: The context table of the same picture is stored in the CIB area (short-time reference CIB memory) in the short-time reference picture memory, and the process is terminated.

[Flow of CIB management processing (when memory management command is used)]
FIG. 9 is a flowchart of CIB management processing when a memory management command is used.

  [Step S60]: The decoded image buffer DPB is read.

  [Step S61]: Next, a control signal related to the decoded image data in the decoded image buffer DPB is read.

  [Step S62]: The following processing is performed according to the meaning of the control signal.

  When the meaning of the control signal is an instruction to release the designated short-time reference picture memory: The designated short-time reference picture memory is released and changed to a non-reference picture memory (step S63). Next, the short-time reference CIB memory for the same picture is released (step S64).

  If the meaning of the control signal is an instruction to release the designated long-time reference picture memory: the designated long-time reference picture memory is released and changed to a non-reference picture memory (step S65). Next, the long-time reference CIB memory for the same picture is released (step S66).

  When the meaning of the control signal is an instruction to change the designated short time reference picture memory to the long time reference picture memory: the designated short time reference picture memory is released and changed to the long time reference picture memory (step S67). ). Next, the short-time reference CIB memory for the same picture is released and changed to the long-time reference CIB memory (step S68).

  When the meaning of the control signal is an instruction to change the number of long-time reference picture memories: Change to the long-time reference picture memory so that a designated number of pictures can be stored (step S69). Next, the long-term reference CIB memory is changed so that the designated number of pictures can be stored (step S610).

  When the meaning of the control signal is an instruction to initialize all reference pictures: All reference pictures are released, and the longest reference picture maximum number is changed to 0 (step S611). Next, all short-time reference CIB memories and long-time reference CIB memories are released, and the maximum number of long-time reference CIB memories is changed to 0 (step S612).

  When the meaning of the control signal is an instruction to store the picture in the long-time reference picture memory: the decoded picture of the picture is stored in the long-time reference picture memory, and the context table of the picture is stored in the long-time reference CIB memory (Step S613).

  FIG. 10 shows an example of a hardware configuration when this system is realized by using a software program. This system includes a CPU 30 that executes a program, a memory 31 such as a RAM that stores programs and data accessed by the CPU 30, and an image signal input unit 32 (disk device) that inputs an image signal to be encoded from a camera or the like. Or a storage unit for storing image signals obtained from the CPU 30 or the like), and a software program to be executed by the CPU 30. An encoded stream generated by executing the program storage device 33 in which the moving image encoding program 34 represented by H.264 is stored and the moving image encoding program 34 loaded in the memory 31 by the CPU 30; For example, an encoded stream output unit 36 (which may be a storage unit that stores an encoded stream by a disk device or the like) that outputs via a network is connected by a bus.

  In particular, the context adaptive entropy encoding program 35 for causing the CPU 30 to execute the context adaptive entropy encoding according to the present embodiment is a part of the video encoding program 34 or separately from the video encoding program 34. It is stored in the program storage device 33 and loaded into the memory 31 for execution at the time of encoding.

  Even when the context adaptive entropy decoding method according to the present embodiment is realized by a software program, it can be realized by using a similar hardware configuration.

  FIG. 11 is a diagram for explaining the effect of the embodiment of the present invention. In this technique, the amount of code is reduced without increasing the amount of distortion. For this purpose, macroblocks belonging to the same subject are classified into classes in space-time, and the CABAC adaptive processing is limited for each class. carry out. As a result, processing specialized for signals within a class is realized, and the amount of codes can be reduced by refining the context classification.

  (A1) in FIG. 11 shows a state in which one subject is moving in the spatiotemporal image, and (B1) in FIG. 11 shows a state in which two subjects are moving in opposite directions. Yes. (A2) and (B2) of FIG. 11 are diagrams showing these spatiotemporal images continuously. Macroblocks having the same characteristics, such as the moving subject and the stationary background image, are classified into the same set (class), and the CABAC is specialized in the set, that is, a context table. By changing the initial value setting from a fixed value to an adaptive setting for each class, the coding efficiency can be improved.

DESCRIPTION OF SYMBOLS 1 Context adaptive entropy encoding device 10,20 Class classification information storage part 11,21 Context table initial value setting part 12,22 Context calculation part 13,23 Context table selection part 14 Binary part 15 Binary arithmetic encoding part 16 , 26 Context information buffer (CIB)
17, 27 Decoded image buffer (DPB)
18, 28 Context table management processing unit 2 Context adaptive entropy decoding device 24 Binary arithmetic decoding unit 25 Syntax element conversion unit

Claims (10)

In a context adaptive entropy encoding method for entropy encoding encoding target data in moving image encoding, using a context table holding probability model information that models conditional occurrence probabilities of encoding target symbols,
A decoded image buffer for storing a decoded image of an encoded image used for inter-frame prediction in moving image encoding;
In association with the frame number of the decoded image stored in the decoded image buffer, using a context information buffer for storing the context table written in block units to be encoded or slice units as a set thereof,
A context table management process for managing the memory area in the context information buffer in synchronization with the management of the memory area in the decoded image buffer;
Refers to class classification information storage means for storing information obtained by classifying encoding target data of a block or symbol to be encoded into a plurality of classes according to a predetermined classification criterion, and at the start of encoding of a certain encoding unit Context table initial value setting process for initializing the corresponding context table in the context information buffer using the probability model information of the encoded target data that belongs to the same class as the target data to be encoded When,
The encoding target data is encoded according to the probability model information of the context table initialized by the context table initial value setting process, the probability model information of the context table is updated according to the encoding result, and the updated A context adaptive entropy encoding method comprising: an encoding process for writing probability model information to a context table in the context information buffer. The context adaptive entropy encoding method according to claim 1,
The context information buffer is provided in a memory area in the decoded image buffer, and the memory for storing the corresponding context table is secured and released in accordance with the timing for securing and releasing the memory for storing the decoded image in the decoded image buffer. A context-adaptive entropy encoding method characterized by: The context adaptive entropy encoding method according to claim 1,
The context information buffer is provided in a memory area different from the decoded image buffer, and a memory for storing a corresponding context table is secured at the timing of securing and releasing the memory for storing the decoded image in the decoded image buffer. And a context-adaptive entropy coding method characterized by performing the release. In a context adaptive entropy encoding device that entropy-encodes encoding target data in moving picture encoding using a context table that stores probability model information that models conditional occurrence probabilities of encoding target symbols,
A decoded image buffer for storing a decoded image of an encoded image used for inter-frame prediction in moving image encoding;
A context information buffer for storing the context table written in block units to be encoded or in slice units as a set thereof in association with the frame number of the decoded image stored in the decoded image buffer;
Context table management processing means for performing management of the memory area in the context information buffer in synchronization with management of the memory area in the decoded image buffer;
Class classification information storage means for storing information obtained by classifying encoding target data of a block or symbol to be encoded into a plurality of classes according to a predetermined classification criterion;
At the start of encoding of a certain encoding unit, referring to the class classification information storage means, using the probability model information of the encoded target data that belongs to the same class as the target data to be encoded, A context table initial value setting means for initializing a corresponding context table in the context information buffer;
The encoding target data is encoded according to the probability model information included in the context table initialized by the context table initial value setting means, and the probability model information included in the context table is updated according to the encoding result. A context adaptive entropy encoding device comprising: encoding means for writing probability model information to a context table in the context information buffer. In a context adaptive entropy decoding method for entropy decoding decoding target data in moving picture decoding, using a context table holding probability model information that models conditional occurrence probability of decoding target symbols,
A decoded image buffer for storing a decoded image used for inter-frame prediction in moving image decoding;
In association with the frame number of the decoded image stored in the decoded image buffer, using a context information buffer for storing the context table written in block units to be decoded or slice units as a set thereof,
A context table management process for managing the memory area in the context information buffer in synchronization with the management of the memory area in the decoded image buffer;
Reference is made to class classification information storage means for storing information obtained by classifying decoding target data of a block or symbol to be decoded into a plurality of classes according to a predetermined classification standard, and decoding is performed at the start of decoding of a certain decoding unit. A context table initial value setting process for initializing a corresponding context table in the context information buffer using probability model information of decoded decoding target data belonging to the same class as the decoding target data;
The decoding target data is decoded according to the probability model information of the context table initialized by the context table initial value setting process, the probability model information of the context table is updated according to the decoding result, and the updated probability model A context adaptive entropy decoding method comprising: a decoding step of writing information to a context table in the context information buffer. The context adaptive entropy decoding method according to claim 5,
The context information buffer is provided in a memory area in the decoded image buffer, and the memory for storing the corresponding context table is secured and released in accordance with the timing for securing and releasing the memory for storing the decoded image in the decoded image buffer. A context adaptive entropy decoding method characterized by: The context adaptive entropy decoding method according to claim 5,
The context information buffer is provided in a memory area different from the decoded image buffer, and a memory for storing a corresponding context table is secured at the timing of securing and releasing the memory for storing the decoded image in the decoded image buffer. And a context adaptive entropy decoding method characterized by performing the release. In a context adaptive entropy decoding device for entropy decoding decoding target data in moving picture decoding, using a context table holding probability model information that models conditional occurrence probabilities of decoding target symbols,
A decoded image buffer for storing a decoded image used for inter-frame prediction in moving image decoding;
A context information buffer for storing the context table written in block units to be decoded or slice units as a set thereof in association with the frame number of the decoded image stored in the decoded image buffer;
Context table management processing means for performing management of the memory area in the context information buffer in synchronization with management of the memory area in the decoded image buffer;
Class classification information storage means for storing information obtained by classifying decoding target data of a block or symbol to be decoded into a plurality of classes according to a predetermined classification criterion;
At the start of decoding of a certain decoding unit, the class classification information storage means is referred to, and using the probability model information of decoded decoding target data belonging to the same class as the decoding target data to be decoded, the corresponding information in the context information buffer A context table initial value setting means for initializing the context table to be
The decoding target data is decoded according to the probability model information held by the context table initialized by the context table initial value setting means, the probability model information held by the context table is updated according to the decoding result, and the updated probability model A context adaptive entropy decoding apparatus comprising: decoding means for writing information to a context table in the context information buffer.   A context adaptive entropy encoding program for causing a computer to execute the context adaptive entropy encoding method according to claim 1, 2 or 3.   A context adaptive entropy decoding program for causing a computer to execute the context adaptive entropy decoding method according to claim 5, 6 or 7.

Images