JP5134048B2 - Decoding apparatus and method, recording medium, and program - Google Patents

Description

The present invention relates to a decoding apparatus and method , a recording medium, and a program . For example, a decoding apparatus and method , a recording medium, and a decoding apparatus that are suitable for use when encoding, transmitting, or storing an image signal at a higher compression rate than before. Regarding the program .

  In recent years, with the aim of handling images as digital signals and efficiently transmitting and storing the digital signals, MPEG (compressed by orthogonal transform such as discrete cosine transform and motion compensation, using redundancy unique to image information. A device compliant with a system such as Moving Picture Expert Group) is becoming popular for both information distribution in broadcasting stations and information reception in general households.

  In particular, the MPEG2 (ISO / IEC 13818-2) compression method is a standard defined as a general-purpose image compression method that covers both interlaced and progressively scanned images, standard resolution images, and high-definition images. For example, as represented by the DVD (Digital Versatile Disk) standard, it is currently widely used in a wide range of applications for professional use and consumer use.

  By using the MPEG2 compression method, for example, 4 to 8 Mbps for a standard resolution interlaced scanning image having 720 × 480 pixels, and 18 to 22 Mbps for a high resolution interlaced scanning image having 1920 × 1088 pixels. By assigning a code amount (bit rate), it is possible to realize a high compression rate and good image quality.

  By the way, MPEG2 was mainly intended for high-quality encoding suitable for broadcasting, but since it did not correspond to a higher compression rate encoding method, MPEG4 code was used as a higher compression rate encoding method. The standardization method was standardized. Regarding the image coding system, the standard was approved as an international standard as ISO / IEC 14496-2 in December 1998.

  Furthermore, in recent years, for the purpose of video coding for video conferences, H.26L (ITU-T Q6 / 16) by ITU-T (International Telecommunication Union-Telecommunication Standardization Sector), which is the telecommunications standardization department of the International Telecommunication Union A standard called VCEG) is being developed.

  H. 26L requires a larger amount of calculation for encoding and decoding compared to conventional encoding methods such as MPEG2 and MPEG4, but is known to achieve higher encoding efficiency. Yes.

  In addition, as a part of MPEG4 activities, H.C. Based on H.26L. Standardization of an encoding technique that realizes higher encoding efficiency that incorporates a function that is not supported by 26L is being promoted as Joint Model of Enhanced-Compression Video Coding.

  Here, a conventional image information encoding apparatus using orthogonal transform such as discrete cosine transform or Karhunen-Loeve transform and motion compensation will be described with reference to FIG. FIG. 1 shows an example of the configuration of a conventional image information encoding apparatus.

  In the image information encoding apparatus, an input image signal which is an analog signal is converted into a digital signal by the A / D conversion unit 1 and then supplied to the screen rearrangement buffer 2. The screen rearrangement buffer 2 rearranges the frame of the image information from the A / D conversion unit 1 according to the GOP (Group of Pictures) structure of the compressed image information output from the image information encoding device.

  First, an image on which intra (intra-image) encoding is performed will be described. In the screen rearrangement buffer 2, the image information is supplied to the orthogonal transform unit 4 via the adder 3 for the image on which intra coding is performed.

  In the orthogonal transform unit 4, orthogonal transform (discrete cosine transform, Karhunen-Loeve transform, or the like) is performed on the image information, and the obtained transform coefficient is supplied to the quantization unit 5. In the quantization unit 5, quantization processing is performed on the transform coefficients supplied from the orthogonal transform unit 4 in accordance with control from the rate control unit 8 based on the data amount of transform coefficients accumulated in the accumulation buffer 7. .

  The lossless encoding unit 6 determines an encoding mode from the quantized transform coefficient, quantization scale, and the like supplied from the quantization unit 5, and performs lossless encoding (variable length code) on the determined encoding mode. Or arithmetic coding) is performed, and information to be inserted into the header portion of the image coding unit is formed. The encoded encoding mode is supplied to and stored in the storage buffer 7. The encoded encoding mode stored in the storage buffer 7 is output to the subsequent stage as image compression information.

  Further, the lossless encoding unit 6 performs lossless encoding on the quantized transform coefficient and causes the accumulation buffer 7 to store the encoded transform coefficient. The encoded transform coefficient stored in the storage buffer 7 is also output to the subsequent stage as image compression information.

  In the inverse quantization unit 9, the transform coefficient quantized by the quantization unit 5 is inversely quantized. In the inverse orthogonal transform unit 10, inverse orthogonal transform processing is performed on the inversely quantized transform coefficient to generate decoded image information. The generated decoded image information is stored in the frame memory 11.

  Next, an image on which inter (inter-image) encoding is performed will be described. In the screen rearrangement buffer 2, the image information is supplied to the adder 3 and the motion prediction / compensation unit 12 for the image to be inter-coded.

  In the motion prediction / compensation unit 12, image information for reference corresponding to an image to be inter-coded from the screen rearrangement buffer 2 is read from the frame memory 11, and is subjected to motion prediction / compensation processing for reference. Image information is generated and supplied to the adder 3. Also, the motion vector information obtained in the motion prediction / compensation process by the motion prediction / compensation unit 12 is supplied to the lossless encoding unit 6.

  In the adder 3, the reference image information from the motion prediction / compensation unit 12 is converted into a difference signal from the image information of the image to be inter-coded from the screen rearrangement buffer 2.

  When processing an image on which inter-coding is performed, the orthogonal transform unit 4 performs orthogonal transform on the difference signal, and the obtained transform coefficient is supplied to the quantization unit 5. In the quantization unit 5, a quantization process is performed on the transform coefficient supplied from the orthogonal transform unit 4 in accordance with the control from the rate control unit 8.

  The lossless encoding unit 6 determines and determines an encoding mode based on the transform coefficient and quantization scale quantized by the quantization unit 5 and the motion vector information supplied from the motion prediction / compensation unit 12. The encoding mode is subjected to lossless encoding, and information to be inserted into the header portion of the image encoding unit is generated. The encoded encoding mode is stored in the storage buffer 7. The encoded encoding mode stored in the storage buffer 7 is output as image compression information.

  Further, the lossless encoding unit 6 performs lossless encoding processing on the motion vector information from the motion prediction / compensation unit 12 to generate information to be inserted into the header portion of the image encoding unit.

  In addition, since the process after the inverse quantization part 9 in the case of processing the image in which inter coding is performed is the same as that in the case of processing the image subjected to intra coding, description thereof is omitted.

  Next, a conventional image information decoding apparatus that restores an image signal using image compression information output from the conventional image information encoding apparatus shown in FIG. 1 as an input will be described with reference to FIG. FIG. 2 shows an example of the configuration of a conventional image information decoding apparatus.

  In the image information decoding apparatus, the input image compression information is temporarily stored in the accumulation buffer 21 and then transferred to the lossless decoding unit 22. The lossless decoding unit 22 performs lossless decoding (variable length decoding, arithmetic decoding, or the like) on the image compression information based on a predetermined format of the image compression information, and stores the encoding mode stored in the header portion. Information is acquired and supplied to the inverse quantization unit 23. Similarly, the lossless decoding unit 22 acquires the quantized transform coefficient and supplies it to the inverse quantization unit 23. Further, when the frame to be decoded is inter-coded, the lossless decoding unit 22 also decodes the motion vector information stored in the header portion of the image compression information, and the information is motion-predicted / compensated. Supplied to the unit 28.

  The inverse quantization unit 23 inversely quantizes the quantized transform coefficient supplied from the lossless decoding unit 22 and supplies the obtained transform coefficient to the inverse orthogonal transform unit 24. The inverse orthogonal transform unit 24 performs inverse orthogonal transform (inverse discrete cosine transform or inverse Karhunen-Labe transform) on the transform coefficient based on a predetermined format of image compression information.

  Here, when the target frame is intra-coded, the image information subjected to the inverse orthogonal transform is stored in the screen rearrangement buffer 26 via the adder 25 and is subjected to D / A conversion. The signal is converted into an analog signal by the unit 27 and output to the subsequent stage. The image information subjected to the inverse orthogonal transform is also stored in the frame memory 29.

  When the target frame is inter-coded, the motion prediction / compensation unit 28 is based on the motion vector information from the lossless decoding unit 22 and the image information stored in the frame memory 29. Thus, a reference image is generated and supplied to the adder 25. In the adder 25, the reference image from the motion prediction / compensation unit 28 and the output of the inverse orthogonal transform unit 25 are combined to generate image information. The other processing is the same as that of the intra-encoded frame, and thus description thereof is omitted.

  H. 26L defines two types of lossless encoding methods: UVLC (Universal Variable Length Code), which is a kind of variable-length coding, and CABAC (Context-based adaptive binary arithmetic coding), which is a kind of arithmetic coding. The user can select and apply either UVLC or CABAC as the lossless encoding method. Information indicating whether the lossless encoding method is UVLC or CABAC is specified in a field called Entropy Coding included in the RTP Parameter Set Packet of the RTP layer in the image compression information.

  Here, the arithmetic coding to which CABAC belongs will be described. In arithmetic coding, an arbitrary message (consisting of a plurality of alphabet symbols) is represented as one point on a half-open interval 0.0 ≦ x <1.0, and a code is generated from the coordinates of this point.

  First, based on the appearance probability of symbols constituting the alphabet, the half-open interval 0.0 ≦ x <1.0 is divided into sub-intervals corresponding to the respective symbols.

FIG. 3 shows an example of the occurrence probabilities of symbols s _{1 to} s ₇ and the division of sub-intervals. In arithmetic coding, as shown in FIG. 3, the upper and lower limits of the sub-interval are determined based on the cumulative appearance probability of each symbol. The lower limit of the subsection for the symbol s _i (i = 1, 2,..., 7) is the upper limit of the subsection of the symbol s _i−1 , and the upper limit of the subsection corresponding to the symbol s _i is the sub This is a value obtained by adding the appearance probability of the symbol s _i to the lower limit of the section.

It is assumed that (s ₂ s ₁ s ₃ s ₆ s ₇ ) is input as a message. However, the symbol s ₇ is a terminal symbol indicating the end of the message, and the message ends when the terminal symbol appears. In the arithmetic coding method, as shown in FIG. 4, the sub-interval corresponding to each symbol constituting the message is calculated for the message (s ₂ s ₁ s ₃ s ₆ s ₇ ). That is, the section allocated in FIG. 3 is divided according to the cumulative appearance probability of the next symbol. The sub-interval finally obtained is an interval including a value representing the message. Therefore, the message can be uniquely restored as long as the value is within this interval. However, in consideration of encoding efficiency, the message is represented by a number that can be expressed as a power of 2 within the half-open interval.

  That is, in this example, when the following expression (1) is considered, the following expression (2) is a value representing a message included in the half-open section 0.21164 ≦ x <0.2117.

2 ^-1
= 0.5
2 ^-2
= 0.25
2 ^-3
= 0.125
2 ^-4
= 0.0625
^2-5
= 0.03125
2 ^-6
= 0.015625
2 ^-7
= 0.0078125
^2-8
= 0.00390625
2 ^-9
= 0.001953125
2 ⁻¹⁰ = 0.0009765625
2 ^-11 = 0.00048828125
2 ^-12 = 0.000244140625
・
・
・

... (1)
2 ⁻³ +2 ⁻⁴ +2 ⁻⁶ +2 ⁻⁷ +2 ⁻¹¹ +2 ⁻¹² = 0.211669921875
... (2)

Therefore, the code length of the code corresponding to the message (s ₂ s ₁ s ₃ s ₆ s ₇ ) may be 12 bits so that 2 ^{−1 to} 2 ⁻¹² can be expressed, and the message (s ₂ s ₁ s ₃ s ₆ s ₇ ) is encoded into (0011011000011).

  Next, H.I. The CABAC defined in 26L will be described. Details of CABAC are disclosed in Non-Patent Document 1. CABAC is also H.264. Compared with UVLC defined by 26L, it has the following three characteristics.

  The first feature is that redundancy between symbols can be eliminated by using an appropriate context model for each symbol to be encoded and performing arithmetic encoding based on independent probability models.

  The second feature is that, in arithmetic coding, a code amount (bit) of a non-integer value can be assigned to each symbol, and coding efficiency close to entropy can be obtained. .

  The third feature is that, for example, statistical data of motion vectors is different not only in bit rate and sequence, but also in space and time, and is not constant. It is possible to follow the encoding.

  FIG. 5 shows a general configuration of a CABAC encoder to which CABAC is applied. In the CABAC encoder, the context modeling unit 31 first converts a symbol (symbol) of a syntax element into an appropriate context model according to a past history regarding an arbitrary syntax element in the image compression information. Such modeling is called context modeling. The context model for each syntax element in the image compression information will be described later.

  The binarization unit 32 binarizes symbols that have not been binarized. In the adaptive binary arithmetic coding unit 33, probability estimation is performed on the binarized symbol by the probability estimation unit 34, and adaptive arithmetic coding based on the probability estimation is performed by the coding engine 35. Since the relevant model is updated after the adaptive arithmetic coding process is performed, each model can perform the coding process according to the statistics of the actual image compression information.

  Here, a context model that arithmetically encodes the macroblock type MB_type (MB_type), motion vector information (MVD), and reference frame parameter (Ref_frame), which are syntax elements in the image compression information, will be described.

  MB_type context model generation will be described separately for intra frames and inter frames.

In the intra frame, when macro blocks A, B, and C are arranged as shown in FIG. 6, the context model ctx_mb_type_intra (C) corresponding to the MB_type of the macro block C is defined by the following equation (3). In the intra frame, the mode of the macro block is Intra 4 × 4 or Intra 16 × 16.
ctx_mb_type_intra (C) = A + B
... (3)

  However, in Expression (3), A is 0 when the macroblock A is Intra 4 × 4, and 1 when the macroblock A is Intra 16 × 16. Similarly, B is 0 when the macroblock B is Intra 4 × 4, and 1 when the macro block B is Intra 16 × 16. Therefore, the context model ctx_mb_type_intra (C) takes one of 0, 1, and 2.

In the inter frame, when macro blocks A, B, and C are arranged as shown in FIG. 6, the context model ctx_mb_type_inter (C) corresponding to the MB_type of the macro block C indicates that the inter frame is a P picture. It is defined by the following equation (4). When the inter frame is a B picture, it is defined by the following equation (5).
ctx_mb_type_inter (C) = ((A == Skip)? 0: 1) + ((B == Skip)? 0: 1)
... (4)
ctx_mb_type_inter (C) = ((A == Direct)? 0: 1) + ((B == Direct)? 0: 1) (5)

  However, in Expression (4), the operator ((A == Skip)? 0: 1) indicates 0 when the macroblock A is in Skip mode, and when the macroblock A is not in Skip mode. 1 shall be shown. Similarly, the operator ((B == Skip)? 0: 1) indicates 0 when the macroblock B is in Skip mode, and indicates 1 when the macroblock B is not in Skip mode. To do.

  In equation (5), the operator ((A == Direct)? 0: 1) indicates 0 when the macroblock A is in the Direct mode, and when the macroblock A is not in the Direct mode. 1 shall be shown. The operator ((B == Direct)? 0: 1) indicates 0 when the macroblock B is in the Direct mode, and indicates 1 when the macroblock B is not in the Direct mode.

  Therefore, the context model ctx_mb_type_inter (C) corresponding to the MB_type of the macroblock C in the inter frame (P picture) takes three values for each of the case of the P picture and the case of the B picture. .

  Next, context vector generation of motion vector information (MVD) will be described.

The motion vector information corresponding to the macro block of interest included in the image compression information is encoded as a prediction error with the motion vector corresponding to the adjacent macro block. Now, of the macroblocks A, B, and C arranged as shown in FIG. 7, the evaluation function e _k (C) for the macroblock C of interest is defined by the following equation (6). Here, k = 0 indicates a horizontal component, and k = 1 indicates a vertical component.
e _k (C) = | mvd _k (A) | + | mvd _k (B) |
... (6)

In Equation (6), mvd _k (A) and mvd _k (B) are motion vector prediction errors for macroblocks A and B adjacent to macroblock C, respectively.

Note that, regarding the equation (6), when the macroblock C exists at the left end of the image frame and one of the macroblocks A and B does not exist, the motion vector prediction error mvd _k (A) or mvd _k (B ) Cannot be obtained, the corresponding term on the right side of equation (6) is ignored. The context model ctx_mvd (C, k) corresponding to e _k (C) defined in this way is defined as the following equations (7-1) to (7-3).
ctx_mvd (C, k) = 0
e _k (C) <3
... (7-1)
ctx_mvd (C, k) = 1
32 <e _k (C)
... (7-2)
ctx_mvd (C, k) = 2
3 ≦ e _k (C) ≦ 32
... (7-3)

The context model generation of the motion vector information (MVD) is performed as shown in FIG. That is, the motion vector prediction error mvd _k (C) for the macroblock C is separated into an absolute value | mvd _k (C) | and a sign. The absolute value | mvd _k (C) | is binarized. The first bin (the leftmost value) of the binarized absolute value | mvd _k (C) | is encoded using the context model ctx_mvd (C, k) described above. The second bin (second value from the left end) is encoded using the context model 3. Similarly, the third and fourth bins are encoded using context models 4 and 5, respectively. The fifth and subsequent bins are encoded using the context model 6. The code of mvd _k (C) is encoded using the context model 7. Thus, motion vector information (MVD) is encoded using eight types of context models.

  Next, a context model for encoding the reference frame parameter (Ref_frame) will be described.

When two or more reference frames are used for an inter frame, information on the reference frame is set for each macroblock of the inter frame. In the macroblocks A, B, and C arranged as shown in FIG. 6, when the reference frame parameters for the macroblocks A and B are A and B, the context model ctx_ref_frame (C) for the macroblock C is Defined by (8).
ctx_ref_frame (C) = ((A == 0)? 0: 1) +2 ((B == 0)? 0: 1)
... (8)

  However, in Expression (8), the operator ((A == 0)? 0: 1) indicates 0 when the reference frame parameter of macroblock A is 0, and the reference frame parameter of macroblock A is 0. If not, 1 is shown. Similarly, the operator ((B == 0)? 0: 1) indicates 0 when the reference frame parameter of the macroblock B is 0, and when the reference frame parameter of the macroblock B is not 0, 1 shall be shown.

  Four types of context models for encoding the reference frame parameter (Ref_frame) are defined by Expression (8). Further, a context model for the second bin and a context model for the third and subsequent bins are defined.

  Next, H.I. A context model for calculating and encoding code block pattern (CBR), intra prediction mode (IPRED), and (RUN, LEVEL) information, which are syntax elements related to texture information included in the 26L image compression information, will be described.

  First, the context model regarding the code block pattern will be described. The handling of code block patterns other than Intra16 × 16 macroblocks is defined as follows.

That is, regarding the luminance signal, a total of 4 CBP bits are included, one for each of the four 8 × 8 blocks included in the Intra 16 × 16 macroblock. When macroblocks A, B, and C are arranged as shown in FIG. 6, the context model ctx_cbp_luma (C) corresponding to the luminance signal of the macroblock C is defined by the following equation (9).
ctx_cbp_luma (C) = A + 2B
... (9)
In Equation (9), A is the CBP bit of the luminance signal of the macroblock A, and B is the CBP bit of the luminance signal of the macroblock B.

The remaining 2 bits of the CBP field relate to the color difference signal. The context model ctx_cbp_chroma_sig (C) corresponding to the color difference signal of the macroblock C is defined by the following equation (10).
ctx_cbp_chroma_sig (C) = A + 2B
... (10)
In Equation (10), A is the CBP bit of the color difference signal of the macroblock A, and B is the CBP bit of the color difference signal of the macroblock B.

Here, when the context model ctx_cbp_chroma_sig (C) corresponding to the color difference signal of the macroblock C is not 0, that is, when the AC component of the color difference signal exists, the color difference of the macroblock C defined by the following equation (11) The context model ctx_cbp_chroma_ac (C) corresponding to the AC component of the signal needs to be encoded.
ctx_cbp_chroma_ac (C) = A + 2B
(11) However, in Expression (11), A is cbp_chroma_ac decision corresponding to the macroblock A, and B is cbp_chroma_ac decision corresponding to the macroblock B.

  Since the context models defined by the equations (9) to (11) are defined separately for each of the intra macro block and the inter macro block, 24 (= 2 × 3 × 4) types of context models are defined. Will be.

  Furthermore, for an Intra 6 × 16 macroblock, one type of context model is defined for the binarized AC decision, and one type of context model is defined for each component of the color difference signal. .

  Next, a context model related to the intra prediction mode (IPRED) will be described. Here, H. Six types (labels 0 to 5) of intra prediction modes defined in H.26L will be described with reference to FIGS. FIG. 9 shows pixels a to p existing in a 4 × 4 block obtained by dividing a macroblock and pixels A to I existing in each adjacent 4 × 4 block. Labels 1 to 5 in FIG. 10 indicate directions of the intra prediction modes of labels 1 to 5, respectively. The intra prediction mode labeled 0 is a DC prediction mode (DC Prediction).

In the intra prediction mode labeled 0, pixels a to p are predicted according to the following equation (12).
Pixels a to p = (A + B + C + D + E + F + G + H) / 8
(12) However, in the expressions (12) to (15), A to I indicate the pixels A to I, respectively, and the symbol “//” means an operation for rounding the result of division. doing.

  In addition, in the intra prediction mode of label 0, when 4 pixels (for example, pixels A to D) out of 8 pixels A to H do not exist in the image frame, Expression (12) is not used and the remaining 4 pixels The average value of the pixels E to H in this case is the predicted value of the pixels a to p. Also, when all of the eight pixels A to H are not present in the image frame, the equation (12) is not used, and a predetermined value (for example, 128) is set as the predicted value of the pixels a to p.

The intra prediction mode of label 1 is called Vertical / Diagonal Prediction. The intra prediction mode of label 1 is used only when 4 pixels A to D are present in the image frame. In this case, each of the pixels a to p is predicted according to the following equations (13-1) to (13-6).
Pixel a
= (A + B) / 2
... (13-1)
Pixel e
= B
(13-2)
Pixel b, i = (B + C) / 2
... (13-3)
Pixel f, m = C
... (13-4)
Pixel c, j = (C + D) / 2
... (13-5)
Pixels d, g, h, k, l, n, o, p
= D
... (13-6)

  The intra prediction mode of label 2 is called Vertical Prediction. The intra prediction mode of label 2 is used only when 4 pixels A to D are present in the image frame. In this case, for example, the pixel A is used as the predicted value of the pixels a, e, i, and m, and the pixel B is used as the predicted value of the pixels b, f, j, and n.

The intra prediction mode labeled 3 is called Diagonal Prediction. The intra prediction mode of label 1 is used only when 9 pixels A to I are present in the image frame. In this case, each of the pixels a to p is predicted according to the following equations (14-1) to (13-7).
Pixel m
= (H + 2G + F) / 4
... (14-1)
Pixel i, n
= (G + 2F + E) / 4
(14-2)
Pixels e, j, o
= (F + 2E + I) // 4
... (14-3)
Pixels a, f, k, p = (E + 2I + A) / 4
... (14-4)
Pixels b, g, l
= (I + 2A + B) / 4
... (14-5)
Pixels c, h
= (A + 2B + C) / 4
... (14-6)
Pixel d
= (B + 2C + D) // 4
... (14-7)

  The intra prediction mode of label 4 is called Horizontal Prediction. The intra prediction mode of label 4 is used only when 4 pixels E to H are present in the image frame. In this case, for example, the pixel E is used as the predicted value of the pixels a, b, c, and d, and the pixel F is used as the predicted value of the pixels e, f, g, and h.

The intra prediction mode labeled 5 is called Horizontal / Diagonal Prediction. The intra prediction mode of label 5 is used only when 4 pixels E to H are present in the image frame. In this case, each of the pixels a to p is predicted according to the following equations (15-1) to (15-6).
Pixel a
= (E + F) / 2
... (15-1)
Pixel b
= F
... (15-2)
Pixel c, e = (F + G) / 2
... (15-3)
Pixel f, d = G
... (15-4)
Pixel i, g = (G + H) / 2
... (15-5)
Pixels h, j, k, l, m, n, o, p
= H
... (15-6)

  Two context models are defined for each of the intra prediction modes labeled 0 to 5. That is, one is the first bin for each mode and the other is the second bin for each mode. In addition to these, one context model is defined for two bits of the Intra 16 × 16 mode. Therefore, a total of 14 context models are defined for the intra prediction mode.

  Next, a context model relating to (RUN, LEVEL) will be described.

  H. In 26L, two types of methods shown in FIGS. 11A and 11B are defined as scan methods for rearranging two-dimensional discrete cosine transform coefficients in one dimension. The single scan method shown in FIG. 11A is a method used for a case where the luminance signal is for an intra macroblock and the quantization parameter QP is smaller than 24. The double scan method shown in FIG. 11B is used when the single scan method is not used.

  In an intra macroblock and an intra macroblock having a quantization parameter QP of 24 or more, on average, there is one non-zero coefficient for a 4 × 4 macroblock, and a 1-bit EOB (End Of Block) signal is sufficient. However, for the luminance signal of an intra macroblock whose quantization parameter QP is smaller than 24, since there are two or more non-zero coefficients, a 1-bit EOB signal is not sufficient. For this reason, the double scan method shown in FIG. 11B is used.

  As shown in FIG. 12, the context model for (RUN, LEVEL) includes the above-described scan method distinction, DC / AC block type distinction, luminance signal / chrominance signal distinction, and intra macroblock / intermacroblock distinction. Nine types are defined accordingly.

  LEVEL information is separated into a sign and an absolute value. Four context models are defined according to the corresponding Ctx_run_level shown in FIG. That is, the first context model is for the code, the second context model is for the first bin, and the second context model is for the second bin. The fourth context model is defined for subsequent bins.

  When LEVEL is not 0 (not EOB), RUN described below is encoded. For RUN, two context models are defined for each of the first bin and the second and subsequent bins for each Ctx_run_level shown in FIG.

  H. A context model for the quantization parameter Dquant that can be set at the macroblock level in the 26L image compression information will be described.

The parameter Dquant is set when the code block pattern for the macroblock includes a non-zero orthogonal transform coefficient, or when the macroblock is 16 × 16 Intra Coded. The parameter Dquant can take a value of -16 to 16. The quantization parameter QUANT _new for the macroblock is calculated by the following equation (16) using the parameter Dquant in the image compression information.
QUANT _new = modulo ₃₂ (QUANT _old + Dquant + 32)
(16) However, in equation (16), QUANT _old is a quantization parameter used for the previous encoding or decoding.

The first context model ctx_dquant (C) for the parameter Dquant of the macroblock C arranged as shown in FIG. 6 is defined as the following equation (17).
ctx_dquant (C) = (A! = 0)
(17) However, in equation (17), A indicates the value of the parameter Dquant of the macroblock A. A second context model is defined for the first bin, and a second context model is defined for the second and subsequent bins.

  For the various context models described above, when the input symbol is not binarized, it is necessary to binarize the symbol before input. Syntax elements other than MB_type are binarized according to the correspondence shown in FIG.

  MB_type defined for 10 types of P picture is binarized by the correspondence shown in FIG. 14A. Also, MB_type defined for 17 types for B picture is binarized by the correspondence shown in FIG. 14B.

  The registers corresponding to the various context models described above are initialized in advance with values calculated in advance, and when encoding each symbol, the occurrence frequency of bins for a series of context models is sequentially updated. This is used for the determination when the symbols are encoded.

  However, when the occurrence frequency for a given context model exceeds a predetermined value, the frequency counter is reduced. By periodically performing scaling processing in this way, it is easy to cope with the dynamic generation of symbols.

  H. In 26L, the method disclosed in Non-Patent Document 2 is currently applied to the arithmetic coding method of binarized symbols.

  By the way, in MPEG2, when an input image signal is an interlaced scanning format, field / frame adaptive encoding processing can be performed at a macroblock level.

  Currently H. Such a specification is not defined in H.26L. It has been proposed to extend the 26L specification to allow field / frame adaptive coding at the macroblock level.

  A field / frame adaptive encoding process at the macroblock level proposed in Non-Patent Document 3 will be described.

  Current H. In 26L, seven types of modes (modes 1 to 7) as shown in FIG. 15 are defined as motion prediction / compensation units in a macroblock.

  Non-Patent Document 3 proposes having a Frame / Field Flag between Run and MB_type as syntax corresponding to a macroblock of image compression information, as shown in FIG. When the value of Frame / Field Flag is 0, this indicates that the macroblock is subjected to frame-based encoding. When the value of Frame / Field Flag is 1, field-based encoding is performed. It is shown that.

  When the value of Frame / Field Flag is 1 (that is, when field-based encoding is performed), the pixels in the macroblock are rearranged in units of rows as shown in FIG.

  When the value of Frame / Field Flag is 1, five modes (modes 1a to 5a) shown in FIG. 18 corresponding to modes 3 to 7 in FIG. 15 are defined as motion prediction / compensation units in the macroblock. Yes.

  For example, in mode 2a of FIG. 18, among the 8 × 8 blocks 0 to 3 obtained by dividing the macroblock into four, blocks 0 and 1 belong to the same field parity, and blocks 2 and 3 belong to the same field parity. Also, for example, in mode 3a in FIG. 18, among 4 × 8 blocks 0 to 8 obtained by dividing a macroblock into eight, blocks 0 to 3 belong to the same field parity, and blocks 4 to 7 belong to the same field parity. .

  The intra prediction mode when the value of Frame / Field Flag is 1 will be described. For example, the pixels a to p located in the 4 × 4 block shown in FIG. 9 use the pixels A to I located in the adjacent 4 × 4 block even when the value of the Frame / Field Flag is 1. Intra prediction is performed, but the pixels a to p and the pixels A to I all belong to the same field parity.

  A case where the pixels A to I belong to the same macroblock as the pixels a to p will be described with reference to FIG. The pixels a to p existing in the 4 × 4 block 7 obtained by dividing the macroblock into 16 are subjected to intra prediction using the pixels A to I existing at the ends of the adjacent blocks 2, 3, and 6.

  A case where the pixels A to I belong to a different macroblock from the pixels a to p will be described with reference to FIG.

  FIG. 20A shows a case where the value of the Frame / Field Flag is 1 for the macroblock on the left side of the macroblock to be processed and the macroblock on the upper side. In this case, the intra prediction of the pixels existing in the 4 × 4 block C obtained by dividing the macroblock to be processed into 16 blocks is performed using the pixels existing in the block 4 × 4 block A obtained by dividing the left macroblock into 16 blocks and the upper macroblock. This is performed using pixels existing in a 4 × 4 block B obtained by dividing the block into 16 blocks. Intra prediction of pixels existing in the 4 × 4 block C ′ is performed using pixels existing in the 4 × 4 block A ′ and pixels existing in the 4 × 4 block B ′.

  FIG. 20B shows a case where the value of Frame / Field Flag for the macroblock to be processed is 1, and the values of Frame / Field Flag for the left and upper macroblocks are 0, respectively. In this case, the intra prediction of the pixels existing in the 4 × 4 block C obtained by dividing the macroblock to be processed into 16 blocks is performed on the pixels existing in the 4 × 4 block A obtained by dividing the left macroblock into 16 blocks and the upper macroblock. Is performed using pixels existing in a 4 × 4 block B obtained by dividing 16 into 16 blocks. Intra prediction of pixels existing in the 4 × 4 block C ′ is performed using pixels existing in the 4 × 4 block A ′ and pixels existing in the 4 × 4 block B.

  Next, intra prediction of color difference signals will be described with reference to FIG. When the value of Frame / Field Flag is 1, only one type of intra prediction mode for color difference signals is defined.

In FIG. 21, A to D indicate 4 × 4 blocks of color difference signals, respectively. Blocks A and B belong to the first field, and blocks C and D belong to the second field. s _{0 to} s ₂ are the total values of the color difference signals existing in the blocks belonging to the first field parity among the blocks adjacent to the blocks A to D. s ₃ to s ₅ are the total values of the color difference signals existing in the blocks belonging to the second field parity among the blocks adjacent to the blocks A to D.

The predicted values A to D corresponding to the blocks A to D are predicted according to the following equation (18) when all of s _{0 to} s ₅ are present in the image frame.
A = (s ₀ + s ₂ +4) / 8
B = (s ₁ +2) / 4
C = (s ₃ + s ₅ +4) / 8
D = (s ₄ +2) / 4
... (18)

However, when only s ₀ , s ₁ , s ₃ , and s ₄ are present in the image frame among s _{0 to} s ₅ , predicted values A to D corresponding to the blocks A to D are expressed by the following equation (19 ) To be predicted.
A = (s ₀ +2) / 4
B = (s ₁ +2) / 4
C = (s ₃ +2) / 4
D = (s ₄ +2) / 4
... (19)

Furthermore, when only s ₂ s ₅ out of s _{0 to} s ₅ are present in the image frame, the predicted values corresponding to the blocks A to D are predicted according to the following equation (20).
A = (s ₂ +2) / 4
B = (s ₂ +2) / 4
C = (s ₅ +2) / 4
D = (s ₅ +2) / 4
... (20)

  FIG. 22 shows a method of encoding the residual component of the color difference signal after intra prediction as described above. That is, after orthogonal transform processing is performed on each 4 × 4 block, a 2 × 2 block as illustrated is generated using the DC components of the first field and second field, and orthogonal transform processing is performed again. Is done.

  Next, motion prediction / compensation processing when the value of Frame / Field Flag is 1 will be described. When the value of the Frame / Field Flag is 1, the motion prediction / compensation mode includes 6 modes of inter 16 × 16 mode, inter 8 × 16 mode, inter 8 × 8 mode, inter 4 × 8 mode, and inter 4 × 4 mode. There are different types of modes.

  For example, the inter 16 × 16 mode is a mode in which the motion vector information for the first field, the motion vector information for the second field, and the reference frame in the inter 8 × 16 mode are equivalent.

  Code_Number 0 to 5 are assigned to these six types of motion prediction compensation modes, respectively.

  Current H. In H.26L, multiple frame prediction capable of providing a plurality of reference frames as shown in FIG. 23 is defined. Current frame-based H.264 In the 26L standard, information on the reference frame is defined at the macroblock level, Code_Number0 is assigned to the frame encoded immediately before, and the frame encoded one to five times before is assigned to the frame. On the other hand, Code_Number 1 to 5 are assigned, respectively.

  On the other hand, when performing field-based encoding, Code_Number0 is assigned to the first field of the frame encoded immediately before, and Code_Number1 is assigned to the second field of the frame. Code_Number2 is assigned to the first field of the frame encoded one time before, and Code_Number3 is assigned to the second field of the frame. Furthermore, Code_Number4 is assigned to the first field of the frame encoded once before, and Code_Number5 is assigned to the second field.

  In addition, for a macroblock subjected to field-based coding, a reference field for the first field and a reference field for the second field are separately defined.

  Next, a motion vector information prediction method when the value of Frame / Field Flag is 1 will be described. The median prediction defined in 26L will be described with reference to FIG. The 16 × 16, 8 × 8, or 4 × 4 motion vector information corresponding to the 16 × 16 macroblock E shown in FIG. 24 is predicted using the median of the motion vector information of the adjacent macroblocks A to C.

  However, for macroblocks A to C that do not exist within the image frame, the median is calculated assuming that the value of the corresponding motion vector information is zero. For example, when the macroblocks D, B, and C do not exist in the image frame, motion vector information corresponding to the macroblock A is used as a predicted value. When the macro block C does not exist in the image frame, the median is calculated using the motion vector information of the macro block D instead.

  Note that the reference frames of the macroblocks A to D are not necessarily the same.

  Next, the case where the block size of the macroblock is 8 × 16, 16 × 8, 8 × 4, or 4 × 8 will be described with reference to FIG. Note that the macro block E of interest and the macro blocks A to D adjacent thereto are arranged as shown in FIG.

  FIG. 25A shows a case where the block sizes of the macroblocks E1 and E2 are 8 × 16. Regarding the left macroblock E1, when the macroblock A adjacent to the left refers to the same frame as the macroblock E1, the motion vector information of the macroblock A is used as a prediction value. When the macroblock A adjacent to the left refers to a frame different from the macroblock E1, the median prediction described above is applied.

  Regarding the right macroblock E2, when the macroblock C adjacent to the upper right refers to the same frame as the macroblock E2, the motion vector information of the macroblock C is used as a predicted value. When the macroblock C adjacent to the upper right refers to a frame different from the macroblock E2, the median prediction described above is applied.

  FIG. 25B shows a case where the block sizes of the macroblocks E1 and E2 are 16 × 8. As for the upper macroblock E1, when the macroblock B adjacent on the upper side refers to the same frame as the macroblock E1, the motion vector information of the macroblock B is used as a predicted value. When the macroblock B adjacent on the top refers to a frame different from the macroblock E1, the median prediction described above is applied.

  Regarding the lower macroblock E2, when the macroblock A adjacent to the left refers to the same frame as the macroblock E2, the motion vector information of the macroblock A is used as a predicted value. When the macroblock A adjacent to the left refers to a frame different from the macroblock E2, the median prediction described above is applied.

  FIG. 25C shows a case where the block sizes of the macroblocks E1 to E8 are 8 × 4. The median prediction described above is applied to the left macroblocks E1 to E4, and the motion vector information of the left macroblocks E1 to E4 is used as a prediction value for the right macroblocks E5 to E8.

  FIG. 25D shows a case where the block sizes of the macroblocks E1 to E8 are 4 × 8. The median prediction described above is applied to the upper macroblocks E1 to E4, and the motion vector information of the upper macroblocks E1 to E4 is used as a prediction value for the lower macroblocks E5 to E8. .

  Even when the value of the Frame / Field Flag is 1, the prediction of the horizontal direction component of the motion vector information conforms to the above method. However, regarding the vertical component, since field-based blocks and frame-based blocks coexist, the following processing is performed. Note that the macro block E of interest and the macro blocks A to D adjacent thereto are arranged as shown in FIG.

  When the macroblock E is frame-based encoded and any of the adjacent macroblocks A to D is field-based encoded, the vertical direction component of the motion vector information for the first field and the second field Twice the average value of the vertical component of the motion vector information for the motion vector information is calculated, and the prediction processing is performed assuming that this is equivalent to the frame-based motion vector information.

  When the macroblock E is field-based encoded and any of the adjacent blocks A to D is frame-based encoded, the quotient obtained by dividing the value of the vertical direction component of the motion vector information by 2, Prediction processing is performed assuming that it corresponds to a field-based motion vector.

"Video Compression Using Context-Based Adaptive Arithmetic Coding", Marpe et al, ICIO1 "Arithmetic Coding for Data Compression", (Witten et al. Comm. Of the ACM, 30 (6), 1987, pp520-541) "Interlace Coding Tools for H.26L Video Coding (L. Wang et al., VCEG-O37, Dec. 2001)"

  By the way, in Non-Patent Document 3, syntax elements necessary for field / frame encoding at the macroblock level are added, and the semantics of syntax elements such as motion vector information are also changed. On the other hand, the introduction of a new context model and the change of the existing context model have not been made, and only with the information proposed in Non-Patent Document 3, the field / frame code at the macroblock level using the CABAC method is used. It is impossible to make it.

  The CABAC method requires a larger amount of calculation processing than the UVLC method, but is known to achieve higher encoding efficiency, and the input image information is in an interlaced scanning format. In addition, it is desirable to be able to implement macroblock level field / frame coding using the CABAC method.

  The present invention has been made in view of such circumstances, and enables macroblock level field / frame encoding using the CABAC method even when the input image information is in an interlaced scanning format. For the purpose.

One aspect of the present invention shows, on a macroblock level, whether the image information is encoded in a frame mode or a field mode in a decoding device that decodes encoded data in which the image information is encoded. Using the coding mode information, the motion vector prediction error of an adjacent macroblock adjacent to the target macroblock to be decoded is converted to match the coding mode of the target macroblock, and the motion of the target macroblock Context model means for calculating a context model corresponding to a vector prediction error; and context adaptive arithmetic decoding means for performing context adaptive arithmetic decoding on the encoded data using the context model calculated by the context model means. It is a decoding device.

One aspect of the present invention is also a decoding method of a decoding apparatus for decoding encoded data in which image information is encoded, wherein the context model means determines whether the image information is encoded in a frame mode or a field mode. The motion vector prediction error of the adjacent macroblock adjacent to the target macroblock to be decoded is matched with the coding mode of the target macroblock using the coding mode information indicating whether the data is encoded with the macroblock level. And a context model corresponding to the motion vector prediction error of the target macroblock is calculated, and context adaptive arithmetic decoding means uses the calculated context model to perform context adaptive arithmetic decoding on the encoded data. Is a decoding method for performing

According to another aspect of the present invention, there is provided a computer that decodes encoded data in which image information is encoded, and whether the image information is encoded in a frame mode or a field mode at a macroblock level. Using the coding mode information shown, the motion vector prediction error of an adjacent macroblock adjacent to the target macroblock to be decoded is converted to match the coding mode of the target macroblock, and the target macroblock Context model means for calculating a context model corresponding to a motion vector prediction error, and using the context model calculated by the context model means to function as context adaptive arithmetic decoding means for performing context adaptive arithmetic decoding on the encoded data Recording program for A computer-readable recording medium.

According to another aspect of the present invention, there is provided a computer that decodes encoded data in which image information is encoded, and whether the image information is encoded in a frame mode or a field mode at a macroblock level. Using the coding mode information shown, the motion vector prediction error of an adjacent macroblock adjacent to the target macroblock to be decoded is converted to match the coding mode of the target macroblock, and the target macroblock Context model means for calculating a context model corresponding to a motion vector prediction error, and using the context model calculated by the context model means to function as context adaptive arithmetic decoding means for performing context adaptive arithmetic decoding on the encoded data It is a program for.

In one aspect of the present invention, target macroblocks to be decoded using encoding mode information indicating at a macroblock level whether image information is encoded in a frame mode or a field mode. The motion vector prediction error of the adjacent macroblock adjacent to the target macroblock is converted so as to match the encoding mode of the target macroblock, and a context model corresponding to the motion vector prediction error of the target macroblock is calculated, and the calculated context model Is used to perform context adaptive arithmetic decoding on the encoded data.

  As described above, according to the present invention, the image information in the interlaced scanning format is decoded by decoding the compressed image information in which the image information in the interlaced scanning format is field / frame encoded at the macroblock level using the CABAC method. It can be restored.

It is a block diagram which shows the structure of the conventional image information encoding apparatus which implement | achieves image compression by orthogonal transformation processing and motion compensation processing. It is a block diagram which shows the structure of the image information decoding apparatus corresponding to the image information encoding apparatus of FIG. It is the figure which showed an example of the correspondence of the generation | occurrence | production probability of a symbol and the allocated sub-section in arithmetic coding processing. It is a figure which shows an example of an arithmetic encoding process. It is a block diagram which shows the general structure of a CABAC encoder. It is a figure for demonstrating the context model of MB_type. It is a figure for demonstrating the context model of motion vector information MVD. It is a figure for demonstrating the process which encodes the motion vector information MVD based on a context model. H. It is a figure for demonstrating the intra prediction mode defined by 26L. It is a figure for demonstrating the direction of the intra prediction modes of the labels 1-5. H. It is a figure for demonstrating the single scan system and the double scan system defined by 26L. H. It is a figure which shows the context model corresponding to (RUN, LEVEL) defined by 26L. H. It is a figure for demonstrating the process which binarizes syntax elements other than MB_type in 26L. H. It is a figure for demonstrating the process which binarizes MB_type of P picture and B picture in 26L. H. It is a figure which shows seven types of modes as a unit of the motion estimation and compensation in a macroblock defined in 26L. It is a figure which shows the syntax of the image compression information extended so that the field / frame adaptive encoding of a macroblock level could be performed. It is a figure for demonstrating rearrangement of the pixel of a macroblock in the case of encoding a macroblock on a field basis. It is a figure which shows five types of modes defined as a unit of motion prediction and compensation in the case of encoding a macroblock on a field basis. It is a figure for demonstrating the operation principle which performs intra prediction in a macroblock in the case of encoding a macroblock on a field basis. It is a figure for demonstrating the operation principle which performs intra prediction across macroblocks in the case of encoding a macroblock on a field basis. It is a figure for demonstrating the operation principle which performs the intra prediction with respect to a colour-difference signal in the case of encoding a macroblock on a field basis. It is a figure for demonstrating the operation | movement principle which encodes the residual component of a colour-difference signal in the case of encoding a macroblock on a field basis. H. It is a figure for demonstrating the multiple frame prediction prescribed | regulated in 26L. It is a figure for demonstrating the prediction method of motion vector information in the case of encoding a macroblock on a field basis. H. It is a figure for demonstrating the process which produces | generates the predicted value of the motion vector information in each prediction mode defined by 26L. It is a block diagram which shows the structural example of the image information encoding apparatus which is one embodiment of this invention. FIG. 27 is a block diagram illustrating a configuration example of an arithmetic encoding unit 58 in FIG. 26. It is a figure which shows the table for binarizing MB_type of the macroblock which belongs to P picture and B picture in the case of encoding a macroblock on a field basis. FIG. 27 is a block diagram illustrating a configuration example of an image information decoding device according to an embodiment of the present invention, corresponding to the image information encoding device of FIG. 26.

  Hereinafter, an image information encoding apparatus to which the present invention is applied will be described with reference to FIG. The image information encoding apparatus can perform encoding processing using the CABAC method even when input image information is in an interlaced scanning format.

  In the image information encoding apparatus, the A / D conversion unit 51 converts an input image signal that is an analog signal into a digital signal and outputs the digital signal to the screen rearrangement buffer 52. The screen rearrangement buffer 52 rearranges the input image information from the A / D conversion unit 51 according to the GOP structure of the image compression information output from the image information encoding device, and outputs the rearranged image information to the adder 54.

  The field / frame determination unit 53 determines whether the macroblock of the processing target image is field-based or frame-based encoded and has the higher encoding efficiency, and the corresponding Frame / Field A flag is generated and output to the field / frame conversion unit 55 and the arithmetic coding unit 58.

  When the macroblock to be processed is inter-coded, the adder 54 generates a difference image between the input image via the field / frame determination unit 53 and the reference image from the motion prediction / compensation unit 64, and / Output to frame conversion unit 55 and orthogonal conversion unit 56. Further, when the macroblock to be processed is intra-coded, the adder 54 outputs the input image that has passed through the field / frame determination unit 53 to the field / frame conversion unit 55 and the orthogonal transformation unit 56 as they are.

  When the processing target macroblock is encoded on a field basis, the field / frame conversion unit 55 converts the input image from the adder 54 into a field structure and outputs it to the orthogonal transformation unit 56. The orthogonal transform unit 56 performs orthogonal transform (discrete cosine transform, Karoonen-Loeve transform, etc.) on the input image information, and supplies the obtained transform coefficient to the quantization unit 57. The quantization unit 57 performs a quantization process on the transform coefficient supplied from the orthogonal transform unit 56 under the control of the rate control unit 65 and the like.

  The arithmetic coding unit 58 arithmetically codes and stores each syntax element input from the quantization unit 57 and the motion prediction / compensation unit 64 and the Frame / Field Flag from the field / frame determination unit 53 based on the CABAC method. The data is supplied to the buffer 59 and accumulated. The accumulation buffer 59 outputs the accumulated image compression information to the subsequent stage.

  The inverse quantization unit 60 performs inverse quantization on the quantized orthogonal transform coefficient and outputs the result to the inverse orthogonal transform unit 61. The inverse orthogonal transform unit 61 performs inverse orthogonal transform processing on the inversely quantized transform coefficients to generate decoded image information, which is supplied to the frame memory 62 and stored. The field / frame conversion unit 63 converts the decoded image information stored in the frame memory 62 into a field structure and outputs the macroblock to be processed to the motion prediction / compensation unit 64 when encoding a macroblock to be processed on a field basis. .

  The motion prediction / compensation unit 64 generates optimal prediction mode information and motion vector information by motion prediction processing and outputs them to the arithmetic encoding unit 58, and also generates a prediction image and outputs it to the adder 54. The rate control unit 65 performs feedback control of the operation of the quantization unit 57 based on the amount of data stored in the storage buffer 59. The control unit 66 controls each unit of the image information encoding apparatus according to the control program recorded on the recording medium 67.

  Next, the operation principle of the arithmetic encoding unit 58 will be described with reference to FIG. FIG. 27 shows a configuration example of the arithmetic encoding unit 58. In the arithmetic encoding unit 58, among the syntax elements of the input image compression information, first, the frame / field flag shown in FIG. 16 is encoded by the frame / field flag context model 91.

  When the macroblock to be processed is frame-based encoded, the current H.264 A frame-based context model 92 defined in the 26L standard is applied. Note that a syntax element having a value that has not been binarized is binarized by the binarizing unit 93 and then subjected to arithmetic coding.

  On the other hand, when a macroblock to be processed is field-encoded, a field-based context model 94 is applied to the following syntax elements. Note that a syntax element having a value that has not been binarized is binarized by the binarization unit 95 and then subjected to arithmetic coding. That is, the first syntax element is MB_type for I picture, the second syntax element is MB_type for P / B picture, the third syntax element is motion vector information, and the fourth syntax element is a reference. It is a field parameter, and the fifth syntax is an intra prediction mode.

Hereinafter, it is assumed that macroblocks A, B, and C are arranged as shown in FIG. A context model related to frame / field flag will be described. The context model ctx_fifr_flag (C) regarding the frame / field flag of the macroblock C is defined by the following equation (21).
ctx_fifr_flag (C) = a + 2b
(21) However, in equation (21), a and b are the values of the frame / field flag of macroblocks A and B, respectively.

Next, a context model regarding MB_type for an I picture will be described. When the frame / field flag is 1, the context model ctx_mb_type_intra_field (C) corresponding to the MB_type of the macroblock C included in the I picture is defined by the following equation (22) similarly to the equation (3).
ctx_mb_type_intra_field (C) = A + B
(22) However, A and B in the equation (22) are the same as those in the equation (3). Adjacent macroblocks A and B may be field-based encoded or frame-based encoded.

Next, a context model related to MB_type for P / B pictures will be described. When the macroblock C is included in the P picture, the context model ctx_mb_type_inter_field (C) corresponding to the MB_type of the macroblock C is defined by the following equation (23). When included in a B picture, it is defined by the following equation (24).
ctx_mb_type_inter_field (C) = ((A == skip)? 0: 1) +2 ((B == skip)? 0: 1)
(23)
ctx_mb_type_inter_field (C) = ((A == Direct)? 0: 1) +2 ((B == Direct)? 0: 1)
... (24)

  However, the operators ((A == skip)? 0: 1) and ((A == skip)? 0: 1) in the expression (23) are the same as those in the expression (4), and the expression (24 ) Operators ((A == Direct)? 0: 1) and ((B == Direct)? 0: 1) are the same as those in equation (5). Adjacent macroblocks A and B may be field-based encoded or frame-based encoded.

  Note that the MB_type of the P picture that has not been binarized is binarized by the table shown in FIG. 28A. The MB_type of the B picture that has not been binarized is binarized by the table shown in FIG. 28B.

  In the adaptive binary arithmetic coding unit 96, probability estimation is performed on the binarized symbol by the probability estimating unit 97, and adaptive arithmetic coding based on the probability estimation is performed by the encoding engine 98. Since the relevant model is updated after the adaptive arithmetic coding process is performed, each model can perform the coding process according to the statistics of the actual image compression information.

  For macroblocks to be frame-based encoded, 10 types of MB_type are defined when belonging to a P picture. On the other hand, for a macroblock to be field-based encoded, when belonging to a P picture, the 16 × 16 mode and the 8 × 16 mode are not defined among the 16 types. That is, eight types of MB_type are defined for P pictures for field-based macroblocks.

  For macroblocks to be frame-based encoded, 18 types of MB_type are defined for B pictures. On the other hand, for a macroblock to be field-based encoded, when belonging to a B picture, among the 18 types, the forward 16 × 16 mode, the backward 16 × 16 mode, the forward 8 × 16 mode, and the backward Direction 8x16 mode is not defined. That is, 14 types of MB_type are defined for B pictures for field-based encoded macroblocks.

Next, a context model of motion vector information will be described. When the value of the frame / field flag is 1, the first to third context models ctx_mvd_field (C, k) corresponding to the motion vector information of the macroblock C are expressed by the following equations (25-1) to (25-3) Defined by
ctx_mvd_field (C, k) = 0
e _k (C) <3
... (25-1)
ctx_mvd_field (C, k) = 1
32 <e _k (C)
... (25-2)
ctx_mvd_field (C, k) = 2
3 ≦ e _k (C) ≦ 32
... (25-3) where the evaluation function e _k in Equation (25-1) to (25-3) is defined by the following equation (26). Macroblocks A and B are in the same parity field.
e _k (C) = | mvd _k (A) | + | mvd _k (B) |
... (26)

Here, when the macroblock A is frame-based encoded, regarding the motion vector information mvd ₁ (A) of the vertical direction component, mvd _{1_field} (A) calculated using the following equation (27) Applies to (26). The same applies to the case where the macroblock B is frame-based encoded.
mvd _{1_field} (A) = mvd _{1_frame} (A) / 2
... (27)

On the other hand, when the macroblock C is frame-based encoded and the adjacent block A is field-based encoded, the horizontal component and the vertical component of mvd _k (A) Mvd _{k_frame} (A) calculated using (28-1) and (28-2) is applied to equation (26).
mvd _{0_frame} (A)
= (Mvd _{0_top} (A) + mvd _{0_bottom} (A)) / 2 (28-1)
mvd _{1_frame} (A)
= Mvd _{1_top} (A) + mvd _{1_bottom} (A)
(28-2)

Next, a context model related to the reference field parameter will be described. When the value of frame / field flag is 1, the first context model ctx_ref_field_top (C) corresponding to the first field is defined by the following equation (29-1). Also, the first context model ctx_ref_field_bot (C) corresponding to the second field is defined by the following equation (29-2).
_{ctx_ref_field_top (C) = a t +} 2b t
... (29-1)
ctx_ref_field_bot (C) = a _b + 2b _b
(29-2)

However, the formula (29-1) and (29-2), the parameter a _t is related to the first field of the neighboring macroblock A, the parameter a _b is related to the second field of the neighboring macroblock A The parameter b _t relates to the first field of the adjacent macroblock B, the parameter b _b relates to the second field of the adjacent macroblock B, and the following equation (30-1), It is defined as (30-2).
a _t , a _b , b _t , b _b
= 0
The reference field is the most recently encoded one
... (30-1)
a _t , a _b , b _t , b _b
= 1
In cases other than the above
... (30-2)

  The context models corresponding to the second and subsequent bins are defined in the same manner as the context model ctx_ref_frame (C) shown in Expression (8). However, Code_number to be encoded is not assigned to the frame but assigned to the field.

  Next, a context model related to the intra prediction mode will be described. When the value of the frame / field flag is 1, the context model ctx_intra_pred_field (C) for the intra prediction mode corresponding to the macroblock C is defined in the same manner as the context model ctx_intra_pred (C) for the macroblock in the frame mode. Adjacent macroblocks A and B may be field-based encoded or frame-based encoded.

  As described above, by introducing a new context model and changing an existing context model, field / frame encoding using the CABAC method can be performed.

  Next, FIG. 29 shows a configuration example of an image information decoding apparatus corresponding to the image information encoding apparatus of FIG.

  In the image information decoding apparatus, the accumulation buffer 101 accumulates input image compression information and outputs it to the arithmetic decoding unit 102 as appropriate. The arithmetic decoding unit 102 performs arithmetic decoding processing on the compressed image information encoded based on the CABAC method, outputs the decoded frame / field flag to the field / frame conversion units 105 and 110, and is quantized. The orthogonal transform coefficient is output to the inverse quantization unit 103, and the prediction mode information and motion vector information are output to the motion prediction / compensation unit 111.

  The inverse quantization unit 103 inversely quantizes the quantized orthogonal transform coefficient decoded by the arithmetic decoding unit 102. The inverse orthogonal transform unit 104 performs inverse orthogonal transform on the inversely quantized orthogonal transform coefficient. The field / frame conversion unit 105 converts the output image or the difference image obtained as a result of the inverse orthogonal transform into a frame structure when the macro block to be processed is encoded on a field basis.

  When the macro block to be processed is an inter macro block, the adder 106 combines the difference image from the inverse orthogonal transform unit 104 and the reference image from the motion prediction / compensation unit 111 to generate an output image. The screen rearrangement buffer 107 rearranges the output images according to the GOP structure of the input image compression information and outputs the rearranged output images to the D / A conversion unit 108. The D / A converter 108 converts the output image, which is a digital signal, into an analog signal and outputs it to the subsequent stage.

  The frame memory 109 stores image information that is generated by the adder 106 and serves as a reference image. The field / frame conversion unit 110 converts the image information stored in the frame memory 111 into a field structure when the macroblock to be processed is encoded on a field basis. The motion prediction / compensation unit 111 generates a reference image based on the image information stored in the frame memory based on the prediction mode information and motion vector information for each macroblock included in the image compression information, and the addition unit 106 Output to.

  According to the image information decoding apparatus configured as described above, the original image information can be obtained by decoding the image compression information output by the image information encoding apparatus in FIG.

  The series of processes described above can be executed by hardware, but can also be executed by software. When a series of processing is executed by software, a program constituting the software may execute various functions by installing a computer incorporated in dedicated hardware or various programs. For example, it is installed in a general-purpose personal computer or the like from the recording medium 67 of FIG.

  The recording medium 67 is distributed to provide a program to the user separately from the computer, and includes a magnetic disk (including a flexible disk) on which the program is recorded, an optical disk (CD-ROM (Compact Disc-Read Only Memory). ), DVD (including Digital Versatile Disc), magneto-optical disc (including MD (Mini Disc)), or packaged media consisting of semiconductor memory, etc. It is composed of a ROM, hard disk, etc., where the program is recorded.

  In the present specification, the step of describing the program recorded in the recording medium is not limited to the processing performed in time series according to the described order, but is not necessarily performed in time series, either in parallel or individually. The process to be executed is also included.

  53 field / frame determination unit, 55 field / frame conversion unit, 58 arithmetic coding unit, 63 field / frame conversion unit, 66 control unit, 67 recording medium, 102 arithmetic decoding unit, 105 field / frame conversion unit, 110 field / Frame converter

Claims (4)

In a decoding device that decodes encoded data in which image information is encoded,
The motion of an adjacent macroblock adjacent to a target macroblock to be decoded using encoding mode information indicating at a macroblock level whether the image information is encoded in a frame mode or a field mode Context model means for converting a vector prediction error to match the encoding mode of the target macroblock and calculating a context model corresponding to the motion vector prediction error of the target macroblock;
A decoding apparatus comprising: context adaptive arithmetic decoding means for performing context adaptive arithmetic decoding on the encoded data using the context model calculated by the context model means. A decoding method of a decoding apparatus for decoding encoded data in which image information is encoded,
The context model means is adjacent to the target macroblock to be decoded using encoding mode information indicating at a macroblock level whether the image information is encoded in the frame mode or the field mode. Converting a motion vector prediction error of an adjacent macroblock to match the encoding mode of the target macroblock, and calculating a context model corresponding to the motion vector prediction error of the target macroblock;
A decoding method, wherein context adaptive arithmetic decoding means performs context adaptive arithmetic decoding on the encoded data using the calculated context model. A computer that decodes encoded data in which image information is encoded;
The motion of an adjacent macroblock adjacent to a target macroblock to be decoded using encoding mode information indicating at a macroblock level whether the image information is encoded in a frame mode or a field mode Context model means for converting a vector prediction error to match the encoding mode of the target macroblock and calculating a context model corresponding to the motion vector prediction error of the target macroblock;
A computer-readable recording medium storing a program for causing the encoded data to function as context adaptive arithmetic decoding means for performing context adaptive arithmetic decoding using the context model calculated by the context model means. A computer that decodes encoded data in which image information is encoded;
The motion of an adjacent macroblock adjacent to a target macroblock to be decoded using encoding mode information indicating at a macroblock level whether the image information is encoded in a frame mode or a field mode Context model means for converting a vector prediction error to match the encoding mode of the target macroblock and calculating a context model corresponding to the motion vector prediction error of the target macroblock;
A program for functioning as context adaptive arithmetic decoding means for performing context adaptive arithmetic decoding on the encoded data using the context model calculated by the context model means.

Images