0001]The present disclosure relates to an image processing apparatus and image processing method, in particular, when generating a predictive image of the block based on the motion vectors of two vertices of the block, it is possible to generate a predicted image of a rectangular block with high precision an image processing apparatus and image processing method capable.
BACKGROUND
[0002]ITU-T (International Telecommunication Union Telecommunication Standardization Sector) in JVET (Joint Video Exploration Team) to search the next generation video coding, inter-prediction processing by the reference image based on the motion vectors of the two vertices affine transformation ( performing the Affine motion compensation (MC) prediction) has been proposed (e.g., see non-Patent documents 1 and 2). Accordingly, when inter-prediction processing, translation (translation) and rotation movement between the screens, and can generate a predicted image by compensating for the change in shape such as enlargement or reduction.
[0003]
　Further, the JVET, as a technique for forming a CU (Coding Unit), a technique called QTBT (Quad tree plus binary tree) described in Non-Patent Document 3 is adopted. Thus, the shape of the CU, square well, there is a possibility that a rectangle.
CITATION
Non-patent literature
[0004]
Non-Patent Document 1: Jianle Chen et, "Algorithm Description of Joint Exploration Test Model 4 (JVET-C1001)", JVET of ITU-T SG16 WP3 and ISO / IEC JTC1 / SC29 / WG11,26 May-1 June 2016
Non-Patent Document 2: Feng Zou, "Improved affine motion prediction (JVET-C0062)", JVET of ITU-T SG16 WP3 and ISO / IEC JTC1 / SC29 / WG11,26 May-1 June 2016
non-Patent Document 3: "EE2.1 : Quadtree plus binary tree structure integration with JEM tools (JVET-C0024) ", JVET of ITU-T SG16 WP3 and ISO / IEC JTC1 / SC29 / WG11, 2016 Nian 5 Yue 16 Ri
Summary of the Invention
Problems that the Invention is to Solve
[0005]
　If PU (Prediction Unit) is the same rectangular block and CU, the affine transformation in the inter prediction process is performed based on two motion vectors of the vertices of the short sides of the PU, the two movements of the vertices of the long sides as compared with the case where on the basis of the vector, reduction in prediction accuracy due to errors of the motion vector becomes larger.
[0006]
　However, the position in PU of vertices corresponding to the two motion vectors used in the affine transformation of the inter prediction process, it has not been devised to change according to the shape of the PU. Therefore, if the shape of the PU is rectangular, there may not be able to generate a predictive image with high accuracy.
[0007]
　The present disclosure has been made in view of such a situation, when generating a predictive image of the block based on the motion vectors of two vertices of the blocks, and generates a prediction image of a rectangular block with high precision it is intended to allow.
Means for Solving the Problems
[0008]
　According to an embodiment of the present disclosure, based on vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the block, a prediction unit generating a prediction image of the block an image processing apparatus comprising.
[0009]
　According to another embodiment of the present disclosure corresponds to the image processing apparatus according to an embodiment of the present disclosure.
[0010]
　In one aspect of the present disclosure, based on the large size two motion vectors of the vertices aligned in the direction of one of the vertical and horizontal size of the block, the prediction image of the block is generated.
Effect of the invention
[0011]
　According to one aspect of the present disclosure, it is possible to generate a prediction image. According to an aspect of the present disclosure, may be in the case of generating a prediction image of the block based on the motion vectors of two vertices of the blocks, and generates a prediction image of a rectangular block with high accuracy.
[0012]
　Here, the advantages described in the present invention is not necessarily limited, it may be any of the effects described in the present disclosure.
BRIEF DESCRIPTION OF THE DRAWINGS
[0013]
It is a diagram for explaining inter prediction process for performing motion compensation on the basis of FIG. 1 one motion vector.
Is a diagram of the inter prediction process will be described for performing motion compensation on the basis of FIG. 2 rotation angles and one motion vector.
3 is a diagram of the inter prediction process will be described which performs motion compensation based on two motion vectors.
Is a diagram of the inter prediction process will be described for performing motion compensation on the basis of FIG. 4 the three motion vectors.
Is a diagram illustrating an affine transformation before and after the block based on FIG. 5 three motion vectors.
It is a diagram illustrating the FIG. 6] QTBT.
7 is a diagram for explaining inter prediction processing based on the two motion vectors for the rectangular PU.
8 is a diagram for explaining inter prediction processing based on the two motion vectors error occurs on the PU rectangle.
9 is a diagram for explaining inter prediction processing based on the three motion vectors for the rectangular PU.
It is a block diagram showing a configuration example of an embodiment of FIG. 10 the image coding apparatus.
11 is a view for explaining the two motion vector information.
12 is a diagram for explaining a neighboring vector.
[Figure 13] Affine flag is an example showing an area of CU 1.
[Figure 14] Affine flag is a diagram showing an example of a boundary of the CU region is 1.
[Figure 15] Affine flag is a diagram showing another example of the boundary of the CU region is 1.
It is a flowchart illustrating the FIG. 16 image encoding process.
17 is a flowchart illustrating a first example of the inter prediction operation mode setting process.
18 is a flowchart illustrating a second example of the inter prediction operation mode setting process.
Is a flowchart illustrating the FIG. 19 merge affine transformation mode encoding process.
It is a flow chart illustrating the FIG. 20] AMVP affine transformation mode encoding process.
Is a flow chart illustrating the FIG. 21] Affine flag encoding process.
22 is a block diagram showing a configuration example of an embodiment of an image decoding apparatus.
It is a flowchart illustrating the FIG. 23 image decoding processing.
Is a flowchart illustrating the FIG. 24 merge affine transformation mode decoding process.
It is a flow chart illustrating the FIG. 25] AMVP affine transformation mode decoding process.
FIG. 26 is a block diagram showing a configuration example of hardware of a computer.
FIG. 27 is a block diagram showing an example of a schematic configuration of a television device.
[FIG. 28] is a block diagram showing an example of a schematic configuration of a mobile phone.
FIG. 29 is a block diagram showing an example of a schematic configuration of a recording and reproducing apparatus.
It is a block diagram showing an example of a schematic configuration of a [30] an imaging device.
[FIG. 31] is a block diagram showing an example of a schematic configuration of a video set.
[FIG. 32] is a block diagram showing an example of a schematic configuration of the video processor.
[FIG 33 is a block diagram showing another example of a schematic configuration of the video processor.
FIG. 34 is a block diagram showing an example of a schematic configuration of a network system.
DESCRIPTION OF THE INVENTION
[0014]
　Hereinafter, embodiments of the premise and the disclosure of the present disclosure (hereinafter, referred to as embodiments) will be described. The description will be made in the following order.
　0. Premise of the present disclosure (FIGS.
　1-9) 1. First Embodiment: Image processing apparatus (FIGS. 10 to
　25) 2. Second Embodiment: computer (Fig.
　26) 3. Third Embodiment: a television device (Fig.
　27) 4. Fourth Embodiment: mobile phone (FIG.
　28) 5. Fifth Embodiment: recording and reproducing apparatus (FIG.
　29) 6. Sixth Embodiment: imaging device (Fig. 30)
　7. Seventh Embodiment: Video Set (FIGS. 31 to
　33) 8. Eighth Embodiment: Network System (Figure 34)
[0015]
　
　(Description of the inter prediction process for performing motion compensation based on one motion vector)
　FIG. 1 is a diagram for explaining inter prediction process for performing motion compensation based on one motion vector.
[0016]
　In the following, unless otherwise specified, the image lateral (picture) (horizontal) and x-direction, the vertical direction (vertical direction) and the y-direction.
[0017]
　As shown in FIG. 1, in the inter prediction process for performing motion compensation based on one motion vector, one motion vector v with respect to the prediction target PU11 (current block) c (v cx , v cy ) is determined that. Then, in the reference image at different times than the picture 10 containing PU11, motion vector v from the PU11 c only same size of the block 13 and PU11 present at a position apart, the motion vector v c translate based on by, predicted image PU11 is generated.
[0018]
　That is, in the inter prediction process for performing motion compensation based on one motion vector, an affine transformation is not performed for the reference image, the predicted image compensated only translational movement between the screens are generated. Furthermore, parameters used for the inter prediction process, v cx and v cy is two. Such inter-prediction processing is adopted in such AVC (Advanced Video Coding) or HEVC (High Efficiency Video Coding).
[0019]
　(One explanation of the inter prediction process for performing motion compensation based on a motion vector rotation angle)
　FIG. 2 is a diagram of the inter prediction process will be described for performing motion compensation on the basis of the rotation angle and the one motion vector.
[0020]
　As shown in FIG. 2, in the inter prediction process for performing motion compensation on the basis of the rotational angle and one motion vector is one motion vector v with respect to PU11 of the prediction target c (v cx , v cy ) and the rotation angle θ There are determined. Then, in the reference image at different times than the picture 10 containing PU11, motion vector v from the PU11 c only same size of the block 21 and the PU11 present in the inclination of the rotation angle θ at a distance, the motion vector v c by affine transformation based on the rotation angle θ and the predicted image of PU11 is generated.
[0021]
　That is, in the inter prediction process for performing motion compensation based on the rotational angle and one motion vector, an affine transformation is performed on the reference image based on the rotation angle and the one motion vector. Thus, the predicted image to compensate for translational movement and rotational movement between the screens are generated. Therefore, as compared with the inter prediction process for performing motion compensation based on one motion vector, thereby improving the accuracy of the prediction image. Furthermore, parameters used for the inter prediction process, v cx , v cy , and is three theta.
[0022]
　(Two explanations inter prediction process for performing motion compensation based on the motion vector)
　Fig. 3 is a diagram of the inter prediction process will be described which performs motion compensation based on two motion vectors.
[0023]
　As shown in FIG. 3, in the inter prediction process for performing motion compensation based on two motion vectors for PU31 prediction target, the motion vector v at the apex A of the upper left PU31 0 (v 0x , v 0y ) and top right of the motion vector v at vertex B 1 (v 1x , v 1y ) are determined.
[0024]
　Then, in the reference image at different times than the picture including the PU31, motion vector v from the vertex A 0 and the upper left vertex of the A'points apart, the motion vector v from the vertex B 1 upper right B'point distant by the block 32, the vertex of the motion vector v 0 vector v motion and 1 by affine transformation based on the predicted image PU31 is generated.
[0025]
　Specifically, PU31 is of a predetermined size block (hereinafter, referred to as motion compensation unit block) is divided into. Then, the motion vector v (v in each motion compensation unit block x , v y ) is the motion vector v 0 (v 0x , v 0y ) and the motion vector v 1 (v 1x , v 1y ) on the basis of the following formula (1) it is determined.
[0026]
[Number 1]

[0027]
　Note, W is the size of the x-direction of PU31, H is the y-direction size of PU31. Therefore, if PU31 is square, W and H are equal. Also, x, y, respectively, x-direction, y position of the motion compensation unit block. According to equation (1), based on the position of the motion compensation unit block, the motion vector v of the motion compensation unit block is determined.
[0028]
　Then, in the reference image, each motion compensation from the unit block of the motion vector v apart motion compensation unit block and same-size block, by translation on the basis of the motion vector v, the prediction of the motion compensation unit block to generate an image.
[0029]
　As described above, in the inter prediction process for performing motion compensation based on two motion vectors, the affine transformation is performed on the reference image on the basis of the two motion vectors. This not only movement of the translational and rotational direction of the inter-picture prediction image to compensate for change in shape of such expansion and contraction is generated. Therefore, as compared with the inter prediction process for performing motion compensation on the basis of the rotational angle and one motion vector, it improved the accuracy of the prediction image. Furthermore, parameters used for the inter prediction process, v 0x , v 0y , v 1x , and v 1y is four. Such inter-prediction processing is adopted in JEM (Joint Exploration Model) Referring software.
[0030]
　Note that the affine transformation based on two motion vectors are affine transformation before and after affine transformation block is assumed to be rectangular. In order to perform affine transformation even when affine transformation before and after the block is a square other than rectangular, it requires three motion vectors.
[0031]
　(3 describes the inter prediction process for performing motion compensation based on the motion vector)
　FIG. 4 is a diagram for explaining inter prediction process for performing motion compensation based on three motion vectors.
[0032]
　As shown in FIG. 4, in the inter prediction process for performing motion compensation based on three motion vectors for PU31 prediction target, the motion vector v 0 (v 0x , v 0y ) and the motion vector v 1 (v 1x , v 1y ) as well as the lower left apex C of the motion vector v 2 (v 2x , v 2y ) is determined.
[0033]
　Then, in the reference image at different times than the picture including the PU31, motion vector v from the vertex A 0 and the upper left vertex of the A'points apart, the motion vector v from the vertex B 1 upper right B'point distant by an apex of the motion vector v from the vertex C 2 the block 42 to the C'point distant by a lower left vertex of the motion vector v 0 to v 2 by affine transformation based on the predicted image PU31 is generated that.
[0034]
　That is, in the inter prediction process for performing motion compensation based on three motion vectors, the affine transformation is performed on the reference image based on three motion vectors. Thus, the block 42 or translation (Translation) as shown in A of FIG. 5, or skew (Skew), as shown in B of FIG. 5, rotated (Rotation) as shown in C in FIG. 5 or, or to enlarge or reduce (Scaling) as shown in D of FIG.
[0035]
　As a result, translation and rotation movement between the screens, as well as enlargement or reduction, and a prediction image obtained by compensating for the change in shape such skew is generated. In FIG. 5, a block 42 before the affine transformation by the solid line, shows a block 42 after the affine transformation by the dotted line.
[0036]
　Shape contrast, in the inter prediction process for performing motion compensation based on two motion vectors described with reference to FIG. 3, the prediction image, translational movement and rotational movement between the screens, as well, such as enlargement or reduction can compensate for the change, it is impossible to compensate for the skew. Accordingly, in inter prediction process for performing motion compensation based on three motion vectors, compared to the inter prediction process for performing motion compensation based on two motion vectors, thereby improving the accuracy of the prediction image.
[0037]
　However, in inter prediction process for performing motion compensation based on three motion vectors, parameters used in the inter prediction process, v 0x , v 0y , v 1x , v 1y , v 2x , and v 2y is six . Thus, one motion vector and the rotation angle or, as compared to inter prediction process for performing motion compensation based on two motion vectors, the number of parameters used in the inter prediction process is increased. Thus, the prediction accuracy of the improvement and overhead suppression of inter prediction process using the affine transformation, a tradeoff relationship.
[0038]
　Therefore, in JVET, the control signals, techniques for switching the inter prediction process for performing motion compensation on the basis of the inter prediction operation and three motion vectors for performing the motion compensation based on two motion vectors is proposed.
[0039]
　(QTBT description)
　in the old image encoding method such as MPEG2 (Moving Picture Experts Group 2 ( ISO / IEC 13818-2)) and AVC, encoding processing is executed in the process units called macroblocks. Macro block is a block having a uniform size of 16x16 pixels. In contrast, in HEVC, coding process is performed by processing units called CU (coding unit). CU is the maximum coding unit LCU and (Largest Coding Unit) is formed by dividing recursively, a block having a variable size. The maximum size of the selectable CU is 64x64 pixels. The minimum size of the selectable CU is 8x8 pixels. The minimum size of the CU is referred to as the SCU (Smallest Coding Unit). The maximum size of the CU is not limited to 64x64 pixels, larger 128x128 pixels, or as a block size, such as 256x256 pixels.
[0040]
　Thus, the results of CU having a variable size is employed, in HEVC, it is possible to adjust the image quality and encoding efficiency adaptively depending on the content of the image. Prediction process for predictive coding is performed by the processing units called PU. PU is formed by dividing the CU in one of several split pattern. Further, PU is composed of processing units called luminance (Y) and color difference (Cb, Cr) for each of PB (Prediction Block). Further, orthogonal transform processing is executed in the process units called TU (Transform Unit). TU is formed by dividing to a depth in the CU or PU. Moreover, TU consists luminance (Y) and color difference (Cb, Cr) for each of TB (Transform Block) and referred to the processing unit (transformation blocks).
[0041]
　In the following, there are (not a block of processing unit) When described with reference to "block" as the partial area and processing unit of an image (picture). The "block" in this case refers to any partial area in the picture, its size, shape, and characteristics, etc. are not limited. That is, the "block" in this case, for example, TB, TU, PB, PU, ​​SCU, CU, LCU (CTB), the sub-block, macroblock, tiles, or slice, etc., any partial region (processing units), It is intended to be included.
[0042]
　Figure 6 is a diagram illustrating a QTBT adopted in JVET.
[0043]
　In HEVC, it can be only 4 (= 2x2) subblocks by dividing one block in the horizontal and vertical directions. In contrast, in QTBT, one block 4 (= 2x2) not only sub-blocks, the horizontal direction and only divided into one of the vertical 2 (= 1x2,2x1) sub-blocks it can also be in. That is, in QTBT, formation of CU is carried out by repeating a division into four or two sub-blocks of a block recursively, resulting quadtree (Quad-Tree) form, or, 2 min tree (Binary-tree) like a tree structure is formed. In the following, PU and TU is assumed to be identical to the CU.
[0044]
　(Rectangular Description of the inter-prediction processing based on the two motion vectors for the PU)
　7 and 8 are diagrams for explaining inter prediction processing based on the two motion vectors for the rectangular PU.
[0045]
　In the example of FIG. 7, PU61 prediction target is a vertical rectangle size H in the y direction is larger than the size W in the x direction. In this case, the PU61, as in the case of FIG. 3, when the inter prediction process for performing motion compensation based on two motion vectors is performed, as shown in FIG. 7, a different time from the picture including the PU61 the block 62 in the reference image, the motion vectors v 0 vector v motion and 1 by affine transformation based on the predicted image PU61 is generated. Incidentally, the block 62, the motion vector v from the vertex A 0 the A'point distant by a top left vertex of the motion vector v from the vertex B 1 is a block to the B'point away by the upper right vertex.
[0046]
　Here, as shown in FIG. 8, the motion vector v 0 error e to 0 is generated, the motion vectors v 1 error e to 1 if occurs, a block 71 in the reference image, the motion vectors v 0 + e 0 and motion vector v 1 + e 1 by affine transformation based on the predicted image PU61 is generated. Incidentally, the block 71, the motion vector v from the vertex A 0 + e 0 the A'' apart points the upper left corner of the motion vector v from the vertex B 1 + e 1 is the B'' apart points the upper right corner of the it is a block.
[0047]
　Error of the motion vector v for each motion compensation block PU61 the motion vector v is used in the calculation of the motion vector v 0 error e 0 vector v motion with 1 error e 1 affected. Further, the effect is, the motion vector v 0 vertex A and a motion vector v corresponding to 1 farther from the apex B corresponding to large.
[0048]
　In the example of FIGS. 7 and 8, the apex A and apex B, since that appears in the x direction, which is the short side direction of PU61, distance of the vertex C and vertex A opposite the apex A, and the vertex B distance of the vertex D and vertex B opposite to the larger.
[0049]
　Therefore, displacement of the block 62 and the block 71 is increased. Therefore, it decreases the accuracy of the prediction image, residual between the predicted image and the PU61 increases. As a result, if the orthogonally transformed residual is not zero by the quantization, it exacerbated the coding efficiency of the encoded stream containing residual after quantization. Also, if the orthogonally transformed residual is zero by the quantization, since the accuracy of the predictive image is reduced, the image quality of the decoded image is degraded.
[0050]
　(Description of the inter prediction process based on three motion vectors for the rectangular PU)
　FIG. 9 is a diagram for explaining inter prediction processing based on the three motion vectors for the rectangular PU.
[0051]
　Against PU61 vertically long rectangle, as in the case of FIG. 4, when the inter prediction process for performing motion compensation based on three motion vectors is performed, as shown in FIG. 9 differs from the picture including the PU61 the block 72 in the time of the reference image, the motion vectors v 0 through v 2 by affine transformation based on the predicted image PU61 is generated. Incidentally, the block 72, the motion vector v from the vertex A 0 and the A'point distant by the upper left corner of the motion vector v from the apex B 1 of B'point distant by a upper right corner of the motion vector v from the vertex C 2 the C'only away point is a block to the lower left vertex of.
[0052]
　Here, as shown in FIG. 9, the motion vector v 0 to v 3 , respectively error e to 1 or e 2 If occurs, the block 73 in the reference image, the motion vectors v 0 + e 0 , v 1 + e 1 , and v 2 + e 2 by affine transformation based on the predicted image PU61 is generated. Incidentally, the block 73, the motion vector v from the vertex A 0 + e 0 the A'' apart points the upper left corner of the motion vector v from the vertex B 1 + e 1 is the B'' apart points the upper right corner of, motion vector v from the vertex C 2 + e 2 a C'' spaced points a block to the lower left vertex.
[0053]
　In this case, the motion vector v 2 + e 2 , as in the case of FIG. 8, it is possible to prevent the error of the motion vector v becomes larger toward the lower side of the motion compensation blocks in PU61.
[0054]
　However, as described above, in inter prediction processing based on the three motion vectors, since the number of parameters is six, overhead is increased, the coding efficiency decreases. Accordingly, in the present disclosure, based on the magnitude relation of the size H and the size W, by changing the position of the vertex corresponding to the two motion vectors, thereby improving the prediction accuracy of inter prediction processing based on the two motion vectors.
[0055]
　
　(Configuration example of an image coding apparatus)
　FIG. 10 is a block diagram showing a configuration example of an embodiment of an image coding apparatus as an image processing apparatus according to the present disclosure. The image coding apparatus 100 of FIG. 10, as in the AVC and HEVC, an apparatus for encoding a prediction residual of the image and its prediction picture. For example, the image coding apparatus 100, technique or HEVC, implementing the proposed technique in JVET.
[0056]
　Incidentally, in FIG. 10 shows the main ones, such as the flow of the processing unit and data is not all that shown in Figure 10. That is, in the image encoding apparatus 100, or there is processing unit not shown as a block in FIG. 10 may be or there is a flow of processing and data not shown as arrows or the like in FIG. 10.
[0057]
　The image coding apparatus 100 of FIG. 10, the control unit 101, arithmetic unit 111, conversion unit 112, a quantization unit 113, the coding unit 114, an inverse quantization unit 115, inverse transform unit 116, arithmetic unit 117, frame memory 118 , and a prediction unit 119. The image encoding apparatus 100 performs encoding for each CU against a moving image of a frame to be input picture.
[0058]
　Specifically, the control unit 101 of the image encoding apparatus 100 based on the input and RDO (Rate-Distortion Optimization) or the like from the outside, coding parameter (header information Hinfo, prediction information pinfo, conversion information Tinfo etc.) to set.
[0059]
　Header information Hinfo, for example, a video parameter set (VPS (Video Parameter Set)), sequence parameter set (SPS (Sequence Parameter Set)), picture parameter set (PPS (Picture Parameter Set)), the slice header (SH), such as including the information. For example, the header information Hinfo, the image size (width PicWidth, longitudinal width PicHeight), bit depth (brightness BitDepthY, chrominance BitDepthC), includes information defining such maximum MaxCUSize / minimum value MinCUSize the CU size. Of course, the contents of the header information Hinfo is optional, any information other than the examples described above may be included in the header information Hinfo.
[0060]
　The prediction information pinfo, for example, include split flag indicating the presence or absence of division in the horizontal direction or vertical direction in each divided hierarchy at the time of forming the PU (CU). Further, the prediction information pinfo, each PU, or prediction processing of the PU is intra prediction processing, or include mode information pred_mode_flag indicating which inter prediction process.
[0061]
　When the mode information pred_mode_flag indicates inter prediction process, the prediction information Pinfo, Merge flag, Affine flag, etc. The motion vector information, reference image identification information specifying the reference image. Merge flag, the mode of inter prediction process, either a merge mode, or is information indicating which AMVP mode. The merge mode, the motion vector generated based on the motion vector of the coded adjacent PU adjacent to PU to be processed (hereinafter, referred to as the neighborhood vector) based on the prediction vector selected from among the candidates including it is a mode to perform the inter prediction processing. The AMVP mode is a mode for performing inter prediction processing based on the motion vector of the PU to be processed. Merge flag is 1 to indicate that a merge mode, a 0 to indicate that a AMVP mode.
[0062]
　Affine flag, in the inter prediction operation, whether to perform motion compensation in the affine transformation mode, or is information indicating whether to perform motion compensation in translation mode. The translation mode is a mode for performing motion compensation by performing a translational movement relative to the reference image based on one motion vector. The affine transformation mode is a mode for performing motion compensation by performing affine transformation on the reference image based on two motion vectors. Affine flag (multiple vector prediction information) is 1 to indicate that performing motion compensated affine transformation mode, a 0 to indicate that performing motion compensation in translation mode.
[0063]
　Motion vector information, if Merge flag is 1, the predicted vector information identifying the predictive vector from the candidate including neighboring vector, if Merge flag is 0, the prediction vector information, and the processing and the prediction vector which is the difference between the motion vector of the target of PU. Also, if Affine flag is 1, the prediction information pinfo, 2 one motion vector information is included, if Affine flag is 0, one motion vector information is included.
[0064]
　When the mode information pred_mode_flag indicates an intra prediction process, the prediction information pinfo, etc. intra prediction mode information indicating the intra prediction mode is the mode of the intra prediction process. Of course, the content of the prediction information pinfo is optional, any information other than the examples described above may be included in the prediction information pinfo.
[0065]
　The conversion information Tinfo include such TBSize indicating the size of the TB. Of course, the content of the conversion information TINFO is optional, any information other than the examples described above may be included in the conversion information TINFO.
[0066]
　Calculation unit 111, the order of the encoding target picture to picture to be inputted, based on the split flag of prediction information pinfo, sets the encoding target CU (PU, TU) for the picture to be encoded. Calculation unit 111 from the encoding target PU images I (current block), obtains the prediction residual D by subtracting the predicted image P of the supplied from the prediction unit 119 PU (prediction block), a conversion unit it and supplies to 112.
[0067]
　Converter 112, based on the conversion information Tinfo supplied from the control unit 101 performs orthogonal conversion on the prediction residual D supplied from the arithmetic unit 111, and the transform coefficients Coeff. Conversion unit 112 supplies the transform coefficients Coeff to the quantization unit 113.
[0068]
　Quantization unit 113, based on the conversion information Tinfo supplied from the control unit 101, the transform coefficient Coeff supplied from the conversion unit 112 scales (quantized), to derive a quantized transform coefficient levels level. Quantization unit 113 supplies the quantized transform coefficient level level to encoding section 114 and the inverse quantization unit 115.
[0069]
　Encoding unit 114 encodes the quantized transform coefficient levels level like supplied from the quantization unit 113 in a predetermined manner. For example, the coding unit 114, along with the definition of the syntax table, coding parameters supplied from the control unit 101 (header information Hinfo, prediction information pinfo, conversion information Tinfo etc.) are supplied from the quantization unit 113 the that quantized transform coefficient levels level, converted into the syntax values ​​of each syntax element. The encoding unit 114 encodes each syntax value (e.g., arithmetic coding such as CABAC (Context-based Adaptive Binary Arithmetic Coding)) is.
[0070]
　Encoding unit 114, for example, the coded data is a bit string of each syntax element obtained as a result of coding and multiplexing, and output as an encoded stream.
[0071]
　Inverse quantization unit 115, based on the conversion information Tinfo supplied from the control unit 101, the scaling values ​​of the quantized transform coefficient levels level supplied from the quantization unit 113 (inverse quantization) and, after dequantization to derive the conversion coefficient Coeff_IQ. Inverse quantization unit 115 supplies the transform coefficients Coeff_IQ inverse transformation unit 116. Inverse quantization performed by the inverse quantization unit 115, an inverse process of the quantization performed by the quantization unit 113 is the same processing as the inverse quantization performed in the image decoding apparatus described later.
[0072]
　Inverse transform unit 116, based on the conversion information Tinfo supplied from the control unit 101 performs inverse orthogonal conversion on the transform coefficients Coeff_IQ supplied from the inverse quantization unit 115 derives a prediction residual D ' . Inverse transform unit 116 supplies the prediction residual D 'to the arithmetic unit 117. Inverse orthogonal transform performed by the inverse transformation unit 116, an inverse process of the orthogonal transformation performed by the transformation unit 112, an inverse orthogonal transformation and the same processing performed in the image decoding apparatus described later.
[0073]
　Calculation unit 117, the prediction residual D supplied from the inverse transform unit 116 'and is supplied from the prediction unit 119, the prediction residual D' local decoded image by adding the predicted image P corresponding to the to derive the Rec. Operation unit 117 supplies the local decoded image Rec in the frame memory 118.
[0074]
　Frame memory 118, reconstructs the decoded image in units of pictures by using a local decoded image Rec supplied from the arithmetic unit 117 is stored in the buffer in the frame memory 118. The frame memory 118 is read out from the buffer the decoded image is designated by the prediction unit 119 as a reference image, and supplies the prediction unit 119. The frame memory 118, the header information Hinfo according to generation of the decoded image, the prediction information pinfo, and conversion information TINFO, may be stored in a buffer in the frame memory 118.
[0075]
　Prediction unit 119, based on the mode information pred_mode_flag prediction information pinfo, acquires the decoded image stored coded the CU same time in the frame memory 118 as a reference picture. Then, the prediction unit 119 uses the reference image, with respect to PU to be coded, performs intra prediction processing of the intra-prediction mode indicated by the intra prediction mode information.
[0076]
　Also, the prediction unit 119, based on the mode information pred_mode_flag the reference image identification information of the prediction information pinfo, acquires a reference image decoded image at different times from the CU of stored coded in the frame memory 118. Prediction unit 119, Merge flag, based on Affine flag, and motion vector information, performs motion compensation in translation mode or affine conversion mode for the reference image, it performs the inter prediction process of the merge mode or AMVP mode.
[0077]
　Prediction unit 119 supplies the predicted image P of the encoding target PU generated as a result of the intra prediction processing or the inter prediction operation to the operation unit 111 and operation unit 117.
[0078]
　(Two motion description of vector information)
　FIG. 11 is a diagram illustrating the two motion vector information is set based on the RDO by the control unit 101.
[0079]
　As shown in A of FIG. 11, when PU121 prediction target is a horizontally long rectangular x-direction size W is larger than the size H in the y-direction, the control unit 101, based on the RDO, the upper left PU121 movement of the vertex a vector v 0 motion vector v of the upper right apex B 1 sets the motion vector information. That is, the control unit 101, based on the RDO, size larger W motion vector v of the two vertices A and vertex B arranged in the x direction which is the direction of one of the H size W 0 and v 1 of the motion vector information to set.
[0080]
　Thus, the prediction unit 119, motion vector v corresponding to the two motion vector information block 122 in the reference image at different times, it sets the PU121 0 vector v motion and 1 by affine transformation based on, to generate a prediction image of PU121. Incidentally, the block 122, the motion vector v from the vertex A 0 the A'point distant by a top left vertex of the motion vector v from the vertex B 1 is a block to the B'point away by the upper right vertex.
[0081]
　Here, as shown in A of FIG. 11, the motion vector v 0 error e to 0 is generated, the motion vectors v 1 error e to 1 when occurs, the prediction unit 119, a block 123 in the reference image, motion vector v 0 + e 0 vector v motion and 1 + e 1 by affine transformation based on, and generates a prediction image of PU121. Incidentally, the block 123, the motion vector v from the vertex A 0 + e 0 the A'' apart points the upper left corner of the motion vector v from the vertex B 1 + e 1 is the B'' apart points the upper right corner of the it is a block.
[0082]
　Error of the motion vector v for each motion compensation block PU121 the motion vector v is used in the calculation of the motion vector v 0 error e 0 vector v motion with 1 error e 1 affected. Further, the effect is, the motion vector v 0 vertex A and a motion vector v corresponding to 1 farther from the apex B corresponding to large.
[0083]
　However, in A of FIG. 11, the vertex A and vertex B, since that appears in the x direction is a longitudinal direction of the PU121, distance of the vertex C and vertex A opposite the apex A, and faces the vertex B distance of the vertex D and vertex B is small. Therefore, the deviation of the block 122 and the block 123, as compared with a case where the affine transformation based on the motion vector of the vertex A and vertex C arranged in the short side direction of PU121, smaller.
[0084]
　On the other hand, as shown in B of FIG. 11, when PU131 prediction target is a vertical rectangle size H in the y direction is larger than the size W in the x-direction, the control unit 101, based on the RDO, the PU131 movement of the upper left vertex a vector v 0 motion vector v of the lower left vertex C 2 to set the motion vector information. That is, the control unit 101, based on the RDO, size W and size larger two vertices A and movement of the vertex C vectors v aligned in the y direction is the direction of the H of the H 0 and v 2 motion vector information to set.
[0085]
　Thus, the prediction unit 119, motion vector v corresponding to the two motion vector information block 132 in the reference image at different times, it sets the PU131 0 vector v motion and 2 by affine transformation based on, to generate a prediction image of PU131. Note that block 132, the motion vector v from the vertex A 0 the A'point distant by a top left vertex of the motion vector v from the vertex C 2 is a block to the C'point distant by a lower left vertex.
[0086]
　Here, as shown in B of FIG. 11, the motion vector v 0 error e to 0 is generated, the motion vector v 2 to the error e 2 If occurs, the prediction unit 119, a block 133 in the reference image, motion vector v 0 + e 0 vector v motion and 2 + e 2 by affine transformation based on, and generates a prediction image of PU131. Incidentally, the block 133, the motion vector v from the vertex A 0 + e 0 the A'' apart points the upper left corner of the motion vector v from the vertex C 2 + e 2 is the C'' apart points and lower left vertex of the it is a block.
[0087]
　In this case, the motion vector v (v in each motion compensation block PU131 x , v y ) is determined by the following equation (2), the error of the motion vector v, the motion vectors used in the calculation of the motion vector v V 0 error e of 0 and a motion vector V 2 error e of 2 affected by the. Further, the effect is, the motion vector v 0 vertex A and a motion vector v corresponding to 2 farther from the vertex C corresponding to large.
[0088]
[Number 2]

[0089]
　However, in B of FIG. 11, the vertex A and vertex C, since that appears in the y direction is a longitudinal direction of the PU131, distance vertex B and vertex A opposite the apex A, and faces the vertex C distance of the vertex D and vertex C is small. Therefore, displacement of the block 132 and the block 133, as compared with a case where the affine transformation based on the motion vector of the vertex A and vertex B arranged in the short side direction of PU131, smaller.
[0090]
　Incidentally, the motion vector v 0 to v 2 If the errors are not generated, the motion vector v 0 Vector v motion between 1 and predicted image generated by the affine transformation based on the motion vector v 0 Vector v motion and 2 based on predicted image generated by the affine transformation is the same.
[0091]
　(Description of the adjacent vectors)
　FIG. 12 is a diagram illustrating a neighbor vectors, which are candidates of the prediction vector.
[0092]
　Prediction unit 119, FIG. 12 the motion vector v of the upper left vertex A of the prediction of the target PU151 of 0 prediction vector pv of 0 adjacent vectors, which are candidates of already coded in the upper left of PU151 whose vertices apex A of PU blocks a, is generated based on the motion vector of the block c is a block b or left coded PU, which is above the coded PU is.
[0093]
　Also, the prediction unit 119, the motion vector v upper right apex B of PU151 1 prediction vector pv of 1 adjacent vectors, which are candidates, the upper coded PU of PU151 whose vertices vertices B block d or is generated based on the motion vector of the block e is the upper right side of the coded PU.
[0094]
　Prediction unit 119, motion vector v of the vertex C 2 predicted vector pv of 2 neighboring vectors, which are candidates of a left coded PU of PU151 whose vertices vertices C block f, or lower left side generated based on the motion vector of the block g is coded PU. Note that the motion vector of the block a to g is held in the prediction unit 119, which is one of the motion vectors for the block.
[0095]
　Thus, the prediction vector pv 0 to pv 2 candidate combinations of motion vectors used to generate the neighboring vectors, which are candidates of, 12 (= 3 × 2 × 2) streets. Prediction unit 119, among the candidates for the combination of the 12 kinds, combinations that DV obtained by the following equation (3) becomes minimum, the prediction vector pv 0 to pv 2 used to generate a neighbor vectors, which are candidates of selecting as a combination of a motion vector.
[0096]
[Number 3]

[0097]
Incidentally, v 0x ', v 0y ', the prediction vector pv 0 either the x-direction of blocks a to c used to generate the a y direction of the motion vector. v 1x ', v 1y ' is predicted vector pv 1 either x direction of the block d and e are used to generate a y-direction of the motion vector. v 2x ', v 2y ', the prediction vector pv 2 either x direction of the block f and g used for generating a y-direction of the motion vector.
[0098]
　According to equation (3), the motion vector v 0 '(v 0x ', v 0y ') to v 2 ' (v 2x ', v 2y ') by affine transformation based, the affine transformation based on two motion vectors If anything other than a non-skew is performed, DV is reduced.
[0099]
　(Affine Description of coding flag)
　FIG. 13 is an example showing an area of CU (PU) is Affine flag is 1.
[0100]
　13, the white rectangles in the image 170, Affine flag represents CU (PU) is 0, the hatched rectangle, Affine flag represents CU (PU) 1. Further, in FIG. 13, for the sake of clarity, it shows only a part of the CU in the image 170.
[0101]
　As shown in FIG. 13, region 171 of CU (PU) Affine flag in the image 170 is 1, it is presumed to be present collectively.
[0102]
　Thus, for example, as shown in A of FIG. 14, if the size W is horizontally long PU191 exists larger than the size H, the block a to e are adjacent in the x-direction of the upper side of the vertex A and vertex B of PU191 when Affine flag is 1, it is likely the lower side of the PU191 is a boundary 192 of the region 171. Therefore, it is likely Affine flag is one-PU191.
[0103]
　Further, as shown in B of FIG. 14, when Affine flag of block f and g is adjacent to the vertex C of the x-direction of the lower side of the PU191 is 1, it is likely the upper side of the PU191 is a boundary 192. Therefore, it is likely Affine flag is one-PU191.
[0104]
　On the other hand, as shown in A of FIG. 15, the size if the H is elongated PU193 exists larger than the size W, blocks a to c is adjacent to the vertex A and vertex C of the left side of the y direction PU193, f, and when g of Affine flag is 1, it is likely the right side of PU193 is a boundary 194 of the region 171. Therefore, it is likely Affine flag is one-PU193.
[0105]
　Further, as shown in B of FIG. 15, when Affine flag of block d and e adjacent to the vertex B in the y direction of the right side of PU193 is 1, it is likely the left side of PU193 is a boundary 194. Therefore, it is likely Affine flag is one-PU193.
[0106]
　Thus, the coding unit 114, Affine flag adjacent PU adjacent to large vertex of the side in the direction of the size of the x-direction size W in the y direction size H of PU (CU) is whether a 1 based on, it switches the context of CABAC probabilistic model of Affine flag for that PU.
[0107]
　Specifically, the encoding unit 114, when encoding with CABAC the Affine flag of oblong PU191, blocks a to e, or, in Affine flag is 1 having a predetermined number or more blocks of the blocks f and g On one occasion, Affine flag is used as the context probability models are likely to be 1.
[0108]
　Meanwhile, blocks a to e, or, when Affine flag of a predetermined number less than the blocks of the blocks f and g is 1, the coding unit 114, a probability model that Affine flag is not likely a 1 used as a context.
[0109]
　Also, when encoding with CABAC the Affine flag of elongated PU193, coding section 114, block a to c, f, and g, or, the Affine flag of a predetermined number or more blocks of the blocks d and e when it is 1, we use the Affine flag is likely to be 1 as the context probability models.
[0110]
　On the other hand, when the block a to c, f, and g, or, the Affine flag of a predetermined number less than the blocks of the blocks d and e is 1, the coding unit 114, the potential Affine flag is 1 low that used as the context probability models.
[0111]
　Furthermore, if the PU is a square, the coding unit 114, when Affine flag of a predetermined number or more blocks of the blocks a to e is 1, a probability model that Affine flag is likely to be 1 used as a context.
[0112]
　On the other hand, when the Affine flag of a predetermined number less than the block of the blocks a to e is 1, the coding unit 114 uses the Affine flag is not likely a 1 as the context probability models.
[0113]
　Then, the encoding unit 114, when encoding with CABAC the Affine flag, a probability model of the CABAC, probability is 1 increases with the Affine flag is likely to be 1 as the context probability models performing encoding set such. Accordingly, the code amount in the case Affine flag is 1, smaller than the code amount in the case Affine flag is 0.
[0114]
　The encoding unit 114, when encoding with CABAC the Affine flag using the Affine flag is not likely a 1 as the context, setting a probability model of the CABAC, as the probability is zero is high coding is performed with. Accordingly, the code amount in the case Affine flag is zero, smaller than the code amount in the case Affine flag is 1.
[0115]
　As a result, the encoding unit 114 reduces the amount of code of Affine flag is overhead, it is possible to improve the coding efficiency.
[0116]
　The number of blocks Affine flag is 1 in whether more than a predetermined number, instead of switching the context, the number of blocks Affine flag is 1, may be switched context. In this case, for example, Affine flag in response to the number of blocks is 1, the probability is 1 probabilistic model of CABAC is changed.
[0117]
　The encoding unit 114, based on Affine flag blocks a to g, instead of switching the context probability model CABAC, may be switched code (bit string) to be assigned to Affine flag.
[0118]
　In this case, the coding unit 114, a probability model of the CABAC, instead of setting so that the probability is 1 increases, the code length of the code to be assigned to Affine flag is 1 (bit length) is 0 Affine shorter than the flag. The encoding unit 114, a probability model of the CABAC, instead of setting so that the probability is zero is high, the code length of a code to be assigned to Affine flag is 0, shorter than the Affine flag is 1 to.
[0119]
　(Process description of the image processing apparatus)
　FIG. 16 is a flowchart for explaining an image coding process of the image encoding apparatus 100 of FIG. 10.
[0120]
　In step S11 in FIG. 16, the control unit 101 based on the input and RDO or the like from the outside, to set the encoding parameters (header information Hinfo, prediction information pinfo, conversion information Tinfo etc.). Control unit 101 supplies the encoding parameter set to each block.
[0121]
　In step S12, the prediction unit 119 determines whether the mode information pred_mode_flag prediction information Pinfo indicates the inter prediction process. If it is determined that indicates the inter prediction process in step S12, in step S13, the prediction unit 119 determines whether the Merge flag of prediction information Pinfo is 1.
[0122]
　If the Merge flag is determined to be 1 in step S13, in step S14, the prediction unit 119 determines whether Affine flag is 1 prediction information pinfo. If Affine flag is determined to be 1 in step S14, the process proceeds to step S15.
[0123]
　In step S15, the prediction unit 119 performs motion compensation by affine transformation mode, merging affine transformation encoding an image I to be coded using the predicted image P generated by performing inter prediction process of the merge mode the mode coding process. Details of the merge affine transformation mode encoding processing will be described with reference to FIG. 19 described later. After completion of the merge affine transformation mode encoding processing, the image coding process is ended.
[0124]
　On the other hand, if the Affine flag is determined not to be 1 at step S14, i.e., if the Affine flag is 0, the process proceeds to step S16.
[0125]
　In step S16, the prediction unit 119 performs motion compensation in translation mode, merge mode code encoding an image I to be coded using the predicted image P generated by performing inter prediction process of the merge mode carry out the process. After completion of the merge mode coding, the image coding process is ended.
[0126]
　Also, if the Merge flag is determined not to be 1 at step S13, i.e., if the Merge flag is 0, in step S17, the prediction unit 119 determines whether Affine flag is 1 prediction information Pinfo . If Affine flag is determined to be 1 in step S17, the process proceeds to step S18.
[0127]
　In step S18, the prediction unit 119 performs motion compensation by affine transformation mode, AMVP affine transformation encoding an image I to be coded using the predicted image P generated by performing inter prediction process AMVP mode the mode coding process. Details of this AMVP affine transformation mode encoding processing will be described with reference to FIG. 20 to be described later. AMVP affine transformation mode encoding process after the completion of the image coding process is ended.
[0128]
　On the other hand, if the Affine flag is determined not to be 1 at step S17, i.e., if the Affine flag is 0, the process proceeds to step S19.
[0129]
　In step S19, the prediction unit 119 performs motion compensation in translation mode, AMVP mode code encoding an image I to be coded using the predicted image P generated by performing inter prediction process AMVP mode carry out the process. After completion of AMVP mode encoding processing, the image coding process is ended.
[0130]
　When it is determined not to show the inter prediction operation in step S12, i.e., the mode information pred_mode_flag may show intra prediction process, the process proceeds to step S20.
[0131]
　In step S20, the prediction unit 119 performs intra encoding process of encoding an image I to be coded using the predicted image P generated by the intra prediction process. Then, the image coding process is ended.
[0132]
　Figure 17 is a flowchart illustrating a first example of the inter prediction operation mode setting process of setting of the processing in step S11 in FIG. 16, the Merge flag and Affine flag. The inter prediction operation mode setting process is performed, for example, PU (CU) unit.
[0133]
　In step S41 of FIG. 17, the control unit 101 controls the respective blocks, with respect to the processing target PU (CU), the merge mode encoding process for each prediction information Pinfo non Merge flag and Affine flag as a candidate was carried out, RD cost J MRG is calculated. The calculation of the RD cost, resulting generated bit amount of the coded (code quantity) is performed based on such SSE decoded image (Error Sum of Squares).
[0134]
　In step S42, the control unit 101 controls the respective blocks, performs AMVP mode encoding process for each prediction information Pinfo non Merge flag and Affine flag that are candidates with respect to the processing target PU (CU), RD cost J AMVP to calculate.
[0135]
　In step S43, the control unit 101 controls the respective blocks, with respect to the processing target PU (CU), the merge affine transformation mode encoding process for each prediction information Pinfo non Merge flag and Affine flag as a candidate done, RD cost J MRGAFFINE to calculate.
[0136]
　In step S44, the control unit 101 controls the respective blocks, performs AMVP affine transformation mode encoding process for each prediction information Pinfo other than the candidate with respect to the processing target PU (CU) Merge flag and Affine flag , RD cost J AMVPAFFINE to calculate.
[0137]
　In step S45, the control unit 101, RD costs J MRG is, RD cost J MRG , J AMVP , J MRGAFFINE , and J AMVPAFFINE determines whether the smallest of.
[0138]
　RD cost J in step S45 MRG when it is determined to be the minimum in step S46, the control unit 101 sets the Merge flag of PU to be processed in 1, set the Affine flag to 0. Then, the inter-prediction processing mode setting process is completed.
[0139]
　RD cost J in step S45 MRG when it is determined not to be a minimum, the process proceeds to step S47. In step S47, the control unit 101, RD costs J AMVP is, RD cost J MRG , J AMVP , J MRGAFFINE , and J AMVPAFFINE determines whether the smallest of.
[0140]
　RD cost J in step S47 AMVP when it is determined to be the minimum, in step S48, the control unit 101, a Merge flag and Affine flag of PU to be processed is set to 0, ends the inter prediction operation mode setting process to.
[0141]
　Meanwhile, RD cost J in step S47 AMVP when it is determined not to be a minimum, the process proceeds to step S49. In step S49, the control unit 101, RD costs J MRGAFFINE is, RD cost J MRG , J AMVP , J MRGAFFINE , and J AMVPAFFINE determines whether the smallest of.

claims

Based on the large size two motion vectors of the vertices aligned in the direction of one of the vertical and horizontal size of the block, the prediction unit generating a prediction image of the block
　image processing apparatus comprising a.
[Requested item 2]
　The prediction unit, the prediction image of the adjacent block adjacent to the large apex direction side of the size of the vertical and horizontal size of the block, out of the vertical and horizontal size of the adjacent block when it produced based on the motion vectors of two vertices aligned in the direction of the large size, based on the large size two motion vectors of the vertices aligned in the direction of one of the vertical and horizontal size of the block, the generating a predictive image of the block
　configured to
　image processing apparatus according to claim 1.
[Requested item 3]
　Encoding unit for encoding a plurality vector prediction information indicating that generating a prediction image of the block based on large size two motion vectors of the vertices aligned in the direction of one of the vertical and horizontal size of the block
　further comprising an
　image processing apparatus according to claim 1.
[Requested item 4]
　The encoding unit, the prediction image of the adjacent block adjacent to the large apex direction side of the size of the vertical and horizontal size of the block, out of the vertical and horizontal size of the adjacent block large size based on whether they are generated based on two motion vectors of the vertices aligned in the direction of, the plurality vector prediction information coding
　constructed as
　an image processing apparatus according to claim 3.
[Requested item 5]
　The encoding unit, the predicted image of the neighboring blocks, based on whether they are generated based on the vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the adjacent block Te, it switches the context probability model in the encoding of the plurality vector prediction information
　configured as
　an image processing apparatus according to claim 4.
[Requested item 6]
　The encoding unit, the predicted image of the neighboring blocks, based on whether they are generated based on the vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the adjacent block Te, wherein the plurality vector switching the sign of the prediction information
　configured to
　image processing apparatus according to claim 4.
[Requested item 7]
　The encoding unit, when the prediction image of the neighboring block is generated based on the vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the adjacent block, said adjacent as predicted image block, the code amount is smaller than when it is not generated based on the vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the adjacent block, wherein a plurality vector prediction information encoding
　configured as
　an image processing apparatus according to claim 4.
[Requested item 8]
　The prediction unit, based on the vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the block, the reference picture of the block by affine transformation, prediction of the block It generates an image
　configured to
　image processing apparatus according to claim 1.
[Requested item 9]
　Said block is generated by repeating the division of one block horizontal and vertical to at least one of recursively
　constructed as
　an image processing apparatus according to claim 1.
[Requested item 10]
　The image processing apparatus,
　based on the vertical and horizontal large size two motion vectors of the vertices aligned in the direction of one of the size of the block, the step of generating a predictive image of the block
　image processing method comprising.