JP4519723B2 - Encoding or decoding apparatus for moving image data using motion vectors - Google Patents

Description

  The present invention relates to an apparatus for encoding or decoding moving image data using a motion vector, and more particularly to easily generate a motion vector for prediction by improving a memory for storing a motion vector and its memory control unit. The present invention relates to an encoding or decoding apparatus.

  MPEG2 and H261 are widely used as a method for encoding and decoding moving image data. According to this method of encoding / decoding moving image data, when processing image data of a frame in units of 16 × 16 macroblocks MB and encoding image data of the current frame, the image data of the current frame And a difference vector (prediction error data) between the predicted image data generated from the previous frame image data and the motion vector and the current frame image data, for example, are encoded and compressed. When decoding, the difference data is added to the predicted image data generated from the reproduced image data and the motion vector of the current frame to reproduce the image data of the current frame. The reproduced image data is further used for reproducing image data of subsequent frames. In this way, the amount of encoded data can be reduced by encoding and compressing only the motion vector and the difference data, rather than encoding and compressing the image data itself. In addition, the difference (prediction error) can be reduced by using the image at the position in consideration of the moving direction of the moving image as the reference image data, rather than using the difference data of the image data at the same position. Can be increased.

  Such encoding and decoding of moving image data are described in Patent Documents 1 and 2, for example.

On the other hand, when generating predicted image data based on motion vectors, the motion vectors of the macroblocks (candidate macroblocks) around the reference destination macroblock are used instead of the motion vectors of the current target macroblock. It has been proposed to calculate a median value of motion vectors as a compensated motion vector and generate predicted image data based on the compensated motion vector. This is called a median prediction calculation. By adding information on motion vectors of candidate macroblocks around the reference macroblock, a more accurate motion vector can be obtained and the compensated motion vector is used. By doing so, the difference data (prediction error data) is made smaller, and the coding rate is further improved.
JP 2000-102016 A JP 2004-48552 A

  However, MPEG4 and H.264, which are next-generation video encoding / decoding systems. In the H.263 method, the macroblock is provided with two motion vectors, and the followability to more intense or fine motion is enhanced for the moving image vector, thereby reducing the prediction error. In addition, H. In the H.264 system, a macroblock can have up to 16 motion vectors, and there are seven types of block types for motion compensation (hereinafter referred to as MC block types) depending on how the motion vectors are provided. .

  Since the motion compensation block type (MC block type) varies depending on the macroblock as described above, there is a problem that the address management of the motion vector of the candidate block at the time of median prediction calculation is complicated and the processing efficiency is lowered.

  Accordingly, an object of the present invention is to provide a moving image data encoding / decoding device that facilitates access to motion vectors of candidate blocks at the time of median prediction calculation and improves processing efficiency.

  In order to achieve the above object, according to the first aspect of the present invention, for a macroblock consisting of a predetermined pixel unit, a motion vector corresponding to the motion of a moving image in the macroblock, and the motion vector Moving image data for generating predicted image data based on the motion vector from encoded data having prediction error data that is a difference from predicted image data to be predicted, and adding the prediction error data to reproduce the image data Depends on the motion vector memory that stores the motion vector of the macroblock corresponding to the macroblock and the motion compensation block type corresponding to the type of motion vector assigned to the macroblock Without moving the motion vector of the macroblock in units of the minimum motion compensation block. Stored in Kutorumemori, and a memory controller for reading the motion vector of the candidate block in the median prediction calculation from the motion vector memory based on the address corresponding to the motion compensation block type of the current macroblock.

  In the first aspect described above, in a preferred embodiment, the motion compensation block type includes a first type having a single motion vector for the macroblock and a plurality of motion compensation blocks obtained by subdividing the macroblock. At least a second type having a plurality of motion vectors, and the minimum motion compensation block is a block that can be subdivided into a minimum in the second type.

  Here, the candidate blocks are macroblocks located around the reference macroblock specified based on the motion vector of the current macroblock, for example, at the upper, upper right, and left. According to the recommended encoding / decoding method, the prediction error, which is a difference from the predicted image data, can be reduced by obtaining the motion vector for prediction in consideration of the motion vector of the candidate block.

  According to the present invention, since the motion vector is stored in units of the minimum motion compensation block without depending on the motion compensation block type of the macroblock, the reading of the motion vector of the candidate block at the time of median prediction calculation can be simplified, and the processing efficiency Can be increased.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the technical scope of the present invention is not limited to these embodiments, but extends to the matters described in the claims and equivalents thereof.

  FIG. 1 is a configuration diagram of a moving image data encoding circuit and a decoding circuit according to the present embodiment. In the encoding circuit of FIG. 1A, image data IMG for each frame constituting a moving image is supplied, and encoded data CD is output. The motion vector generation unit 28 generates a motion vector MV for each macroblock composed of 16 × 16 pixels with respect to the supplied image data IMG. The frame memory 24 stores image data of the previous frame, and the predicted image creation unit 26 generates predicted image data 27 from the motion vector MV of the current frame and, for example, image data of the previous frame. Then, the subtracter 10 generates difference data between the image data IMG and the predicted image data 27 as the prediction error data 11. The prediction error data 11 is subjected to discrete cosine transform by the DCT unit 12, quantized by the quantizer 14, and the quantized data is variable-length encoded by the variable-length encoder 16 and encoded and compressed. The encoded data CD is supplied to the decoding circuit. On the other hand, the quantized data is inversely quantized by the inverse quantizer 18, decoded by inverse discrete cosine transform by the inverse DCT unit 20, and the decoded prediction error data is predicted by the adder 22. The original image data is added to the data 27 and stored in the frame memory 24 as local decoded image data.

  On the other hand, in the encoding circuit of FIG. 1B, the encoded data CD is decoded by the variable length decoder 30. The decoded prediction error data 31 is subjected to inverse quantization and inverse discrete cosine transform by an inverse quantizer 32 and an inverse DCT unit 34, respectively. The decoded motion vector MV is supplied to the prediction motion vector generator 42. The motion vector generator 42 for prediction uses, for example, a motion vector memory 44 that stores a motion vector for each macroblock or motion compensation block (MC block) of the previous frame, and the surroundings of the reference destination macroblock corresponding to the motion vector MV. The block motion vectors MV1, 2, and 3 are read out to obtain a prediction motion vector PMV. The predicted image creation unit 38 generates predicted image data 39 based on the motion vector for prediction PMV and, for example, image data of the previous frame. Then, the adder 36 adds the decoded prediction error data 35 and the predicted image data 39 and outputs decoded image data IMG. The decoded image data IMG is stored in the frame memory 40 and is used for generating a predicted image of an image of the next frame.

  FIG. 2 is a diagram for explaining a motion vector, a predicted image, and a prediction error. As shown in FIG. 2A, when the image at time a moves to the right at the subsequent time b, the movement information is assigned to the macroblock at time b as a motion vector mv. The motion vector mv is generated corresponding to the position of the image closest to the macro block image at time b in the image at time a. Therefore, some difference (prediction error) occurs between the predicted image reproduced with the motion vector mv and the image at time b. Therefore, the moving image encoding circuit encodes the motion vector and the prediction error data.

  As shown in FIG. 2B, in the decoding circuit, in order to reproduce the image 50 at time b, the position corresponding to the motion vector mv from the position 52 corresponding to the image 50 in the image at time a. 54 images are set as predicted images, and a prediction error is added to the predicted images to reproduce the image 50.

  FIG. 3 is a diagram illustrating the relationship between macroblocks and motion vectors. As shown in FIG. 3A, the frame image 56 is divided into 16 × 16 pixel macroblocks MB, and motion vector information indicated by arrows is given to the macroblocks MB. In addition, as shown in FIG. 3B, a plurality of pieces of motion vector information may be given to the macroblock MB. As described above, by dividing one macro block MB into a plurality of blocks and giving motion vector information to each of the divided blocks, encoding can be performed following a rapidly changing moving image. In the example of FIG. 3B, some macro blocks MB are divided into upper and lower blocks, and motion vectors are given to the respective blocks. A single motion vector is given to the other macroblocks MB.

  FIG. 4 is a diagram for explaining the motion vector for prediction. FIG. 5 is a flowchart of the decoding process using the motion vector for prediction. The decoding process will be described with reference to both figures. FIG. 4 shows the previous frame FMn-1 and the current frame FMn. Here, for simplification, it is assumed that predicted image data is generated from image data of the previous frame.

  Now, when the image of the macro block MB of the current frame is reproduced, the position MMC of the reference macro block is obtained based on the motion vector MV of the macro block MB, and the reference macro of the previous frame image at the position corresponding to the position MMC is obtained. Block RMB is detected (S10). Then, the motion vectors MV0, MV2 and MV3 of the upper, upper right and left candidate blocks around the reference macroblock RMB of the previous frame are read from the motion vector memory (S12). Further, the median value of the three read motion vectors MV1, 2, 3 is obtained as the prediction motion vector PMV (S14). Then, the image data of the macro block of the previous frame FMn−1 located at the position based on the motion vector for prediction PMV is used as the predicted image data, and difference data (prediction error data) is added to the image data to add the macro block MB of the current frame FMn. Image data is reproduced (S16).

  FIG. 6 is a diagram for explaining the motion compensation block type. As described with reference to FIG. 3B, in the moving image encoding / decoding method, a plurality of motion vectors may be given to one macroblock. In FIG. The motion compensation block type (MC block type) employed in the H.264 system is shown. The image signal is stored in an image signal memory (frame memory) in units of 16 × 16 pixel macroblocks 60. On the other hand, a motion vector MV is given to each macroblock, and the motion vector MV is assigned as a single or a plurality according to the motion compensation block type (MC block type).

  The MC block type shown in FIG. 6 is MC block type 61 in which one motion vector MV is assigned to 16 × 16 pixels, MC block type 62 in which motion vectors MV are assigned to upper and lower 16 × 8 pixel blocks, and left and right 8 × 16 pixel blocks, respectively. MC block type 63 to which motion vector MV is assigned, MC block type 64 to which motion vector MV is assigned to each of four 8 × 8 pixel blocks, MC block type 65 to which motion vector MV is assigned to each of eight 8 × 4 pixel blocks, 7 types of MC block type 66 in which motion vector MV is assigned to each of 8 4 × 8 pixel blocks and MC block type 67 in which motion vector MV is assigned to 16 4 × 4 pixel blocks. A. In the MC block type 67, a motion vector is assigned to each minimum motion compensation block (4 × 4).

  Therefore, the number of pieces of motion vector MV information stored in the motion vector memory varies depending on the MC block type of each macro block. Considering this in the memory address space, this means that the number of motion vector information storage addresses of each macroblock differs from 1, 2, 4, 8, and 16 according to the MC block type. Therefore, when reading the motion vector of the candidate block from the motion vector memory, it is necessary to read the motion vector of the corresponding address in consideration of the MC block type.

  FIG. 7 is a diagram illustrating problems between different MC block types. In this example, the MC block type of the current macroblock 70 has motion vectors mv0 to 3 for each 8 × 8 block, and one candidate macroblock 72 to be read from the current macroblock 70 is one in a 16 × 16 block. Assume that a motion vector mv0 is included. In this case, the motion vector memory 71 of the current macroblock 70 stores four motion vectors mv0-3 in addition to the MC block type information, and the motion vector memory 73 of the macroblock 72 stores the MC block type information. One motion vector mv0 is stored.

  The motion vectors of candidate macroblocks 72 to be read for the three blocks of motion vectors mv0-3 of the current macroblock 70 are all the same motion vector information mv0. However, if the MC block type of the candidate macroblock 72 is different, the motion vector of the candidate macroblock to be read is different. That is, it is necessary to calculate which motion vector should be read out in accordance with each MC block type, and complicated processing is required.

  FIG. 8 is a diagram illustrating reference to candidate macroblocks in median prediction according to the present embodiment. As described with reference to FIG. 7, it will be described that the motion vector to be referred to by the candidate macroblock differs depending on the combination of the MC block type of the current macroblock and the MC block type of the candidate macroblock. In FIG. 8, MB is the current macro block, and MB 1, 2, and 3 are the upper, upper right, and left candidate macro blocks.

  FIGS. 8A to 8D show MC block types in which a motion vector is assigned to the current macroblock MB corresponding to four 8 × 8 blocks, and candidate macroblock MB1 to be referred to corresponding to four 8 × 8 blocks, respectively. -3 blocks are indicated by thick frames. In this case, the upper candidate block MB1 is an MC block type of 16 × 16 blocks, the upper right candidate block MB2 is two MC block types of 8 × 16 blocks, and the left candidate block MB3 is an MC block type of two 16 × 8 blocks.

  In FIG. 8 (a), as indicated by the thick frame, for the block 3x0 of the current macroblock MB, the motion vector of one block in the macroblock MB1, the motion vector of the left 8x16 block in MB2, and the block in MB3 The motion vectors of the upper 16 × 8 block are read out. Similarly, in FIG. 8B, with respect to the block 3x1 of the current macroblock MB, the motion vector of one block in the macroblock MB1, the motion vector of the right 8x16 block in MB2, and the upper 16x8 block in MB3 Are respectively read out. In FIG. 8C, with respect to the block 3x2 of the current macroblock MB, the motion vector of one block in the macroblock MB1, the motion vector of the left 8x16 block in MB2, and the motion of the lower 16x8 block in MB3 Each vector is read out. In FIG. 8D, with respect to the block 3x3 of the current macroblock MB, the motion vector of one block in the macroblock MB1, the motion vector of the right 8x16 block in MB2, and the lower 16x8 block in MB3 Are respectively read out.

  FIG. 8E shows an MC block type in which one motion vector is assigned to the current macroblock MB corresponding to 16 × 16 blocks. The candidate block MB1 has four 8 × 8 blocks, and the candidate block MB2 has upper and lower 16 × 8 MC blocks. The candidate block MB3 is an MC block type of 8 × 16 blocks on the left and right. In this case, the motion vector of the lower left block 0x2 in the macro block MB1, the motion vector of the lower block 1x1 in MB2, and the motion vector of the right block 2x1 in MB3 are compared to the 16x16 block of the current macro block MB, respectively. Read out.

  FIG. 8 (f) shows an MC block type in which two motion vectors are assigned to the current macroblock MB corresponding to the upper and lower 16 × 8 blocks, the candidate block MB1 is four 8 × 8 blocks, and the candidate block MB2 is the upper and lower 16 × 8 MC. Candidate block MB3 is an MC block type of 8 × 16 blocks on the left and right. In this case, with respect to the upper block 3x0 of the current macroblock MB, the motion vector of the lower left block 0x2 in the macroblock MB1, the motion vector of the lower block 1x1 in MB2, and the motion vector of the right block 2x1 in MB3. Each is read out.

  As described above, the position of the motion vector to be read differs depending on the combination of the MC block type of the current macroblock and the MC block type of the candidate macroblock. Then, the median value of the read motion vectors is calculated and used as a compensated motion vector for generating predicted image data.

  FIG. 9 is a diagram illustrating a configuration example of the motion vector memory. This indicates that the position of the motion vector to be read is different. FIG. 9 shows examples of FIGS. 8B and 8C. In the motion vector memory, the upper candidate macroblock MB1 has a motion vector of block 0x0, the upper right candidate macroblock MB2 has a motion vector of blocks 1x0 and 1x1, and the left candidate macroblock MB3 has a block 2x0, 2 × 1 motion vectors are stored, and the current macroblock MB stores motion vectors 3 × 0 to 3 × 3. Then, for the current macroblock block 3x1 as shown in FIG. 8 (b), it is necessary to read the motion vector of the previous block indicated by the solid line, whereas as shown in FIG. 8 (c). For the block 3x2 of the macro block, it is necessary to read the motion vector of the previous block indicated by the broken line. As described above, according to the MC block type of both macroblocks, the storage position (address in memory) of the motion vector to be read differs as shown by the solid line and the broken line. Therefore, it is necessary to obtain the address of the motion vector to be read based on the MC block type information of the current macroblock and the candidate macroblock, which increases the number of processing steps.

  FIG. 10 is a diagram showing a motion vector writing method to the motion vector memory in the present embodiment. In the present embodiment, 16 types of motion vectors are stored in the motion vector memory for each macro block without depending on the MC block type. That is, the motion vector is written in the motion vector memory in units of the minimum motion compensation block.

  FIG. 10A shows a case where the MC block type is a type in which one motion vector is given to a 16 × 16 block, and the same motion vector is redundantly stored in the motion vector memory corresponding to 16 blocks 0 to 15. . FIG. 10B shows a case in which the motion vector is given to the 16 × 8 blocks of the upper and lower two MC block types, but the same two motion vectors corresponding to the 16 blocks 0 to 15 are stored in the motion vector memory. Store duplicates. That is, in the motion vector memory, the first motion vector is redundantly stored corresponding to the blocks 0 to 7, and the second motion vector is redundantly stored corresponding to the blocks 8-15.

  FIG. 10 (c) shows a case where the motion vector is assigned to four 8 × 8 blocks each having an MC block type. In the motion vector memory, four motion vectors corresponding to 16 blocks 0 to 15 are overlapped. Store. That is, the motion vector memory stores the first motion vector redundantly corresponding to blocks 0, 1, 4 and 5, and the second motion vector corresponding to blocks 2, 3, 6 and 7 The third motion vector is stored correspondingly to the blocks 8, 9, 12, and 13, and the fourth motion vector is stored corresponding to the blocks 10, 11, 14, and 15. Store.

  Further, FIG. 10 (d) shows a case where a motion vector is given to each of 16 4 × 4 blocks whose MC block type is a minimum motion compensation block, and these 16 motion vectors are stored in the motion vector memory in the minimum. Stored corresponding to motion compensation blocks 0-15.

  FIG. 11 is a diagram showing a motion vector writing method to the motion vector memory in the present embodiment. 11A, 11B, and 11C correspond to FIGS. 10A, 10B, and 10C. As shown in FIG. 11A, when the MC block type is 16 × 16 blocks, one motion vector mv0 is stored in the motion vector memory in duplicate at addresses 0x0 to 0xF (hexadecimal). As shown in FIG. 11B, when the MC block type is two 16x8 blocks, the motion vector memory has the motion vector mv0 overlapping at addresses 0x0 to 0x7 and the motion vector mv1 overlapping at addresses 0x8 to 0xF. Stored. As shown in FIG. 11C, when the MC block type is four 8x8 blocks, the motion vector mv0 is at addresses 0x0 to 0x3, the motion vector mv1 is at addresses 0x4 to 0x7, and the addresses 0x8 to The motion vector mv2 is stored redundantly at 0xB, and the motion vector mv3 is stored redundantly at addresses 0xC to 0xF.

  As described above, regardless of the MC block type, motion vectors are stored in 16 storage address areas of the motion vector memory for each macro block. As described above, by storing the motion vector in the memory in units of the minimum motion compensation block, in the motion vector reading operation of the candidate macroblock, which motion vector should be read only depending on the MC block type of the current macroblock. You can decide if it ’s good.

  FIG. 12 is a diagram illustrating a motion vector memory reading method according to the present embodiment. In FIG. 12, the readout positions in the cases of FIGS. 8B and 8C are indicated by solid lines and broken lines. According to the present embodiment, a motion vector is written at 16 addresses in the motion vector memory for each macroblock. As shown in FIG. 8, in the case of FIG. 8 (b), the block of the motion vector to be read corresponding to the block 3x1 of the current macroblock MB is as shown by the hatched rectangle. In the upper block MB1, the address 0xC, the upper right block The motion vector at address 1xE is read in MB2 and the address 2x4 is read in left block MB3. In the case of FIG. 8C, the motion vector of the address 0xE in the upper block MB1, the address 1xC in the upper right block MB2, and the address 2xB in the left block MB3 is read out corresponding to the block 3x2 of the current macroblock MB. The same applies to FIGS. 8A and 8D.

  Further, as shown in FIGS. 8E and 8F, when the MC block type of the current macro block MB is different, the address to be read is also different correspondingly. However, the read address can be determined only depending on the type of the current macroblock without depending on the MC block type of the macroblocks MB1, 2, and 3 to be read. This is because motion vectors are stored at 16 address positions for each macroblock.

  As described above, in the present embodiment, when the MC block type of the current macroblock is found in the motion vector read operation in the candidate macroblock in the median prediction process, the MC block of the candidate macroblock to be read is based on the MC block type. The read address in the motion vector memory can be determined without depending on the type.

  Corresponding to the MC block type of the current macroblock, the position of the block in the candidate macroblock to be read in each block is MPEG4 or H.264. It is specified by recommendations such as H.264. Therefore, the motion vector of the designated block can be easily read out by reading out, for example, the motion vector at the head address at the position of the block designated by the recommendation. Since the motion vector is stored at all 16 address positions without depending on the MC block type of the macro block to be read, the specification of the read address is unambiguous according to the MC block type of the current macro block. There is no need for special arithmetic processing when generating.

  FIG. 13 is a configuration diagram of the memory control unit in the present embodiment. The memory control unit 45 is a memory interface for the motion vector memory 44 (see FIG. 1B). The memory control unit 45 controls writing of a motion vector MV to be stored in the memory 44 to a predetermined address in the memory and reads it from the memory 44. The power vector is read from the corresponding address in the memory and controlled. The read data MV1, 2, 3 are supplied to the prediction motion vector generation unit 42, and a prediction motion vector PMV is generated by an arithmetic process such as obtaining the median value.

  The memory control unit 45 includes circuits 451, 452, and 453 for writing a motion vector MV to be stored in the macroblock in the memory 44, and a circuit for reading out the motion vectors MV1, 2, and 3 to be read out for the current macroblock. 454, 455.

  As shown in FIG. 1B, the motion vector MV decoded by the variable length decoder 30 is once written in the motion vector memory 44. At this time, when the address of the macro block MB being processed is given to the write address control unit 451, it sequentially generates 16 write addresses WAAdd. Further, the MC block type of the macro block is supplied to the MV block type identifying unit 452, and in the example of FIG. 6, it is identified that it is one of seven types, and timing adjustment and data are determined according to the identification result. The extension processing unit 453 supplies the stored motion vector MV to the motion vector memory 44 as write data WData in synchronization with the timing of the write address WAAdd.

  FIG. 14 is a timing chart of a write operation and a read operation of the memory control unit in the present embodiment. 14A and 14B show the write operation. First, when the MC block type of the macro block MB being processed is 16 × 16, there is only one motion vector. Therefore, in synchronization with the timing at which the write address control unit 451 supplies 16 addresses to the memory 44, the timing adjustment and data expansion processing unit 453 gives one motion vector to the memory 44 as write data Wdata. Thereby, one motion vector (16 × 16 (0)) is written to 16 addresses of the memory 44. As a result, it becomes as shown in FIG.

  Next, in the case of FIG. 11C, in synchronization with the timing at which the write address control unit 451 supplies 16 addresses to the memory 44, the timing adjustment and data expansion processing unit 453 has four types of motion vectors. (8 × 8 (0) to (3)) is given to the memory 44 as the write data Wdata. In FIG. 14, the data latch selection signal is a control signal in the timing adjustment and data extension processing unit 453. In response to the MC block type (4 × 4), the timing adjustment and data extension processing unit 453 receives the data latch selection signal. In response to 0-3, the stored motion vector MV is latched in a corresponding internal register (not shown), and the latched motion vector is supplied to the memory 44 as write data Wdata.

  Further, when the MC block type is 4 × 4, 16 kinds of stored motion vectors 4 × 4 (0) to (15) are given to the memory 44 as write data WData corresponding to 16 write addresses WAAdd. In response to the internal data latch signal, 16 types of motion vectors are latched in the built-in latch circuit.

  Next, the reading operation will be described. The reading operation of the motion vector memory is as described with reference to FIG. The MC block type identification unit 455 is supplied with the MC block type of the current macroblock and identifies it, and the read address control unit 454 determines the MC block type identified as the address of the reference macroblock corresponding to the current macroblock. In response, three read addresses are generated. The motion vectors MV1, 2, 3 read from the memory 44 corresponding to the read address are output to the motion vector generator 42 for prediction.

  FIG. 14C shows the read operation when the MC block type is 16 × 16 blocks (example in FIG. 8E) and 8 × 8 blocks (examples in FIGS. 8A to 8D). In the case of a 16 × 16 block, the block positions of three motion vectors to be read are specified corresponding to the MC block type of the current macroblock. Therefore, the read address control unit 454 supplies the addresses A1, A2, and A3 as the read address RAdd to the memory 44, and correspondingly, the motion vectors mv1, 2, and 3 are output as the read data RData. The prediction motion vector generation unit 42 generates a prediction motion vector PMV from them. Similarly, in the case of the 8 × 8 block, the block positions of the three motion vectors to be read are specified corresponding to the MC block type of the current macro block. Therefore, the read address control unit 454 generates addresses A11, A12, and A13, and the motion vectors mv11, 12, and 13 are read correspondingly to generate the prediction motion vector PMV.

  As shown in FIG. 12, the read address control unit 454 can specify which address's motion vector should be read according to the MC block type of the current macroblock. The read address can be generated uniformly without depending on it.

  In the above embodiment, the memory control for the motion vector memory 44 of the decoding circuit of FIG. 1 has been described. However, the encoding circuit of FIG. 1 can also be applied to a prediction motion vector generation process in the motion vector generation unit 28 when generating local decoded image data.

  The above embodiment is summarized as follows.

(Supplementary note 1) For a macroblock consisting of a predetermined pixel unit, a prediction error data which is a difference between a motion vector corresponding to a motion of a moving image in the macroblock and predicted image data predicted by the motion vector In the decoding or encoding device of moving image data, which generates predicted image data based on the motion vector from the given encoded data and adds the prediction error data to reproduce the image data,
A motion vector memory for storing a motion vector of the macroblock corresponding to the macroblock;
The motion vector of the macroblock is stored in the motion vector memory in units of the minimum motion compensation block without depending on the motion compensation block type indicating the type of motion vector assigned to the macroblock, and the candidate block in the median prediction calculation And a memory control unit that reads out the motion vector from the motion vector memory based on an address corresponding to the motion compensation block type of the current macroblock.

(Appendix 2) In Appendix 1,
The motion compensation block type includes a first type having a single motion vector for the macroblock and a second type having a plurality of motion vectors corresponding to a plurality of motion compensation blocks obtained by subdividing the macroblock. And at least
The apparatus for encoding or decoding moving image data, wherein the minimum motion compensation block is a block that can be subdivided into a minimum in the second type.

(Appendix 3) In Appendix 1,
The memory control unit repeatedly stores the motion vector at a plurality of addresses corresponding to the minimum motion compensation block of the motion vector memory when a motion vector is assigned to a block larger than the minimum motion compensation block. An apparatus for encoding or decoding moving image data.

(Appendix 4) In Appendix 1,
The moving image, wherein the memory control unit reads the motion vector based on an address corresponding to a position of a motion compensation block of the current macroblock when reading the motion vector of the candidate block from the motion vector memory. Data encoding / decoding device.

(Appendix 5) In Appendix 1,
Furthermore, the apparatus for encoding or decoding moving image data, further comprising a prediction motion vector generation unit that calculates a prediction motion vector from the motion vectors of the plurality of read candidate blocks.

It is a block diagram of the encoding circuit and decoding circuit of moving image data in the present embodiment. It is a figure explaining a motion vector, a prediction image, and a prediction error. It is a figure which shows the relationship between a macroblock and a motion vector. It is a figure explaining the motion vector for prediction. It is a flowchart figure of the decoding process using the motion vector for prediction. It is a figure explaining a motion compensation block type. It is a figure which shows the problem between different MC block types. It is a figure which shows the reference to the candidate macroblock in the median prediction in this Embodiment. It is a figure which shows the structural example of a motion vector memory. It is a figure which shows the motion vector writing method to the motion vector memory in this Embodiment. It is a figure which shows the motion vector writing method to the motion vector memory in this Embodiment. It is a figure which shows the reading method of the motion vector memory in this Embodiment. It is a block diagram of the memory control part in this Embodiment. FIG. 6 is a timing chart of a write operation and a read operation of the memory control unit in the present embodiment.

Explanation of symbols

42: Prediction motion vector generation unit 44: Motion vector memory 45: Memory control unit 451: Write address control unit 454: Read address control unit MV: Stored motion vector MV1, 2, 3: Read motion vector PMV: Motion vector for prediction

Claims (1)

Coding in which a macroblock composed of a predetermined pixel unit is provided with a motion vector corresponding to a motion of a moving image in the macroblock and prediction error data which is a difference between predicted image data predicted by the motion vector In the decoding or encoding device for moving image data, which generates predicted image data based on the motion vector from the data and adds the prediction error data to reproduce the image data,
A motion vector memory for storing a motion vector of the macroblock corresponding to the macroblock;
Assigning a plurality of motion vectors to the macro block and not depending on a compensation block type indicating the block size of a plurality of types of motion compensation blocks having different block sizes, the motion vector of the macro block in units of the minimum motion compensation block. Coding of moving image data having a memory control unit stored in the motion vector memory and reading out the motion vector of the candidate block in the median prediction calculation from the motion vector memory based on the address corresponding to the motion compensation block type of the current macroblock Or in the decryption device,
The memory control unit repeatedly stores the motion vector at a plurality of addresses corresponding to the minimum motion compensation block of the motion vector memory when a motion vector is assigned to the macro block that is larger than the minimum motion compensation block. An apparatus for encoding or decoding moving image data.

Images