JPWO2011034148A1 - Encoding device, decoding device, moving image encoding device, moving image decoding device, and encoded data - Google Patents

Abstract

The encoding apparatus uses an information indicating which PMV is selected from a plurality of PMV candidates, and extracts an MV candidate in the MV existence area from the temporary MV candidates, and an MV An MV index encoding unit (1153) that encodes each MV index when there are two or more MV indexes specifying the MV candidates extracted by the flag determination unit (1115).

Description

  The present invention mainly relates to a moving image encoding device and a moving image decoding device that efficiently encode a motion vector.

  In the field of moving image encoding technology, generally, image data is divided into a plurality of blocks, and encoding is performed in units of divided blocks. In block unit coding, prediction signal generation of an input signal, calculation of a residual signal that is a difference between the input signal and the prediction signal, conversion of the calculated residual signal, quantization of a transform coefficient obtained by the conversion, An encoding process such as variable length encoding of the quantized transform coefficient is performed.

  As a method for generating a prediction signal, there are intra prediction using a reproduced image around the target block and inter prediction using a reproduced image of an already encoded / decoded frame.

  Intra prediction includes a plurality of intra prediction modes such as DC prediction, horizontal prediction, and vertical prediction, and information indicating which intra prediction mode is used is encoded.

  In inter prediction, a frame ID for specifying a reference frame used for prediction and a motion vector (MV) are encoded.

  MV encoding is performed by first calculating a predicted motion vector (PMV) that is a predicted value of MV, and then encoding the difference between PMV and MV. FIG. 16 is a diagram illustrating a configuration of MV encoded data of Non-Patent Document 1. As shown in FIG. 16, the MV is encoded by being divided into a PMV index indicating PMV, an absolute value of the difference between PMV and MV (MVD absolute value), and an MVD code indicating whether the difference is positive or negative. The The MVD absolute value and the MVD code are encoded in the order of the X component and the Y component, and when the MVD absolute value for each component is 0, the MVD code is not encoded.

  Here, the MVD absolute value is binarized and efficiently encoded by an arithmetic code using an adaptive probability. The MVD code is also binarized and encoded by an arithmetic code. However, in Non-Patent Document 1, the probability of occurrence of 0 and 1 of the binarized code is always 0.5, and as a result of encoding, The code amount is the same as in the case of fixed length coding.

  Non-Patent Document 2 discloses a method of encoding an MV as it is without encoding the MVD absolute value and the MVD code separately. In this method, the existence probability of MV is estimated by a model using the sum of distribution functions using a generalized Gaussian function by calculating the distance between PMV and MV for each of a plurality of PMVs. Then, encoding is performed using a multi-valued arithmetic code using the probability calculated from the estimated model.

  Non-Patent Document 1 discloses a technique called Motion Vector Completion (MVCOMP). In MVCOMP, encoding efficiency is improved by using a plurality of motion vector predictor candidates (PMV candidates). Specifically, a PMV close to the MV to be encoded is selected from among the PMV candidates, and an index (PMV index) that identifies the PMV candidate is encoded. Next, the difference between PMV and MV is encoded. In MVCOMP, by selecting a suitable PMV from a plurality of PMV candidates, the difference between MV and PMV can be made smaller than when one PMV is used, so that the code amount of the difference can be reduced.

Competition-Based Schema for Motion Vector Selection and Coding, ITU-T Q. 6 / SG16 VCEG, VCEG-AC06, 2006 Takeuchi et al., Arithmetic Coding of Motion Vectors Using Multimodal Probabilistic Models, IPSJ Annual Meeting, 5-3, 2009

  Non-Patent Document 1 that calculates a plurality of motion vector predictor candidates (PMV candidates), selects one motion vector predictor (PMV) from the PMV candidates, and encodes the difference between the motion vector (MV) and the PMV. In MV encoding, since the information of the selected PMV is not used for encoding the difference between MV and PMV, the code amount is large.

  More specifically, the efficiency is reduced because the relationship between the selected PMV and the PMV candidate does not use the property that the region where the encoding target MV can exist is limited.

  Furthermore, in the MV coding of Non-Patent Document 1, the MVD absolute value that is the absolute value of the difference between PMV and MV is efficiently coded by arithmetic coding using an adaptive probability. As for the encoding of portions other than the MVD absolute value, etc., since a fixed probability is used, efficient encoding has not been performed.

  On the other hand, in the MV encoding of Non-Patent Document 2, since the MVD absolute value is not used, encoding can be performed according to the existence probability of MV, but a parameter (for example, generalization) that determines characteristics of the distribution function in advance. Since it is necessary to calculate the standard deviation and shape parameters of the Gaussian function, it is difficult to apply the MV encoding of Non-Patent Document 2 to an arbitrary image. Further, since it is necessary to repeatedly calculate the value of the cumulative distribution function, there is a problem that the amount of calculation is large even if the table reference described in Non-Patent Document 2 is used. Moreover, although the magnitude of the MVD absolute value tends to have a high spatial correlation, it is difficult to use such a tendency because the MVD absolute value is not separated.

In order to solve the above problems, the present invention has the following features. That is,
The encoding apparatus according to the present invention generates a prediction data from already encoded data and encodes a difference absolute value between the predetermined data and the prediction data when encoding the predetermined data. When the number of data candidates calculated by the encoding unit, the data candidate calculation unit calculating the data candidate from the difference absolute value and the prediction data, and the data candidate calculation unit is 2 or more, for each data candidate And a data candidate specifying flag encoding unit that encodes the flag set for each data candidate.

  The encoding apparatus of the present invention generates a plurality of prediction data candidates from already encoded data when encoding predetermined data, and selects prediction data to be used for encoding from the prediction data candidates A determination unit; a difference absolute value encoding unit that encodes a difference absolute value between the predetermined data and the prediction data; a data candidate calculation unit that calculates a data candidate from the difference absolute value and the prediction data; A data candidate determination unit that determines the validity of the data candidate calculated from the absolute value of the difference and the prediction data from the prediction data and the prediction data candidate, and the data candidate that has been validated by the data candidate determination unit A data candidate specifying flag encoding unit is provided that encodes a flag that specifies the data candidate when the number is two or more.

  The encoding apparatus of the present invention generates a plurality of prediction data candidates from already encoded data when encoding predetermined data, and selects prediction data to be used for encoding from the prediction data candidates A determination unit, and a difference absolute value encoding unit that encodes the difference absolute value by using the difference absolute value between the predetermined data and the prediction data as a symbol string and arithmetically encoding the symbol string; The difference absolute value encoding unit changes the appearance probability of each symbol used for arithmetic encoding of the symbol string according to the prediction data selected by the prediction data determination unit.

  The decoding apparatus of the present invention, when decoding predetermined data, generates a difference absolute value decoding unit that decodes an encoded difference absolute value, prediction data from already encoded data, and the difference absolute value When the number of data candidates generated by the data candidate calculation unit and the data candidate calculation unit that calculates data candidates from the value and the prediction data is two or more, each data candidate differs for each data candidate A data candidate specifying flag decoding unit that decodes a flag for which a value is set.

  When decoding predetermined data, the decoding device of the present invention decodes an encoded difference absolute value, generates a plurality of prediction data candidates from already encoded data, and generates a code from among the prediction data candidates. A prediction data determination unit that selects prediction data to be used for conversion, a data candidate determination unit that determines the validity of the data candidate calculated from the absolute value of the difference and the prediction data from the prediction data and the prediction data candidate, And a data candidate specifying flag decoding unit for decoding a flag for specifying the data candidate when the number of data candidates validated by the data candidate determining unit is 2 or more.

  A decoding apparatus according to the present invention, when decoding predetermined data, generates a plurality of prediction data candidates from already encoded data, and selects a prediction data to be used for decoding from the prediction data candidates; A difference absolute value decoding unit that decodes the difference absolute value by arithmetically decoding a symbol string that is a difference absolute value between the predetermined data and the prediction data, and the difference absolute value decoding unit includes: The appearance probability of each symbol used for decoding the symbol string is changed according to the prediction data selected by the prediction data determination unit.

  By using information on which prediction motion vector is selected from a plurality of motion vector predictor candidates, and limiting the motion vector existence possible area, by encoding or decoding the motion vector using the limitation, It is possible to reduce the code amount of the motion vector in the encoded data. By reducing the code amount of the motion vector, the coding efficiency in moving picture coding is improved.

1 is a block diagram of a video encoding apparatus 100 according to a first embodiment of the present invention. It is a block diagram of the moving image decoding apparatus 200 of the 2nd Embodiment of this invention. It is a block diagram of the motion vector encoding part 111 of the 1st Embodiment of this invention. It is a block diagram of the motion vector decoding part 211 of the 2nd Embodiment of this invention. It is a block diagram of the MV flag determination part 1115 by the 1st Embodiment of this invention. It is a block diagram of the MV selection part 2115 by the 2nd Embodiment of this invention. It is a figure which shows the operation | movement flow of the MV candidate calculation part 5001 by the 1st, 2nd embodiment of this invention. It is a figure which shows the operation | movement flow of the MV candidate determination part 5002 by the 1st, 2nd embodiment of this invention. It is a figure which shows the operation | movement flow of the MV index determination part 5003 by the 1st Embodiment of this invention. It is a figure explaining the prediction motion vector of this invention. It is a figure explaining the example of the possible area | region of MV of this invention. It is a figure explaining another example of the possible area | region of MV of this invention. It is a figure explaining another example of the possible area | region of MV of this invention. It is a figure explaining the relationship between the MV existence possible area | region of this invention, and an MV candidate. It is a figure which shows the structure of the encoding data of MV of this invention. It is a figure which shows the structure of the encoding data of the conventional MV. It is a block diagram of the moving image encoder 100 'of the 3rd Embodiment of this invention. It is a block diagram of the motion vector encoding part 111 'of the 3rd Embodiment of this invention. It is a block diagram of the principal part which comprises the moving image encoder of the 3rd Embodiment of this invention. (A) shows the MV flag determination unit 1115 ′ according to the third embodiment of the present invention, and (b) shows the MV candidate sorting unit 5004. It is a block diagram of the moving image decoding apparatus 200 'of the 4th Embodiment of this invention. It is a block diagram of the motion vector decoding part 211 'of the 4th Embodiment of this invention. It is a block diagram of MV selection part 2115 'by the 4th Embodiment of this invention. It is another figure which shows the structure of the encoding data of MV of this invention. It is a schematic diagram which shows the 3rd Embodiment of this invention and shows how the value of the binary data allocated to each MV index is determined. (A) schematically shows a relationship between a PMV candidate and an easily selectable MV candidate, and (b) shows an example of binary data used as a value of an MV index to be arithmetically encoded. It is a block diagram of the moving image encoder 100 "of the 5th Embodiment of this invention. It is a block diagram of the variable-length encoding part 115 'of the 5th Embodiment of this invention. It is a block diagram of the MVD absolute value encoding part 1152 'by the 5th Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Moving picture encoding device)
FIG. 1 is a block diagram showing a configuration of a moving image encoding apparatus 100 according to the first embodiment of the present invention. The moving image encoding apparatus 100 includes a motion vector encoding unit 111, a conversion unit 112, a quantization unit 113, a variable length encoding unit 115, an inverse quantization unit 116, an inverse conversion unit 117, a frame memory 118, and an intra prediction unit 121. , An inter prediction unit 123, a motion estimation unit 124, a prediction mode determination unit 125, a subtraction unit 101, and an addition unit 102. Reference numeral 103 in the figure is a local decoded signal extracted from the frame memory 118.

  In the moving image encoding apparatus 100, input image data is divided into a plurality of blocks, and encoding is performed in units of the divided blocks. In block unit coding, a subtraction signal is calculated by the subtraction unit 101 using a prediction signal obtained as an output of the prediction mode determination unit 125. The residual signal is converted into a conversion coefficient by DCT or the like by the conversion unit 112, and the conversion coefficient is further quantized by the quantization unit 113. On the other hand, the quantized transform coefficient is input to the variable length encoding unit 115 and encoded. On the other hand, the quantized transform coefficient is inversely quantized by the inverse quantization unit 116 and subjected to inverse DCT transform or the like by the inverse transform unit 117 to obtain a decoded residual signal. The decoded residual signal is added to the prediction signal by the adding unit 102 and stored in the frame memory 118 as a local decoded signal.

(Prediction signal generation)
The generation of the prediction signal is performed according to the following procedure. The intra prediction unit 121 generates an intra prediction signal that is a prediction signal of the input signal, and calculates an encoding cost for evaluating good prediction from the intra prediction signal and the input signal. The intra prediction signal and the coding cost (intra prediction cost) are input to the prediction mode determination unit 125.

  For the coding cost, the sum of absolute differences between the input signal and the prediction signal may be used, or the sum of squared differences may be used. Furthermore, the code amount of the prediction parameter may be taken into consideration.

  The inter prediction unit 123 responds to the frame ID for specifying the reference frame and the motion vector (MV) indicating the extraction position on the reference frame as a relative position from the encoding target block, which are input from the motion estimation unit 124. An inter prediction signal is generated. The generated inter prediction signal is input to the motion estimation unit 124 and the prediction mode determination unit 125.

  The motion estimation unit 124 uses the inter prediction signal generated by the inter prediction unit 123 for a predetermined frame ID and a prediction parameter within a predetermined MV range, and performs encoding to evaluate the goodness of prediction. Calculate the cost.

  Then, the motion estimation unit 124 inputs, to the prediction mode determination unit 125, a prediction parameter that minimizes the calculated coding cost and the cost (inter prediction cost) among the prediction parameters described above.

  The prediction mode determination unit 125 compares the intra prediction cost and the inter prediction cost, and selects the prediction mode with the smaller cost. The selected prediction mode and prediction parameter are input to the variable length coding unit 115. In addition, the prediction signal corresponding to the selected prediction mode is input to the subtraction unit 101 and the addition unit 102.

(Variable length encoding unit 115)
The variable length coding unit 115 codes not only the quantized transform coefficient but also the prediction mode and the prediction parameter output from the prediction mode determination unit 125. When the prediction mode is inter prediction, the frame ID of the prediction parameters is encoded by the variable length encoding unit 115. The MV is subjected to processing necessary for encoding by the motion vector encoding unit 111, and is encoded as encoded data by the variable length encoding unit 115.

(Motion vector encoding unit 111)
As shown in FIG. 3, the motion vector encoding unit 111 includes an MV storage unit 1111, a PMV candidate calculation unit 1112, a PMV determination unit 1113, an MVD calculation unit 1114, and an MV flag determination unit 1115. Reference numeral 301 in the figure indicates a PMV candidate and the number of PMV candidates, and 302 indicates a PMV.

  The MV storage unit 1111 stores the MV input to the motion vector encoding unit 111. The PMV candidate calculation unit 1112 calculates a motion vector predictor candidate (PMV candidate) with reference to the MV stored in the MV storage unit 1111. The calculated PMV candidates and the number of PMV candidates are input to the PMV determination unit 1113 and the MV flag determination unit 1115. The number of PMV candidates is also input to a PMV index encoding unit 1151 provided in the variable length encoding unit 115.

  The PMV determination unit 1113 receives the MV, the PMV candidates, and the number of PMV candidates. Among the PMV candidates, the PMV candidate that minimizes the cost function PMVCOST (MV, PMV) calculated from the MV and the PMV candidates is used as the PMV. select. A flag (PMV index) for specifying the selected PMV is input to the PMV index encoding unit 1151, and the PMV index encoding unit 1151 encodes the PMV index based on the number of PMV candidates (described later). . The selected PMV is input to the MVD calculation unit 1114 and the MV flag determination unit 1115.

  The MVD calculation unit 1114 calculates the absolute value (MVD absolute value) of the difference (MVD) between the input MV and PMV, and the MVD absolute value encoding unit 1152 and the MV flag determination unit in the variable length encoding unit 115. 1115 is entered. The MVD absolute value encoding unit 1152 encodes the MVD absolute value.

  The MV flag determination unit 1115 calculates the MV candidate and the number of MV candidates from the input MV, PMV, PMV candidate, and the number of PMV candidates, and further specifies a flag (MV for specifying the MV from the calculated MV candidates. Index). The number of MV candidates and the MV index are input to the MV index encoding unit 1153 in the variable length encoding unit 115. The MV index encoding unit 1153 encodes the MV index based on the number of MV candidates.

(Encoding of PMV index and MV index)
The PMV index and the MV index are each encoded according to the number of candidates.

  Specifically, the index is not encoded when the number of candidates is 1. When the number of candidates is 2, “0” or “1” is encoded in binary. When the number of candidates is 3, “0”, “10”, or “11” is encoded. When the number of candidates is 4, one of “00”, “01”, “10”, and “11” is encoded. If the selection ratio of each candidate is biased, when the number of candidates is 4, any one of “0”, “10”, “110”, and “111” may be encoded. . For binary encoding, a fixed-length code having the above-described value as encoded data as it is may be used, but it is more preferable to use an arithmetic code that adaptively changes the occurrence probability of 0 and 1.

  Hereinafter, the PMV candidate calculation method and the function PMVCOST will be described, and the MV existence area, which is the point of the present invention, will be described. Finally, details of the MV flag determination unit 1115 will be described.

(PMV candidate calculation method)
FIG. 10 is a diagram for explaining a PMV candidate calculation method. The MVs of blocks adjacent to the left, top, and top right of the encoding target block are expressed as mv_a, mv_b, and mv_c, respectively. In addition, the MV of the block at the same spatial position on the encoding target block and the reference image is expressed as mv_col. In addition, the median value of MVs of spatially adjacent blocks is expressed as mv_med. Specifically, calculation is performed by taking the median value for each of the X component and Y component of mv_a, mv_b, and mv_c.

  As the PMV candidates, it is preferable to use mv_med and mv_a, or mv_med and mv_col. However, the PMV candidates of the present invention are not limited to the above two types, and the number of PMV candidates and PMV candidates are as follows. Other values can also be used. For example, the PMV candidate calculation unit 1112 may use mv_med, mv_a, and mv_col as PMV candidates.

(Function PMVCOST)
The function PMVCOST (MV, PMV) is a function that inputs MV and PMV and calculates a cost corresponding to the distance between MV and PMV. When encoding MV using PMV, the code amount of MV can be reduced by selecting PMV whose distance is close to MV. The function PMVCOST calculates this distance.

PMVCOST is a function calculated from an MVD absolute value that is a difference absolute value between MV and PMV. That is, using arbitrary functions F and G,
PMVCOST = F (| MV_X−PMV_X |) + G (| MV_Y−PMV_Y |)
A function that can be expressed as Here, || is a symbol indicating an absolute value. By defining the cost as such a function, even if MV and PMV cannot be obtained, the cost can be calculated if an MVD absolute value that is a difference absolute value between MV and PMV is obtained.

For example, city value distance PMVCOST = | MV_X−PMV_X | + | MV_Y−PMV_Y |
Alternatively, the square of the Euclidean distance PMVCOST = | MV_X−PMV_X | ^ 2 + | MV_Y−PMV_Y | ^ 2
Can be used as a distance.

As another distance, a cost function considering the MV code amount can be considered. More specifically,
PMVCOST = R (| MV_X−PMV_X |) + R (| MV_Y−PMV_Y |). Here, R (X) is a function for inputting an absolute value of the MV as X and calculating a code amount or an amount corresponding to the code amount.

  The distribution of MVD absolute values (= X) is often concentrated around X = 0. In this case, it is preferable to use a code such that the code amount is a function R (X) that increases as X increases. Therefore, LOG (X + 1) is preferably used as R (X). Further, when the MVD absolute value X is 0, encoding for indicating positive / negative is unnecessary, and therefore, the MVD can be encoded by 1 bit less than when the MVD absolute value X is other than 0. In consideration of this property, it is also preferable to use R (X) = 1 (when X is 0) and R (X) = X + 2 (when X is other than 0).

(Area where MV can exist)
FIG. 11 is a diagram for explaining the MV existence possible area when the number of PMV candidates is two. The PMV candidate calculation unit 1112 calculates PMV1 and PMV2 as PMV candidates, and the PMV determination unit 1113 indicates that PMV1 is selected as PMV. In addition, the city value distance is used for PMVCOST. The fact that PMV1 is selected by the PMV determination unit 1113 means that the distance between MV and PMV1 is equal to or smaller than the distance between MV and PMV2 with PMVCOST as the distance. That is,
PMVCOST (MV, PMV1) <= PMVCOST (MV, PMV2)
Holds. Therefore, the area where MV can exist is limited to the above-described area (shaded area in FIG. 11).

  When PMVCOST is a city value distance, a line dividing the MV-existing area and the other area is a perpendicular bisector L1 of PMV1 and PMV2.

FIG. 12 is a diagram for explaining a possible MV existence area when the number of PMV candidates is three. Also in this case, the MV can exist area (shaded area in FIG. 12) is expressed by the following two formulas PMVCOST (MV, PMV1) <= PMVCOST (MV, PMV2)
PMVCOST (MV, PMV1) <= PMVCOST (MV, PMV3)
It is limited to the area where both of the above hold. Thus, when the number of PMV candidates is m (m is an integer equal to or greater than 1), the MV existence region is limited as a solution of simultaneous equations composed of m-1 inequalities.

  FIG. 13 is a diagram illustrating a case where PMVCOST is not a city value distance. As shown in the figure, when PMVCOST is not the city value distance, the line dividing the MV-existing region and the other region may be a curve. Although not shown, the line is not necessarily a smooth curve, and may be a broken line.

(MV flag determination unit 1115)
FIG. 5 is a block diagram showing a configuration of the MV flag determination unit 1115. The MV flag determination unit 1115 includes an MV candidate calculation unit 5001, an MV candidate determination unit 5002, and an MV index determination unit 5003. In the MV flag determining unit 1115, information on the absolute value of the difference (MVD absolute value) between PMV and PMV and MV is known.

  FIG. 14 is a diagram showing the relationship between MV candidates and MV existence possible areas when the number of PMV candidates is 2 (PMV1 and PMV2). In the information of only the PMV and the MVD absolute value, the MV candidates usually include four MV candidates from MV1 to MV4 (when both the X component and the Y component of the MVD absolute value are other than 0). Therefore, in order to specify the MV from the PMV and MVD absolute values, an MV index for specifying one of the four MV candidates is necessary. At this time, the code amount necessary for specifying the MV from the four MV candidates is 2 bits.

  If the PMV to be encoded is PMV1 and the MV existence area is limited as in FIG. 11, MV4 is out of the MV existence area in FIG. If the MV candidates outside the MV existable area are excluded, the number of valid MV candidates is 3. In this case, the code amount necessary for encoding the MV index can be reduced to within 2 bits.

  Hereinafter, in order to distinguish between MV candidates determined from PMV and MVD absolute values and MV candidates in the MV existence area in the above-mentioned MV candidates, the former is a temporary MV candidate and the latter is an MV candidate. I will call it.

(MV candidate calculation unit 5001)
The MV candidate calculation unit 5001 receives the PMV and the MVD absolute value as inputs, and calculates the temporary MV candidates and the number of temporary MV candidates. FIG. 7 is a flowchart showing an operation flow of the MV candidate calculation unit 5001. The X component and Y component of the MVD absolute value and PMV are expressed as (mvd_x, mvd_y) and (pmv_x, pmv_y), respectively.

  S701 If mvd_x == 0 and mvd_y == 0, the process proceeds to S702, and if not, the process proceeds to S703.

  S702: The temporary MV candidate number n is set to 1, and the temporary MV candidate tMV1 is calculated as follows.

tMV1 = (pmv_x, pmv_y)
S703 If mvd_x == 0, the process proceeds to S704; otherwise, the process proceeds to S705.

  S704: The temporary MV candidate number n is set to 2, and temporary MV candidates tMV1 and tMV2 are calculated as follows.

tMV1 = (pmv_x, pmv_y + mvd_x)
tMV2 = (pmv_x, pmv_y−mvd_x)
S705 If mvd_y == 0, the process proceeds to S706, and if not, the process proceeds to S707.

  S706: The number of temporary MV candidates n is 2, and temporary MV candidates tMV1 and tMV2 are calculated as follows.

tMV1 = (pmv_x + mvd_x, pmv_y)
tMV2 = (pmv_x−mvd_x, pmv_y)
S707: The number of temporary MV candidates n is 4, and the temporary MV candidates tMV1, tMV2, tMV3, and tMV4 are calculated as follows.

tMV1 = (pmv_x + mvd_x, pmv_y + mvd_y)
tMV2 = (pmv_x + mvd_x, pmv_y−mvd_y)
tMV3 = (pmv_x−mvd_x, pmv_y + mvd_y)
tMV4 = (pmv_x−mvd_x, pmv_y−mvd_y)
(MV candidate determination unit 5002)
The MV candidate determination unit 5002 receives the PMV, the temporary MV candidates, the number of temporary MV candidates, and the MVD absolute value, and determines whether or not the temporary MV candidate is in the MV existence area. By this determination, the MV candidates and the number of MV candidates obtained by eliminating the temporary MV candidates outside the MV existence area are output to the MV index determination unit 5003.

  FIG. 8 is a flowchart showing an operation flow of the MV candidate determination unit 5002.

  In step S <b> 801, the distance min_dist between PMV and MV is calculated using the cost function PMVCOST used in the PMV determination unit 1113. Although no MV is input to the MV candidate determination unit 5002, as described above, the function PMVCOST can be calculated from an MVD absolute value that is a difference absolute value between PMV and MV.

  The distance between the temporary MV candidate and the PMV candidate is always greater than or equal to min_dist when the temporary MV candidate is in the MV existence possible area. Conversely, if the distance between the temporary MV candidate and the PMV candidate is less than min_dist, the temporary PMV candidate can be excluded as not being an MV candidate.

  S802: A variable nn indicating an MV candidate index is initialized to 1.

  L801 A variable i indicating a temporary MV candidate index is set as a loop variable, and a loop between L802 and S806 is performed n times until i becomes equal to the number of temporary MV candidates n.

  L802 A variable j indicating a PMV candidate index is set as a loop variable, and a loop between S803 and S804 is performed m times until j is equal to the number of PMV candidates m.

  S803: The distance dist_ij between the temporary MV candidate tMVi indicated by the temporary MV candidate index i and the PMV candidate PMVj indicated by the PMV candidate index j is calculated using the cost function PMVCOST used by the PMV determination unit 1113.

  S804: When dist_ij is less than min_dist, exit the loop of variable j indicating the PMV index (L802 loop). If not, the L802 loop is continued. Note that dist_ij is less than min_dist means that the provisional MV candidate tMVi is outside the MV existence region.

  S805 Since it is determined that the temporary MV candidate tMVi is in the MV existence possible area, it is added to the MV candidate fMV. Specifically, tMVi is substituted for fMVnn.

S806: The variable nn indicating the MV candidate index is incremented by 1.
The number of MV candidates nn and the MV candidate fMV are obtained by the above operation flow.

  Note that, by integrating the MV candidate calculation unit 5001 and the MV candidate determination unit 5002, it is possible to calculate MV candidates without calculating all the temporary MV candidates. In this case, every time a temporary MV candidate is calculated, the determination by the loop of L802 is performed.

(MV index determination unit 5003)
The MV index determining unit 5003 determines whether the MV is one of the MV candidates from the input MV (MV of the encoding target block), the MV candidates, and the number of MV candidates, and determines an MV index indicating the MV index. To do.

  FIG. 9 is a flowchart showing an operation flow of the MV index determination unit 5003.

  L901 S901 is looped from 1 to the number of MV candidates nn for the loop variable i indicating the index of the MV candidate.

  S901: It is determined whether the MV candidate fMVi matches MV. If they match, the process proceeds to S902. If they do not match, the loop continues until the loop end condition.

  S902 Assign i to the MV index ii.

  The MV index ii is obtained by the above operation flow.

  FIG. 15 is a diagram showing a structure of MV encoded data according to the present invention. As shown in FIG. 15, the MV is encoded by being divided into a PMV index indicating PMV, an absolute value of a difference between PMV and MV (MVD absolute value), and an MV index. The MVD absolute value is encoded in the order of the X component and the Y component. When the number of PMV candidates is 1, the PMV index is not encoded. When the number of MV candidates is 1, the MV index is not encoded.

  As described above, in the motion picture encoding apparatus 100 described above, in the motion vector encoding method for calculating a plurality of PMV candidates as MV prediction values and selecting and encoding a PMV suitable for MV encoding, the PMV candidate and the PMV The MV can be efficiently encoded by limiting the MV existence possible area using

  More specifically, after encoding the difference absolute value (MVD absolute value) between MV and PMV, the temporary MV candidates determined from the PMV and MVD absolute values are excluded from the MV existing area. The MV candidates are limited by and an MV index that uniquely identifies the MV from the limited MV candidates is encoded.

  By limiting the MV existence area, it is possible to reduce the code amount of the MV index for identifying the MV from the MV candidates.

(Decryption device)
FIG. 2 is a block diagram showing the configuration of the moving picture decoding apparatus 200 according to the second embodiment of the present invention. The moving picture decoding apparatus 200 includes a motion vector decoding unit 211, a variable length coding decoding unit 201, an inverse quantization unit 116, an inverse transform unit 117, a frame memory 118, an intra prediction unit 121, an inter prediction unit 123, and a prediction mode selection unit. 225 and the addition unit 102. In addition, the code | symbol 203 in a figure means frame ID among prediction parameters.

  Similar to the moving image encoding apparatus 100, the moving image decoding apparatus 200 divides input image data into a plurality of blocks, and decoding is performed in units of divided blocks. In block unit decoding, the variable length coding / decoding unit 201 decodes the prediction mode and the quantized transform coefficient.

  The prediction mode is a mode indicating whether the block is intra prediction or inter prediction. When the prediction mode is inter prediction, the frame ID and MV which are prediction parameters are decoded. The frame ID is decoded by the variable length coding / decoding unit 201. The MV is decoded by the variable length coding / decoding unit 201 and the motion vector decoding unit 211.

  The intra prediction unit 121 and the inter prediction unit 123 generate an intra prediction signal and an inter prediction signal from the decoded image (reproduced image) stored in the frame memory 118 using the decoded prediction parameter. The generated prediction signal is input to the prediction mode selection unit 225. The prediction mode selection unit 225 selects a prediction signal corresponding to the prediction mode and outputs it to the addition unit 102.

  The decoded quantized transform coefficient is inversely quantized by the inverse quantization unit 116, converted by the inverse transform unit 117, such as inverse DCT, and the decoded residual signal is output to the addition unit 102. Adder 102 adds the prediction signal and the decoded residual signal to generate a local decoded signal. The generated local decoded signal is output to the frame memory 118. The reproduced image output to the frame memory 118 is used as a reference image by the intra prediction unit 121 and the inter prediction unit 123 and is output to the outside of the video decoding device 200.

  FIG. 4 is a block diagram showing a configuration of the motion vector decoding unit 211. The motion vector decoding unit 211 includes an MV storage unit 1111, a PMV candidate calculation unit 1112, a PMV selection unit 2113, and an MV selection unit 2115.

  The MV storage unit 1111 stores the MV decoded by the motion vector decoding unit 211. The PMV candidate calculation unit 1112 calculates a PMV candidate. The calculated PMV candidates and the number of PMV candidates are input to the PMV selection unit 2113 and the MV selection unit 2115. The number of PMV candidates is input to the PMV index decoding unit 2011 in the variable length coding / decoding unit 201. The PMV index decoding unit 2011 decodes the PMV index using the number of PMV candidates and inputs the PMV index to the PMV selection unit 2113. The PMV selection unit 2113 selects a candidate indicated by the PMV index from the PMV candidates, and inputs the candidate as the PMV to the MV selection unit 2115.

  Further, the variable length coding / decoding unit 201 decodes the MVD absolute value by the internal MVD absolute value decoding unit 2012 following the PMV index, and inputs the MVD absolute value to the MV selection unit 2115.

  FIG. 6 is a block diagram illustrating a configuration of the MV selection unit 2115. The MV selection unit 2115 includes an MV candidate calculation unit 5001, an MV candidate determination unit 5002, and an MV extraction unit 6004. The MV candidate calculation unit 5001 and the MV candidate determination unit 5002 have been described in the first embodiment and thus will not be described. However, among these provisional MV candidates calculated from PMV and MVD absolute values, PMV and PMV candidates are used. Are calculated in the possible MV existence area (= MV candidate). The MV extraction unit 6004 decodes the MV by selecting a candidate indicated by the MV index from the MV candidates.

  As described above, the moving picture decoding apparatus 200 calculates a plurality of PMV candidates as MV prediction values, selects a PMV suitable for MV encoding, and uses the PMV candidate and PMV in the motion vector decoding method used for decoding. Thus, the encoded MV can be efficiently decoded by limiting the MV existence area.

  More specifically, after the absolute difference value (MVD absolute value) between MV and PMV is decoded, the temporary MV candidates determined from the PMV and the MVD absolute value are excluded from the MV existence area. Thus, the MV candidates are limited, and the MV is decoded by decoding the MV index that identifies the MV from the limited MV candidates.

  By limiting the possible MV areas, it is possible to decode MVs from encoded data in which the code amount of the MV index necessary for specifying MVs from MV candidates is reduced.

(Moving picture encoding apparatus 100 ′)
Next, a moving picture coding apparatus 100 ′ according to a third embodiment of the present invention will be described below with reference to FIGS. 17 to 19 and FIG. FIG. 23 is a diagram illustrating a configuration of MV encoded data. As shown in FIG. 23, the MV encoded data is composed of “MVD” and “other than MVD”.

  The moving image encoding apparatus 100 ′ according to the present embodiment performs arithmetic encoding in consideration of the appearance probability of the MV candidate, so that the higher the existence probability of the MV candidate, the more the MV that is a flag for specifying the MV candidate. The tendency that the code length after index encoding becomes short is strengthened, and the code amount of the MV index is reduced as compared with the case where the appearance probability is not taken into consideration.

  That is, the moving picture encoding apparatus 100 ′ of the present embodiment further reduces the code amount of the portion other than “MVD” in FIG. 23 compared to the moving picture encoding apparatus 100 of the first embodiment, The code amount of MV encoding can be further reduced. For this purpose, the moving picture coding apparatus 100 ′ according to the present embodiment performs an operation of rearranging the MV candidates in an order in which it is estimated that the existence probability of the MV is high.

  FIG. 17 is a block diagram showing a configuration of a moving image encoding apparatus 100 ′ according to the third embodiment of the present invention. Similar to the video encoding device 100 of the first embodiment, the video encoding device 100 ′ includes a transform unit 112, a quantization unit 113, a variable length coding unit 115, an inverse quantization unit 116, and an inverse transform unit. 117, a frame memory 118, an intra prediction unit 121, an inter prediction unit 123, a motion estimation unit 124, a prediction mode determination unit 125, a subtraction unit 101, and an addition unit 102, but the configuration of the motion vector encoding unit 111 ′ is This is different from the motion vector encoding unit 111 of the first embodiment. The other members excluding the motion vector encoding unit 111 ′ perform the same operations as those in the first embodiment, and thus description of these members is omitted, and the motion vector encoding unit 111 ′ is illustrated in FIGS. Will be described with reference to.

  As shown in FIG. 18, the motion vector encoding unit 111 ′ includes an MV storage unit 1111, a PMV candidate calculation unit 1112, a PMV determination unit 1113, and an MVD calculation unit 1114, like the motion vector encoding unit 111. The MV flag determination unit 1115 ′ is different from the MV flag determination unit 1115 of the first embodiment.

  As shown in (a) of FIG. 19, the MV flag determining unit 1115 ′ is provided with an MV candidate sorting unit 5004 between the MV candidate calculating unit 5001 and the MV candidate determining unit 5002. The MV candidate sorting unit 5004 includes an MV candidate cost calculating unit 5004a and a sorting unit 5004b as shown in FIG.

  Note that the MV candidate sorting unit 5004 may be provided after the MV candidate determination unit 5002.

  The MV candidate sorting unit 5004 receives the temporary MV candidates and the number of temporary MV candidates as input values from the MV candidate calculation unit 5001. The MV candidate cost calculation unit 5004a of the MV candidate sorting unit 5004 calculates the cost of the temporary MV candidate by applying a function called MVCOST to each temporary MV candidate with the coordinate value of the temporary MV candidate as an input. MVCOST is a function that gives a larger value (or smaller value) as the appearance probability of an MV candidate is higher, and various functions can be applied. Specifically, it is appropriate that the coordinate value of the MV candidate and the coordinate value of each PMV candidate (or PMV) are input and the cost value increases as the distance between the temporary MV candidate and the PMV candidate increases. A specific example of such a function will be described later.

  The sorting unit 5004b rearranges the temporary MV candidates in ascending order of cost. Alternatively, the sorting unit 5004b may rearrange the temporary MV candidates in descending order of cost.

  The MV candidate determination unit 5002 determines MV candidates from the temporary MV candidates by the same method as in the first embodiment, and outputs the MV candidates and the number of MV candidates to the MV index determination unit 5003.

  The MV index determination unit 5003 receives the MV candidates and the number of MV candidates, and determines the value of binary data of the MV index to be assigned to each MV candidate. Here, how to determine the value of binary data to be assigned to each MV index will be described below with reference to FIG.

  FIG. 24A is a schematic diagram showing a relationship between a PMV candidate and an MV candidate that can be easily selected. In FIG. 24, PMV0 indicates that the PMV is selected by the PMV determination unit 1113. Further, MV1 to MV4 indicate MV candidates output from the MV candidate determination unit 5002.

  In general, among MV candidates, the smaller the distance from a PMV candidate other than the selected PMV candidate, the more likely that the MV is determined. The MV index determination unit 5003 determines the value of the binary data of the MV index using this tendency.

  That is, as will be described later, when arithmetic coding is performed, a high probability is assigned to “0” and a low probability is assigned to “1”, but the MV index determination unit 5003 uses the MV candidate sorting unit 5004. “00”, “01”, “10”, and “11” are assigned to the MV candidates in the order sorted in the order (in ascending order of cost). As a result, as shown in FIG. 24B, “00” is assigned to MV1, “01” to MV2, “10” to MV3, and “11” to MV4.

  Then, the MV index determination unit 5003 outputs the determined MV index to the MV index encoding unit 1153. Then, the variable length encoding unit 115 arithmetically encodes each MV index based on the probabilities assigned to “0” and “1”.

  Therefore, the code word of the MV index has a tendency that the code length of the code word of MV1 becomes short and the code length of the code word of MV4 becomes long. As a result, it can be seen that the average code length of the code word of the MV index becomes shorter and the code amount becomes smaller than in the case of the first embodiment. Note that the MV index encoding unit 1153 may adaptively change the probability assigned to “0” and “1” when performing arithmetic encoding.

  In addition, the variable length coding unit 115 may assign a short code to the order sorted by the MV candidate sorting unit 5004 (the order from the lowest cost) using fixed length coding. For example, “0”, “10”, “110”, and “111” can be assigned.

  As described above, when binarization is performed by assigning an MV index according to an MV candidate having a high appearance probability, a method of biasing the probability of “0” and “1” when arithmetic coding is performed. Is applicable even when the MV candidate determination unit 5002 is not used.

  Note that when the MV candidate determination unit 5002 is used, it is more efficient to perform encoding according to an MV candidate having a high appearance probability. For example, when the number of MV candidates is 3 by the MV candidate determination unit 5002, a short code (or an arithmetic code in the case of an arithmetic code) such as “0”, “10”, and “11” has a high appearance probability. The amount of codes can be reduced by assigning a binary number sequence.

  In addition, when the number of MV candidates is not an integer power of 2, as in the case where the number of MV candidates is 3, generally the codes or binary number sequences corresponding to the MV candidates are unequal. In this case, the effectiveness of the process of assigning the MV index according to the appearance probability of the MV candidate is large. In consideration of this characteristic, this process is applied only when it is most effective. That is, only when the number of MV candidates matches a specific condition (for example, when it is not an integer power of 2), sorting of MV candidates is performed. It is also possible to do this. Since this process requires the number of MV candidates determined to be valid, the MV candidate sorting unit 5004 is provided after the MV candidate determination unit 5002.

  The process of assigning the value of the MV index according to the MV candidate having a high appearance probability does not need to use the MV candidate literal sort. If the number of MV candidates is at most about 4, processing can be easily performed by combining a plurality of conditional branches.

For example, whether the MV candidate cost is in the order of cost A, cost B, cost C, cost D, cost A <= cost B <= cost C <= cost D,
IF (cost A <= cost B and cost B <= cost C and cost C <= cost D)
Can be determined by the conditional branch statement. When this condition is met, “00”, “01”, “10”, and “11” are assigned to the corresponding MV indexes, that is, MV candidates of cost A to cost D. Whether the order is other can be easily implemented by arranging similar conditional branch statements.

  In this case, the sorting unit 5004b is not necessary, and the MV index determination unit 5003 performs MV index determination by conditional branching using the MV candidate cost as described above.

  Further, if the minimum (or maximum) N MV candidate costs are not allocated in the order of the MV candidate costs but are allocated in the order of the MV candidate costs, the processing becomes easier. For example, the process of calculating the MV candidate with the minimum (or maximum) MV candidate cost and allocating a specific MV index can be realized with only the process of specifying the MV candidate with the minimum value, and thus can be realized with a very simple configuration.

  The present invention is also intended to include such a simplified arrangement that does not provide complete ordering of MV candidates.

(Specific example of MVCOST function)
As specific examples of the MVCOST function, the following functions 1 to 4 can be considered.
Function 1: MV candidate cost = Σ (| mv_x−pmv _i —x | + | mv_y−pmv _i —y |) (i: 1 to N, pmv _i is a PMV candidate not selected)
Function 1 ′: MV candidate cost = Σ ((mv_x−pmv _i _x) ^ 2 + (mv_y−pmv _i _y) ^ 2) (i: 1 to N, pmv _i are unselected PMV candidates, and ^ is a square Indicate)
Function 2: MV candidate cost = | mv_x | + | mv_y |
Function 3: MV candidate cost = | mv_x−pmv ₀ _x | + | mv_y−pmv ₀ _y | (where pmv ₀ is the selected PMV candidate)
Function 4: MV candidate cost = MIN (| mv_x−pmv _i —x | + | mv_y−pmv _i —y |) (i: 1 to N, PMv _i is not selected PMV candidate)
The MVCOST function to be used is preferably function 1, but function 1 ′ or function 2 to function 4 may be used. Further, the function used as the MVCOST function may be changed according to the frame type. For example, the function 1 may be used as the MVCOST function in the P frame, and the function 2 may be used as the MVCOST function in the B frame. As will be described later, the MV index encoding unit 1153 may encode the information indicating the MVCOST function used by the MV candidate cost calculation unit 5004a and record it in the header of the encoded data.

  Note that the PMV used for calculating the MV candidate cost (input to the MVCOST function) need not be limited to the PMV candidate calculated by the PMV candidate calculation unit 1112, that is, the PMV candidate to be selected by the PMV index (so “MV” “PMV candidates used for calculating candidate costs” are simply described as “PMV candidates used for calculating candidate costs”). For example, when the PMV candidate calculation unit 1112 calculates mv_med and mv_col as PMV candidates, the PMV used for MV candidate cost calculation may be mv_med, mv_col, and mv_a.

  In order to reduce the code amount of the PMV index, it is necessary to limit the number of PMV candidates to a small number (2 is appropriate), but in order to estimate the appearance probability of the MV candidate, This is because there is no limit.

  Note that a PMV that is completely different from the PMV candidate to be selected in the PMV index, for example, mv_a, may be used as a PMV candidate used for MV candidate cost calculation. Considering PMV in this way, function 2 is the same as using {0, 0} as PMV.

  The PMV for calculating the MV candidate cost may be configured to use the PMV candidate of the PMV candidate calculating unit 1112 as it is as described above, or the PMV candidate calculating unit 1112 may calculate the MV candidate cost separately from the PMV candidate. It is good also as a structure which calculates PMV. Further, a configuration may be provided in which means for calculating a PMV different from the PMV candidate calculation unit 1112 is provided in the motion vector encoding unit 111 ′.

  In this case, a plurality of PMV candidates are calculated, a PMV suitable for MV encoding is selected, and a single PMV is calculated instead of an encoding / decoding apparatus that encodes / decodes the difference between MV and PMV. The present invention can also be applied to an encoding device / decoding device that encodes / decodes a difference between a signal and a PMV. In this case, the present invention can also be realized by a configuration of an encoding device that does not include the PMV candidate calculation unit 1112, the PMV determination unit 1113, the PMV index encoding unit 1151, and the MV candidate determination unit 5002.

(PMVCOST function used by the PMV determination unit 1113)
As described above, the PMV determination unit 1113 receives the MV, the PMV candidates, and the number of PMV candidates, and selects among the PMV candidates candidates for minimizing the cost function PMVCOST (MV, PMV) calculated from the MV and PMV. Although selected as PMV, the PMV determination unit 1113 may dynamically change the function used as the PMVCOST function according to the load state of the motion vector encoding unit 111.

  That is, as described above, various functions can be applied as the PMVCOST function. Among these functions, the PMVCOST = | MV_X−PMV_X | + | MV_Y-PMV_Y |, which does not improve the coding efficiency so much, but has a small amount of calculation. However, there are various things such as a large amount of calculation. Therefore, the PMV determination unit 1113 uses a function with high encoding efficiency when the load is light, depending on the load state of the motion vector encoding unit 111, and when the load is above a certain level, the amount of computation The PMV may be selected using a function having a small value. In this case, the MV index encoding unit 1151 encodes information (flag) indicating the PMVCOST function used by the PMV determination unit 1113 and records it in the header of the encoded data. The header to be recorded may be a sequence header, a picture header, a slice header, an MB header, or the like. In addition, there is a method in which data such as sequence parameters and picture parameters is encoded first without encoding a header for each sequence or picture, and the data is specified and referred to by an index in each sequence and picture. Since such data is also handled as a header, a flag indicating a function may be encoded with these data.

  Thus, by making PMV selectable according to the load state of the motion vector encoding unit 111, it is possible to balance the encoding efficiency and the calculation amount. Further, when the configuration of the encoding device is simplified or when the encoding load is high, by selecting PMVCOST = 0, it is possible to perform encoding without using PMV restrictions, that is, motion vector code. The conversion unit 111 can freely select a PMV without considering PMV restrictions.

  When selecting the PMVCOST function, the MV candidate determination unit 5002 needs to use the same function as the PMVCOST function used in the PMV determination unit 1113. In order to simplify the encoding apparatus, when PMVCOST = 0 is always selected in the encoding apparatus, the provisional MV candidate becomes an MV candidate as it is, and therefore, an MV candidate determination unit 5002 that determines an MV candidate is provided. No configuration is possible.

(About the MVCOST function used by the MV candidate sort unit 5004)
Similarly to the PMVCOST function, the MVCOST function used in the MV candidate sorting unit 5004 is encoded with information (flag) indicating the MVCOST function and recorded in the header of the encoded data, so that the encoding efficiency and the calculation amount can be calculated. Can be balanced. If there is a strong demand for simplification of the configuration of the encoding device and speeding up of the encoding device, setting MVCOST = 0 can eliminate the sort itself and reduce the amount of calculation.

  When the PMVCOST function or the MVCOST function is changed depending on the frame type, it is appropriate to encode a flag indicating the MVCOST function for each frame type.

(Decoding device 200 ')
Next, a decoding device 200 ′ according to the fourth embodiment of the present invention will be described below with reference to FIGS.

  FIG. 20 is a block diagram showing a configuration of a decoding apparatus 200 'according to the fourth embodiment of the present invention. Similar to the decoding device 200 of the second embodiment, the decoding device 200 ′ includes a variable-length coding / decoding unit 201, an inverse quantization unit 116, an inverse transform unit 117, a frame memory 118, an intra prediction unit 121, and an inter prediction unit. 123, the prediction mode selection unit 225, and the addition unit 102, but the motion vector decoding unit 211 ′ is different from the motion vector decoding unit 211 of the second embodiment. The other members excluding the motion vector decoding unit 211 ′ perform the same operations as those in the second embodiment, so description of these members will be omitted, and refer to FIGS. 21 and 22 for the motion vector decoding unit 211 ′. I will explain it.

  As shown in FIG. 21, the motion vector decoding unit 211 ′ includes an MV storage unit 1111, a PMV candidate calculation unit 1112, and a PMV selection unit 2113, like the motion vector decoding unit 211, but the MV selection unit 2115 ′ This is different from the MV selection unit 2115 of the second embodiment.

  As shown in FIG. 22, the MV selection sorting unit 5004 is provided between the MV candidate calculation unit 5001 and the MV candidate determination unit 5002 in the MV selection unit 2115 ′.

  The MV candidate sorting unit 5004 receives the temporary MV candidates and the number of temporary MV candidates as input values from the MV candidate calculation unit 5001. The MV candidate sorting unit 5004 performs the same sort as the MV candidate sorting unit 5004 of the MV flag determining unit 1115 ′. When the MV candidate sorting unit 5004 of the MV flag determining unit 1115 ′ dynamically changes the MVCOST function, the MV candidate sorting unit 5004 stores information indicating the MVCOST function recorded in the header of the encoded data. Sorting is performed by decoding and referencing.

  In addition, when the PMV determination unit 1113 and the MV candidate determination unit 5002 dynamically change the PMVCOST function, the information indicating the PMVCOST function recorded in the header of the encoded data is decoded and the MV candidate determination unit 5002 Make a decision.

  Then, the MV candidate determination unit 5002 determines MV candidates from the temporary MV candidates by the same method as in the second embodiment, and outputs the MV candidates and the number of MV candidates to the MV extraction unit 6004.

  Finally, the MV extraction unit 6004 receives the MV candidates and the number of MV candidates, and decodes the MV by selecting a candidate indicated by the MV index from the MV candidates.

  Note that in a decoding apparatus that calculates a single PMV and decodes the difference between MV and PMV, a decoding apparatus that does not include the PMV candidate calculation unit 1112, the PMV selection unit 2113, the PMV index decoding unit 2011, and the MV candidate determination unit 5002. It can also be realized with a configuration.

(Moving picture encoding apparatus 100 ")
Finally, a moving picture coding apparatus 100 ″ according to a fifth embodiment of the present invention will be described below with reference to FIGS.

  The moving image coding apparatus 100 ″ of the present embodiment improves the coding efficiency of the MVD absolute value by adaptively switching the probability (context) used for MVD arithmetic coding according to PMV. .

  This switching is based on the observation result that the magnitude of the MVD absolute value has a tendency corresponding to the type of PMV used to calculate the MVD. More specifically, when the PMV candidates are mv_med and mv_col, when mv_col is PMV, the MVD absolute value tends to be smaller than when mv_med is PMV. In the following configuration, except for the initial value to be described later, “0” is adaptively applied to the arithmetic code even if it is not included as an explicit process which PMV tends to have a smaller MVD absolute value. , The effect can be realized because it is learned as the appearance probability of “1”.

  FIG. 25 is a block diagram showing a configuration of a video encoding device 100 ″ according to the fifth embodiment of the present invention. The video encoding device 100 ″ is the same as the video encoding device 100 according to the first embodiment. Similarly, the motion vector encoding unit 111, the conversion unit 112, the quantization unit 113, the inverse quantization unit 116, the inverse conversion unit 117, the frame memory 118, the intra prediction unit 121, the inter prediction unit 123, the motion estimation unit 124, and the prediction Although the mode determination unit 125, the subtraction unit 101, and the addition unit 102 are provided, the configuration of the variable length coding unit 115 ′ is different from the variable length coding unit 115 of the first embodiment. The other members excluding the variable length coding unit 115 ′ perform the same operations as those in the first embodiment, and thus description of these members is omitted, and the variable length coding unit 115 ′ is illustrated in FIGS. Will be described with reference to.

  As shown in FIG. 26, the variable length encoding unit 115 'includes a PMV index encoding unit 1151, an MVD absolute value encoding unit 1152', and an MV index encoding unit 1153. Since the PMV index encoding unit 1151 and the MV index encoding unit 1153 perform the same operation as in the first embodiment, the MVD absolute value encoding unit 1152 'will be described.

  As shown in FIG. 27, the MVD absolute value encoding unit 1152 'includes a context determination unit 1152a, a probability holding unit 1152b, a probability update unit 1152c, a binarization unit 1152d, and an arithmetic encoding unit 1152e.

  The binarization unit 1152d converts the MVD absolute value into a binary sequence of “0” and “1” and outputs the binary sequence to the arithmetic coding unit 1152e. The binarization unit 1152d converts the MVD absolute value “n” into a binary number sequence in which one “0” is added after n consecutive “1” s. For example, when n is 0, the binary sequence is “0”, and when n is 2, the binary sequence is “110”.

  On the other hand, the context determination unit 1152a outputs the context index determined using the PMV index value input from the PMV determination unit 1113 to the probability holding unit 1152b.

  The context index is a value indicating whether it is an X component or a Y component, a value indicating the size of the MVD of an adjacent block that has already been encoded / decoded, and the number in the vicinity of the binary sequence. Is preferably determined by a combination of a value indicating stepwise and a PMV index value. That is, there are 2 components, 3 values for the MVD size stepwise, 3 steps for the binary sequence, and 2 PMV candidates. In this case, a context index to be output is determined from 24 context indexes of 2 × 3 × 3 × 2. In this case, the probability holding unit 1152b is configured to hold 24 probabilities of “1” and “0”. As long as the value of the PMV index is used, the context index may be determined using other conditions.

  The arithmetic encoding unit 1152e encodes the binary sequence based on the probabilities “1” and “0” held in the probability holding unit 1152b corresponding to the context index input to the probability holding unit 1152b. In the present embodiment, a point that adaptively encodes a binary number sequence (symbol sequence), that is, a probability of “1” or “0” every time each bit is encoded when the binary number sequence is encoded. The feature is that it is updated.

With respect to this, the encoding of the binary sequence will be described by taking as an example the case where the binary sequence input to the arithmetic encoding unit 1152e is “110”. It is assumed that the probability of “1” before encoding is P ₁ .

The arithmetic encoding unit 1152e encodes the first bit “1” of “110” using the probability P ₁ . Then, the arithmetic encoding unit 1152e outputs the encoded bit “1” to the probability updating unit 1152c. Probability updating unit 1152c, when coded bits are input, and updates the probability of "1" to P _2.

Next, the arithmetic encoding unit 1152e encodes the second bit “1” of “110” using the probability P ₂ . Then, the arithmetic encoding unit 1152e outputs the encoded bit “1” to the probability updating unit 1152c. Probability updating unit 1152c, when coded bits are input, and updates the probability of "1" to P _3.

Finally, the arithmetic encoding unit 1152e encodes the third bit “0” of “110” using the probability 1-P ₃ . Then, the arithmetic encoding unit 1152e outputs encoded data obtained by encoding “110” with the probabilities P ₁ , P ₂ , and 1-P ₃ .

  Incidentally, the probability holding unit 1152b holds initial values for initializing the probabilities at the sequence head, frame head, slice head, and the like. The initial value may be “0” or “1” with an equal probability, that is, 0.5, but it is appropriate to use an appropriate value such as an average value of appearance probabilities experimentally obtained in many sequences. It is. In this case, the PMV candidate is obtained using the spatial prediction vector (for example, mv_med) obtained from the MV of the already encoded / decoded block of the same frame and the MV of the already encoded / decoded block of the reference frame. In the case of a prediction vector (for example, mv_col), it is appropriate to set an initial value so that an MVD absolute value with a shorter temporal prediction vector is encoded with a smaller code amount. In the above example, the initial value of the probability of “0” is set so that the temporal prediction vector is larger than the spatial prediction vector.

  As described above, the moving image encoding apparatus 100 ″ can switch the appearance probability of “0” and “1” in the process of arithmetic encoding the binary sequence converted from the MVD absolute value. Value encoding efficiency can be improved.

(Additional notes)
Note that the MVD absolute value is not encoded separately for the X component absolute value and the Y component absolute value, but the sum of the X component absolute value and the Y component absolute value may be encoded. . In this case, PMVCOST = F (| MV_X−PMV_X | + | MV_Y−PMV_Y |). Further, the number of temporary PMV candidates is not limited to four, and varies depending on the value of the MVD absolute value. That is, when the sum of the absolute value of the X component and the absolute value of the Y component is 1, the number of MV temporary candidates is 8, and when the sum is 2, the number of MV temporary candidates is 16. When the sum is N, the number of MV provisional candidates is 8N.

  In each of the embodiments described above, a program for realizing each function of the moving image encoding device and the moving image decoding device is recorded on a computer-readable recording medium, and the program is recorded on the recording medium. May be read by a computer system and executed to control the video encoding device or the video decoding device. Here, the “computer system” includes an OS and hardware such as peripheral devices.

  The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” means that a program is dynamically held for a short time, like a communication line when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In this case, it is intended to include those that hold a program for a certain period of time, such as a volatile memory inside a computer system serving as a server or a client in that case. The program may be a program for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in a computer system. .

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and the design and the like within the scope of the present invention are also within the scope of the claims. include.

  The encoding apparatus of the present invention further includes a data candidate cost calculation unit that calculates the cost of each data candidate, and the flag set for each data candidate by the data candidate identification flag encoding unit is the data candidate cost. It is desirable that the flag be in accordance with the cost of the data candidate calculated by the calculation unit. For example, in the encoding apparatus, the data candidate specifying flag encoding unit includes a value of a flag specifying the data candidate so that a symbol having a higher appearance probability is included as the distance between the data candidate and the prediction data is shorter. It is desirable to encode the flag by making the symbol sequence into a simple symbol sequence and arithmetically encoding the symbol sequence.

  In the encoding device of the present invention, the difference absolute value encoding unit encodes the difference absolute value for each component when the predetermined data and the prediction data are data composed of a plurality of components, Alternatively, it is desirable to encode the sum of absolute differences for each component.

  The encoding apparatus according to the present invention includes a distance cost calculation unit that calculates a distance cost that is a value corresponding to a distance between the data candidate and the prediction data candidate, using a function that receives the data candidate and the prediction data candidate. A distance cost calculation unit capable of selecting a plurality of functions having different calculation amounts as the function, and the difference absolute value encoding unit codes an index indicating the function selected by the distance cost calculation unit. It is desirable to make it.

  In the encoding device of the present invention, it is preferable that the predetermined data is motion vector data.

  Also, the present invention provides a motion estimation unit that estimates a motion vector between an input image and a reference image that has already been decoded, a motion vector encoding unit that encodes the motion vector, and the motion vector An inter prediction unit that generates a prediction signal from a reference image, a difference calculation unit that calculates a difference between the prediction signal and the input image, a conversion unit that converts the difference, and a quantization coefficient obtained by the conversion It is preferable that the moving picture coding apparatus including the quantizing unit that uses the feature coding apparatus for the motion vector coding in the motion vector coding unit.

  The decoding device of the present invention further includes a data candidate cost calculation unit that calculates the cost of each data candidate, and the flag decoded by the data candidate identification flag decoding unit is calculated by the data candidate cost calculation unit. It is desirable that the flag corresponds to the cost of the data candidate.

  In the decoding device of the present invention, the difference absolute value decoding unit decodes the difference absolute value for each component when the predetermined data and the prediction data are data composed of a plurality of components, or It is desirable to decode the sum of absolute differences for each component.

  Further, the decoding apparatus of the present invention calculates a distance cost that calculates a distance cost that is a value corresponding to a distance between the data candidate and the prediction data candidate, using a function that receives the data candidate and the prediction data candidate. A distance cost calculation unit capable of selecting a plurality of functions having different calculation amounts as the function, and the difference absolute value decoding unit decodes an index indicating the function selected by the distance cost calculation unit It is desirable to do.

  In the decoding device of the present invention, it is desirable that the predetermined data is motion vector data.

  The present invention also provides a motion vector decoding unit that decodes a motion vector, a reference image that has already been decoded, an inter prediction unit that generates a prediction signal from the motion vector, and an inverse quantization unit that dequantizes transform coefficients. And an inverse transform unit that inversely transforms the transform coefficient to generate a difference signal, adds the predicted signal and the difference signal, and reproduces a decoded signal, wherein a motion in the motion vector decoding unit It is desirable to use the above-described decoding device for decoding vectors.

  Note that encoded data that is output by the described encoding device and includes a flag that specifies a data candidate is also included in the scope of the present invention.

DESCRIPTION OF SYMBOLS 100 ... Moving image encoding device, 101 ... Subtraction part, 102 ... Addition part, 103 ... Local decoded signal, 111 ... Motion vector encoding part, 112 ... Conversion part, 113 ... Quantization part, 115 ... Variable length encoding part , 116: Inverse quantization unit, 117: Inverse transform unit, 118 ... Frame memory, 121 ... Intra prediction unit, 123 ... Inter prediction unit, 124 ... Motion estimation unit, 125 ... Prediction mode determination unit, 200 ... Video decoding device , 201: variable length coding / decoding unit, 203 ... prediction parameter, 211 ... motion vector decoding unit, 225 ... prediction mode selection unit, 301 ... PMV candidate and number of PMV candidates, 302 ... PMV, 1111 ... MV storage unit, 1112 ... PMV candidate calculation unit, 1113 ... PMV determination unit, 1114 ... MVD calculation unit, 1115 ... MV flag determination unit, 1151 ... PMV index code , 1152 ... MVD absolute value encoding unit, 1153 ... MV index encoding unit, 2011 ... PMV index decoding unit, 2012 ... MVD absolute value decoding unit, 2013 ... MV index decoding unit, 2113 ... PMV selection unit, 2115 ... MV Selection unit, 5001 ... MV candidate calculation unit, 5002 ... MV candidate determination unit, 5003 ... MV index determination unit, 6004 ... MV extraction unit

Claims (17)

When encoding predetermined data, a difference absolute value encoding unit that generates prediction data from already encoded data and encodes a difference absolute value between the predetermined data and the prediction data;
A data candidate calculation unit that calculates data candidates from the difference absolute value and the prediction data;
A data candidate identification flag code that sets a different flag for each data candidate and encodes the flag set for each data candidate when the number of data candidates calculated by the data candidate calculation unit is 2 or more And a coding unit. The encoding device according to claim 1, comprising:
A data candidate cost calculation unit for calculating the cost of each data candidate;
Coding characterized in that the flag set for each data candidate by the data candidate specifying flag encoding unit is a flag corresponding to the cost of the data candidate calculated by the data candidate cost calculating unit. apparatus. The encoding device according to claim 1 or 2,
The difference absolute value encoding unit encodes the difference absolute value for each component or the difference absolute value for each component when the predetermined data and the prediction data are data composed of a plurality of components. An encoding device that encodes the sum of the two. When encoding predetermined data, a prediction data determination unit that generates a plurality of prediction data candidates from already encoded data and selects prediction data used for encoding from the prediction data candidates;
A difference absolute value encoding unit that encodes a difference absolute value between the predetermined data and the prediction data;
A data candidate calculator that calculates a data candidate from the absolute difference value and the prediction data;
A data candidate determination unit that determines the validity of the data candidate calculated from the difference absolute value and the prediction data from the prediction data and the prediction data candidate;
An encoding apparatus comprising: a data candidate specifying flag encoding unit that encodes a flag for specifying the data candidate when the number of data candidates validated by the data candidate determination unit is 2 or more . The encoding device according to claim 4,
A distance cost calculation unit that calculates a distance cost that is a value corresponding to a distance between the data candidate and the prediction data candidate by a function having the data candidate and the prediction data candidate as inputs, and calculates the function as the function A distance cost calculation unit capable of selecting a plurality of functions having different amounts is further provided,
The difference absolute value encoding unit encodes an index indicating the function selected by the distance cost calculation unit. When encoding predetermined data, a prediction data determination unit that generates a plurality of prediction data candidates from already encoded data and selects prediction data used for encoding from the prediction data candidates;
A difference absolute value encoding unit that encodes the difference absolute value by arithmetically encoding the symbol string while using the difference absolute value between the predetermined data and the prediction data as a symbol string; and
The difference absolute value encoding unit changes an appearance probability of each symbol used for arithmetic encoding of the symbol sequence according to the prediction data selected by the prediction data determination unit. .   The encoding apparatus according to claim 1, wherein the predetermined data is motion vector data.   A motion estimation unit that estimates a motion vector between an input image and a reference image that has already been decoded, a motion vector encoding unit that encodes the motion vector, and a prediction signal from the reference image using the motion vector An inter prediction unit to be generated; a difference calculation unit that calculates a difference between the prediction signal and the input image; a conversion unit that converts the difference; and a quantization unit that quantizes a conversion coefficient obtained by the conversion. 8. A moving picture coding apparatus according to claim 7, wherein the coding apparatus according to claim 7 is used for coding a motion vector in the motion vector coding unit. A difference absolute value decoding unit for decoding the encoded difference absolute value when decoding predetermined data; and
A data candidate calculation unit that generates prediction data from already encoded data, and calculates a data candidate from the difference absolute value and the prediction data;
A data candidate identification flag decoding unit that decodes a flag in which a different value is set for each data candidate for each data candidate when the number of data candidates generated by the data candidate calculation unit is 2 or more; A decoding device comprising: The decoding device according to claim 9, wherein
A data candidate cost calculation unit for calculating the cost of each data candidate;
The decoding device characterized in that the flag decoded by the data candidate identification flag decoding unit is a flag corresponding to the cost of the data candidate calculated by the data candidate cost calculation unit. The decoding device according to claim 9 or 10,
The difference absolute value decoding unit decodes the difference absolute value for each component or the sum of the difference absolute values for each component when the predetermined data and the prediction data are data composed of a plurality of components. The decoding apparatus characterized by decoding. When decoding predetermined data, the encoded absolute difference value is decoded, a plurality of prediction data candidates are generated from the already encoded data, and the prediction data used for encoding is selected from the prediction data candidates A prediction data determination unit to
A data candidate determination unit that determines the validity of the data candidate calculated from the difference absolute value and the prediction data from the prediction data and the prediction data candidate;
A decoding apparatus comprising: a data candidate specifying flag decoding unit configured to decode a flag for specifying the data candidate when the number of data candidates validated by the data candidate determination unit is two or more. The decoding device according to claim 12, comprising:
A distance cost calculation unit that calculates a distance cost that is a value corresponding to a distance between the data candidate and the prediction data candidate by a function having the data candidate and the prediction data candidate as inputs, and calculates the function as the function A distance cost calculation unit capable of selecting a plurality of functions having different amounts is further provided,
The absolute difference decoding unit decodes an index indicating the function selected by the distance cost calculation unit. When decoding predetermined data, a prediction data determination unit that generates a plurality of prediction data candidates from already encoded data and selects prediction data used for decoding from the prediction data candidates;
A difference absolute value decoding unit that decodes the difference absolute value by arithmetically decoding a symbol string that is a difference absolute value between the predetermined data and the prediction data; and
The difference absolute value decoding unit changes the appearance probability of each symbol used for decoding the symbol string according to the prediction data selected by the prediction data determination unit.   15. The decoding apparatus according to claim 9, wherein the predetermined data is motion vector data.   A motion vector decoding unit that decodes a motion vector, an already decoded reference image, an inter prediction unit that generates a prediction signal from the motion vector, an inverse quantization unit that inversely quantizes the transform coefficient, and an inverse transform coefficient A video decoding device comprising an inverse conversion unit that converts and generates a difference signal, adds the prediction signal and the difference signal, and reproduces a decoded signal, wherein the motion vector decoding unit decodes the motion vector. 15. A moving picture decoding apparatus using the decoding apparatus according to 15.   The encoded data output from the encoding device according to any one of claims 1 to 7, wherein the encoded data includes the flag.

Images