WO2011062082A1 - Video encoder and video decoder - Google Patents

Abstract

Provided is a video encoder which improves the prediction accuracy of prediction vectors when encoding is performed using motion compensated prediction. The prediction accuracy of prediction vectors is improved, the amount of code for differential motion vectors and prediction residuals is reduced, and encoding efficiency is improved by calculating the degree of similarity of the motion vectors of blocks in the vicinity of a target block and setting the motion vectors having a high degree of similarity as prediction vectors.

Description

Moving picture encoding apparatus and moving picture decoding apparatus

The present invention relates to a moving picture coding apparatus and a moving picture decoding apparatus that perform coding using motion compensated prediction.

In a moving image encoding apparatus, motion compensated prediction is used as a technique for generating a predicted image of an image (input image) to be encoded.
In motion compensated prediction, a predicted image of an input image is generated using a motion vector and a reference image obtained by motion estimation between an input image and a locally decoded image (reference image) decoded in the moving image encoding device. .
Thereafter, the data obtained by performing frequency conversion and quantization on the residual image between the input image and the predicted image, and the difference motion vector between the estimated motion vector and the predicted vector are encoded.

On the other hand, the moving image decoding apparatus generates a new predicted image using the motion vector restored by the difference vector and the predicted vector and the reference image, and decodes the predicted image and the residual image to the original image.

<Conventional technology>
In Non-Patent Document 1, an input moving image is divided into a plurality of partitions for each frame, and a target partition among a plurality of partitions adjacent to a partition to be encoded (hereinafter referred to as a “target partition”). Predictive Motion Vector (PMV) is estimated using the median (median) of motion vectors of adjacent partitions on the upper, left, and upper right, and an input image is obtained using the obtained prediction vector and reference image. A predicted image is generated.

Non-Patent Document 2 describes a technique called MV Competition that improves coding efficiency by adaptively selecting an optimal prediction vector for a motion vector of a target partition from a plurality of prediction vector candidates. .

The prediction vector candidates include the median value of the motion vectors of the partitions located in the vicinity of the target partition, the motion vectors of the partitions at the same position as the target partition in the encoded frame, and the like. An optimal prediction vector is selected in consideration of rate distortion characteristics (coding cost).

ITU-T Recommendation H. H.264 (11/07) ITU-T T09-SG16-VCEG-AC06, "Competition-Based-Scheme-for Motion-Vector-Selection-and Coding"

Although Non-Patent Document 1 derives a prediction vector of a target partition using the median value of motion vectors of partitions adjacent to the target partition, this method cannot always obtain an optimal prediction vector.

For example, as shown in FIG. 10, when the target partition and the partition adjacent to the left of the target partition are inside the object, and the other partitions adjacent to the top and right are outside the object (background), the motion vector The median becomes the motion vector on the background side.
However, since the target partition is inside the object, it should be approximate to the motion vector of the partition adjacent to the left.

Therefore, if a prediction vector obtained from the median of motion vectors of partitions adjacent to the target partition is used, a difference motion vector that is the difference between the motion vector estimated for the target partition and this prediction vector, or the difference between the input image and the prediction image As a result, the residual image becomes larger and the coding efficiency is lowered.

In Non-Patent Document 2, a prediction vector having a low encoding cost is selected for each partition from a plurality of prediction vector candidates.
When decoding the encoded data encoded by this method, in order to know the derivation method of the selected prediction vector, it is necessary to encode the derivation method distinction information as a flag.
As in Non-Patent Document 2, when the number of prediction vector candidates is large, the data length representing this distinction information becomes large, so that the code amount of the flag becomes large, and the coding efficiency is lowered.

The present invention has been made in consideration of the above situation, and provides a moving picture coding apparatus and a moving picture decoding apparatus that improve the prediction accuracy of a prediction vector when coding using motion compensated prediction. With the goal.
Another object is to make it possible to reduce the code amount of information (flag) that distinguishes the method of deriving the prediction vector.

In order to solve the above-described problems, the moving picture encoding apparatus and moving picture decoding apparatus of the present invention are configured as follows.

A moving image encoding device is a moving image encoding device that generates a differential motion vector from a motion vector and a prediction vector of a target block and encodes a moving image, and corresponds to one or more directions in a temporal direction or a spatial direction. One or a plurality of sets of motion vectors of two or more blocks are extracted, and a similarity calculation unit that calculates a similarity corresponding to the direction from the set of motion vectors, and the similarity is maximized A prediction vector deriving unit for deriving a motion vector calculated by a method corresponding to a direction as one of the prediction vector of the target block or one of the prediction vectors; a difference from a difference between the prediction vector and the motion vector of the target block A motion vector is generated, and encoded data is created using the difference motion vector.

The motion vector set uses a motion vector according to any of the following.

(1) The similarity calculation unit exists as a set of motion vectors for calculating the similarity corresponding to the direction in the spatial direction, in at least two spatial directions among the neighboring blocks of the target block on the processing target image. A set of two motion vectors for each spatial direction, which is composed of a motion vector of an adjacent neighboring block and a motion vector of a neighboring block existing in the direction away from the target block, is used.

In the case of using this motion vector, specifically, the moving image encoding apparatus performs motion of adjacent neighboring blocks existing in at least two or more spatial directions among neighboring blocks of the target block on the processing target image. A set of two motion vectors for each spatial direction consisting of a vector and a motion vector of a neighboring block existing away from the target block in the direction is extracted, and the distance between the motion vectors calculated from the set of the motion vectors is calculated. A similarity calculation unit that calculates a similarity based on the prediction vector derivation unit that derives a motion vector of a block having the maximum similarity as a prediction vector of the target block, and the prediction vector and the A differential motion vector is generated from the difference from the motion vector of the target block, and encoded data is generated using the differential motion vector. Configured to create.

(2) The similarity calculation unit is present at the same position as the processing target block on the already processed first image as a set of motion vectors for calculating the similarity corresponding to the direction in the time direction, or A motion vector of a block existing at a position around the processing target block and a second image that has already been processed different from the first image, are present at the same position as the processing target block, or the processing A pair with a motion vector of a block existing around the target block is used.

(3) The similarity calculation unit, as a set of motion vectors for calculating the similarity corresponding to the direction in the time direction, on the encoding target image, on the block adjacent to the processing target block, on the already processed image Thus, one or more pairs of motion vectors of blocks existing at the same position as the block adjacent to the processing target block or existing around the block adjacent to the processing target block are used.

The moving image encoding device includes a first prediction vector deriving unit that calculates a first prediction vector from a motion vector of a neighboring block adjacent to the target block on the processing target image, and the prediction vector deriving unit includes the similarity When the degree of similarity calculated by the degree calculation unit is equal to or less than a predetermined threshold, the first prediction vector may be derived as a prediction vector.

Alternatively, in the moving image encoding device, a first prediction vector deriving unit that calculates a first prediction vector from a motion vector of a neighboring block that is adjacent to the target block on the processing target image, and the similarity calculation The motion vector having the maximum similarity calculated by the unit and the first prediction vector derived by the first prediction vector deriving unit are candidates, and the coding costs based on both prediction vectors are compared, and the one with the smaller coding cost is selected. An encoding unit that sets and encodes a flag indicating which of the candidates is the prediction vector as the prediction vector of the target block may be provided.

Alternatively, in the moving image encoding device, a first prediction vector deriving unit that calculates a first prediction vector from a motion vector of a neighboring block that is adjacent to the target block on the processing target image; A second prediction vector derivation unit that calculates a second prediction vector different from the first prediction vector from motion vectors of neighboring blocks that are adjacent to the target block, and the processing target block on the already processed image A time direction motion vector calculation unit that calculates a time direction motion vector from a motion vector of a block located at the same position as the block adjacent to the processing target block or around the block adjacent to the processing target block, and a similarity calculation unit, A class of motion vectors in the time direction using a set of one or more motion vectors corresponding to the time direction. If the degree of similarity in the time direction is equal to or greater than a predetermined value when the degree is obtained, the first prediction vector and the time direction motion vector are taken as prediction vector candidates. The prediction vector candidate selection unit that uses the prediction vector and the second prediction vector as a prediction vector candidate, and the encoding cost based on both prediction vectors are compared, and the one with the lower encoding cost is set as the prediction vector of the target block, An encoding unit that sets and encodes a flag indicating which prediction vector is the prediction vector candidate may be provided.

The moving picture decoding device generates a prediction vector of a target block from a motion vector of a block obtained by decoding encoded data obtained by encoding a moving picture to be processed using motion compensated prediction, the difference motion vector of the target block, and the In a moving picture decoding apparatus that restores a motion vector from a prediction vector and decodes the encoded data, one set of motion vectors of two or more blocks corresponding to one or more directions in the time direction or the spatial direction Alternatively, a plurality of extractions, a similarity calculation unit that calculates the degree of similarity corresponding to the direction from the set of motion vectors, and a motion vector calculated by a method corresponding to the direction where the similarity is maximized, A prediction vector deriving unit that derives the prediction vector of the target block or one of the prediction vectors.

The motion vector set uses a motion vector according to any of the following.

(1) The similarity calculation unit exists as a set of motion vectors for calculating the similarity corresponding to the direction in the spatial direction, in at least two spatial directions among the neighboring blocks of the target block on the processing target image. A set of two motion vectors for each spatial direction, which is composed of a motion vector of an adjacent neighboring block and a motion vector of a neighboring block existing in the direction away from the target block, is used.

(2) The similarity calculation unit is present at the same position as the processing target block on the already processed first image as a set of motion vectors for calculating the similarity corresponding to the direction in the time direction, or A motion vector of a block existing at a position around the processing target block and a second image that has already been processed different from the first image, are present at the same position as the processing target block, or the processing A pair with a motion vector of a block existing around the target block is used.

(3) The similarity calculation unit, as a set of motion vectors for calculating the similarity corresponding to the direction in the time direction, on the encoding target image, on the block adjacent to the processing target block, on the already processed image Thus, one or more pairs of motion vectors of blocks existing at the same position as the block adjacent to the processing target block or existing around the block adjacent to the processing target block are used.

In the above moving image decoding apparatus, the motion vectors of adjacent neighboring blocks existing in at least two spatial directions among the neighboring blocks of the target block on the processing target image, and existing in the direction away from the target block A similarity calculation unit that extracts a set of motion vectors for each spatial direction composed of motion vectors of neighboring blocks and calculates a similarity based on a distance between the motion vectors calculated from the set of motion vectors; and the similarity A prediction vector deriving unit for deriving the motion vector of the block having the maximum value as a prediction vector of the target block, the prediction vector, and a motion vector restoring unit for restoring a motion vector from the differential motion vector of the target block; You may comprise so that it may be provided.

Alternatively, in the above moving image decoding apparatus, a first prediction vector deriving unit that calculates a first prediction vector from a motion vector of a neighboring block that is adjacent to the target block on the processing target image, and on the processing target image A second prediction vector deriving unit that calculates a second prediction vector different from the first prediction vector from motion vectors of neighboring blocks existing adjacent to the target block; and on the already processed image, the processing target block A time direction motion vector calculation unit that calculates a time direction motion vector from a motion vector of a block located at the same position as an adjacent block or around the block adjacent to the processing target block, and a similarity calculation unit Similarity of motion vectors in the time direction using a set of one or more motion vectors corresponding to the direction If the similarity in the time direction is greater than or equal to a predetermined value, the first prediction vector and the time direction motion vector are used as prediction vector candidates; otherwise, the first prediction vector And a prediction vector candidate selection unit that uses the second prediction vector as a prediction vector candidate, a flag selection unit that outputs a PMV flag for the target block decoded by the variable-length code decoding unit, and the prediction vector candidate selection unit The prediction vector candidate may be configured to include a switching unit that switches according to the similarity in the time direction output from the similarity calculation unit and the PMV flag output from the flag selection unit. .

According to the present invention, since a prediction vector having a high similarity is selected, the prediction accuracy of the prediction vector is improved.
In addition, since the selection of the prediction vector is narrowed down by the similarity, it is possible to add a substantial number of prediction vector candidates without increasing the code amount of information (flag) for distinguishing the method for deriving the prediction vector. Therefore, it is possible to select a prediction vector with high prediction accuracy.

Thereby, since a prediction vector with high accuracy can be selected, the difference motion vector and the prediction residual are reduced, and the code amount of the flag is reduced. As a result, the data amount of the encoded bit stream is reduced, Encoding efficiency is improved.

Especially, when a part of the partition adjacent to the target partition belongs to the same object, there is an effect of reducing the difference motion vector and its code amount.

It is a block diagram which shows the function structure of the moving image encoder which concerns on this invention. It is a block diagram which shows the function structure of the motion vector redundancy reduction part which concerns on Embodiment 1. FIG. It is a figure which shows the relationship between the object partition when a partition size is the same, and a neighborhood partition. It is a figure which shows the relationship between the object partition in case partition sizes differ, and a near partition. It is a flowchart explaining operation | movement of a similarity degree PMV derivation | leading-out part. It is the encoded bit stream output in Embodiment 1. It is a block diagram which shows the function structure of the moving image decoding apparatus which concerns on this invention. 3 is a block diagram illustrating a functional configuration of a motion vector restoration unit according to the first embodiment. FIG. It is a figure for demonstrating the effect of similarity PMV. It is another figure for demonstrating the effect of similarity PMV. It is a block diagram which shows the function structure of the motion vector redundancy reduction part which concerns on Embodiment 2. FIG. It is a figure explaining the selection method of a reference partition when the size of an object partition and a collocated partition is the same. It is a figure explaining the selection method of a reference partition when the size of an object partition and a collocated partition differs. 10 is a block diagram illustrating a functional configuration of a motion vector restoration unit according to Embodiment 2. FIG. It is a block diagram which shows the function structure of the motion vector redundancy reduction part which concerns on Embodiment 3. FIG. H. 2 is a diagram illustrating a method for calculating a prediction vector standardized by H.264. It is the encoded bit stream output in Embodiment 3. It is a block diagram which shows the function structure of the motion vector decompression | restoration part which concerns on Embodiment 3. FIG. It is a flowchart explaining operation | movement of a PMV determination part. It is a block diagram which shows the function structure of the motion vector redundancy reduction part which concerns on the modification of the moving image encoder of this invention. It is a block diagram which shows the function structure of the motion vector decompression | restoration part which concerns on the modification of the moving image decoding apparatus of this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In addition, the same code | symbol is attached | subjected about the same component and the overlapping description is abbreviate | omitted.

<< Embodiment 1 >>

<Functional Configuration of Video Encoding Device>
First, the functional configuration of the video encoding apparatus according to the first embodiment will be described.
FIG. 1 is a block diagram illustrating a functional configuration of a video encoding device according to the first embodiment. In FIG. 1, the video encoding device 1 includes an inter-predicted image generation unit 13, an inverse quantization / inverse conversion unit. 14, transform / quantization unit 15, prediction scheme control unit 16, variable length coding unit 17, motion vector estimation unit 18, intra prediction image generation unit 20, addition unit 21, subtraction unit 22, motion vector redundancy reduction unit 24 The buffer memory 12 is configured.

The transform / quantization unit 15 calculates a difference value (prediction residual image) between an input image divided into block images including a plurality of adjacent pixels and a prediction image output from a prediction method control unit 16 described later. The subtraction unit 22 calculates and performs quantized prediction residual data by performing DCT (Discrete) Cosine Transform) conversion and quantization on the prediction residual image. Here, the quantization is an operation for associating the frequency component with an integer value. Hereinafter, a block to be processed is referred to as a “target block”. The size of the block is, for example, 16 × 16 pixels, but the present invention is not limited by the specific size of the block. The quantized prediction residual data is input to the variable length encoding unit 17 on the one hand and to the inverse quantization / inverse transform unit 14 on the other hand.

The inverse quantization / inverse transform unit 14 calculates a prediction residual image by performing inverse quantization and inverse DCT transform on the quantized prediction residual data. A locally decoded image obtained by adding the prediction residual image and the prediction image output from the prediction method control unit 16 by the adding unit 21 is stored in the buffer memory 12.

The intra-predicted image generation unit 20 generates an intra-predicted image from a locally decoded image in the same image as the target block stored in the buffer memory 12.

The motion vector estimation unit 18 divides the target block into one or a plurality of partitions, and sequentially assigns a motion vector to each partition. Specifically, based on the input image and the reference image in which the entire frame has already been decoded and stored in the buffer memory 12, a partition to be processed (hereinafter referred to as “target partition”) among the plurality of partitions. A motion vector is calculated. Also, the relative position information of the reference image (hereinafter referred to as the reference image relative position) with respect to the frame to which each partition belongs is calculated. The calculated motion vector is output to the inter predicted image generation unit 13 and the motion vector redundancy reduction unit 24 and is stored in the buffer memory 12.

The inter prediction image generation unit 13 generates an inter prediction image by motion compensation from the above-described motion vector and the reference image of the buffer memory 12.

The prediction scheme control unit 16 compares the intra prediction image that is the output of the intra prediction image generation unit 20 with the inter prediction image that is the output of the inter prediction image generation unit 13 and the input image, and the prediction mode of the target block is determined. It is determined whether the prediction is intra prediction or inter prediction, and a prediction image corresponding to the prediction mode is output. Here, the prediction mode and the motion vector are also sequentially stored in the buffer memory 12.

The motion vector redundancy reduction unit 24 determines a prediction vector for the target partition, and generates a differential motion vector that is a residual with the motion vector estimated by the motion vector estimation unit 18.

The variable length coding unit 17 performs variable length coding on the quantized prediction residual data, the prediction mode, and the differential motion vector in an output form as described later, and outputs the result to the outside.

<Functional Configuration of Motion Vector Redundancy Reduction Unit 24>
Next, the functional configuration of the motion vector redundancy reducing unit 24 according to the first embodiment will be described in detail.
FIG. 2 is a block diagram illustrating a functional configuration of the motion vector redundancy reduction unit 24 according to the first embodiment. The motion vector redundancy reduction unit 24 is a motion of each encoded partition stored in the buffer memory 12. Calculates multiple similarities indicating the correlation of motion vectors from the vectors, and uses the similarities to determine the prediction vector of the target partition from the motion vectors of the partitions located in the vicinity of the target partition (hereinafter referred to as neighboring partitions) And calculating the difference between the motion vector of the target partition output from the similarity PMV derivation unit 111 and the motion vector estimation unit 18 to be output and the prediction vector of the target partition output from the similarity PMV derivation unit 111; It is comprised from the subtraction part 191 output as a difference motion vector.

<Functional Configuration of Similarity PMV Deriving Unit 111>
As shown in FIG. 2, the similarity PMV derivation unit 111 includes a spatial direction similarity calculation unit 120, a similarity PMV determination unit 121, and a similarity PMV candidate extraction unit 122.

The spatial direction similarity calculation unit 120 calculates and outputs similarities regarding a plurality of spatial directions by using the encoded motion vectors stored in the buffer memory 12 for each target partition.

The similarity PMV candidate extraction unit 122 extracts a motion vector as a prediction vector candidate from the encoded motion vector stored in the buffer memory 12 and outputs the motion vector.

The similarity PMV determination unit 121 selects a prediction vector (similarity) having the maximum similarity of the similarity calculated by the spatial direction similarity calculation unit 120 from the prediction vector candidates extracted by the similarity PMV candidate extraction unit 122. PMV) is determined and output.
Hereinafter, the spatial direction similarity calculation unit 120, the similarity PMV determination unit 121, and the similarity PMV candidate extraction unit 122 will be described in detail.

<Functional Configuration of Spatial Direction Similarity Calculation Unit 120>
The spatial direction similarity calculation unit 120 includes a spatial direction motion vector extraction unit 130 and a similarity calculation unit 131, as shown in FIG.

First, the spatial direction motion vector extraction unit 130 will be described.
In motion compensation prediction, the target block is further divided into units called partitions, and processing is performed. Based on the motion vector of each partition of the target block, a corresponding region on the reference image stored in the buffer memory 12 is determined. By referencing, an inter prediction image is generated.
In the above-mentioned standard of Non-Patent Document 1, 16 × 16, 16 × 8, 8 × 16, 8 × 8, 8 × 4, 4 × 8, and 4 × 4 pixel partition sizes (hereinafter referred to as partition sizes). ) Is available.

In the spatial direction motion vector extraction unit 130, the extraction method differs depending on the size relationship between the target partition and the neighboring partition.

(1) Extraction method of motion vector of reference partition when size of target partition and neighboring partition is the same:
A method of extracting a motion vector of a reference partition when the size of the target partition and the neighboring partition are the same, for example, when one block corresponds to one partition will be described.

FIG. 3 is a diagram illustrating the relationship between the target partition and neighboring partitions.
In the spatial direction motion vector extraction unit 130, a set of two partitions (reference partition 1 and reference partition 2) located in a specific direction (left, upper, upper right, upper left) with respect to the target partition (hatched solid line rectangular region). Extract motion vectors. These reference partitions are determined as follows, for example. Note that the extraction of motion vectors in the spatial direction in the present invention is not limited by the method described below.

A partition adjacent to the target partition is defined as a reference partition 1 (gray-dashed rectangular area).
The reference partition 1 is composed of four partitions with the partition A as the left direction, the partition B as the upper direction, the partition C as the upper right direction, and the partition D as the upper left direction based on the target partition.

Further, the left of partition A is partition E, the top of partition B is partition F, the top right of partition C is partition G, the top left of partition D is partition H, and these are referred to as reference partition 2 (colorless dashed rectangular area).

The reference partition 2 may be, for example, the partition E ′ in the left direction, the partition F ′ in the upward direction, the partition G ′ in the upper right direction, and the partition H ′ in the upper left direction. .

(2) Extraction method of motion vector of reference partition when size of target partition and neighboring partition are different:
A method of extracting the motion vector of the reference partition when the target partition and the neighboring partition are different in size will be described.

The reference partition 1 is selected regardless of the position of the target block inside or outside the target block. That is, partition motion vectors are extracted across block boundaries.
Further, when there are a plurality of partitions adjacent in the left direction and the upward direction, selection is made based on the following criteria.

Left direction: the topmost partition among neighboring partitions Upward direction: the leftmost partition among neighboring partitions

The reference partition 2 is selected in the same manner as the reference partition 1 on the basis of the reference partition 1.
FIG. 4 shows an example of a reference partition selected based on the above processing. The hatched solid line rectangular area indicates the target partition, the gray broken line rectangular area indicates the reference partition 1, and the colorless broken line rectangular area indicates the reference partition 2. ing.

Next, the similarity calculation unit 131 will be described.
The similarity calculation unit 131 calculates a distance between motion vectors from the motion vectors of two reference partitions in the same direction, and uses the distance between the motion vectors to use four neighboring partitions of the target partition, that is, partition A ( Left), partition B (upper), partition C (upper right), and partition D (upper left) are calculated in four types of similarities, and these four types of similarities are output.

The two motion vectors are reference partition 1 and reference partition 2, that is, motion vectors of partitions A and E, motion vectors of partitions B and F, motion vectors of partitions C and G, and motion vectors of partitions D and H.

The distance dist between motion vectors is expressed by the following equation (1) using the horizontal component vx1 and vertical component vy1 of the motion vector v1 of the reference partition 1, and the horizontal component vx2 and vertical component vy2 of the motion vector v2 of the reference partition 2. Can be calculated.

Alternatively, the following equation (2) may be used to reduce the processing amount.

Here, | x | is a symbol indicating the absolute value of x.

The similarity mvdif calculated from the motion vectors of the reference partitions 1 and 2 can be expressed by the reciprocal of the distance dist between the motion vectors as shown in the following equation (3).

The similarity is not limited to the reciprocal, and may be expressed by the following equation (4).

Also, the distance may be used as it is, or the following expression (5) weighted on the distance may be used as the similarity. In this case, the greater the value, the lower the similarity.

Here, w is a weight for the similarity and may be a constant value (for example, w = 1) or may be changed for each direction of the similarity. The weight w may be stored and encoded in a header such as a sequence, a frame, or a slice.

Since the similarity mvdif is calculated based on the distance of the motion vector, the position where the partition exists is also an important factor. That is, the positions of the reference partitions (partitions C and D) in the upper left direction and the upper right direction exist at positions that are √2 times larger than the positions in the reference partitions (partitions A and B) in the left direction and the upper direction. As correction for that, for example, when obtaining the similarity in the upper left and upper right directions, w = 1 / √2, and for obtaining the similarity in the left and upper directions, w = 1 may be set. Good.
Conversely, to reduce the similarity in the upper left and upper right directions, that is, to exclude the motion vectors of the upper left and upper right partitions that are farther away than the left and upper directions, the similarity weights in the upper left and upper right directions are set. It may be large (w = √2).

Thus, the selection of a motion vector in a predetermined direction can be controlled by using the weight w in the calculation formula for obtaining the similarity mvdif. For example, if the weight w is increased, the degree of similarity is reduced, and it becomes difficult to select a motion vector in this direction, and if the weight w is reduced, the degree of similarity is increased, so that a motion vector in this direction is easily selected.

Conversely, an index indicating the degree of dissimilarity may be used for the above-described similarity, and a smaller index may be interpreted as a greater similarity.
For example, for simplification, the distance dist between vectors can be used as an index representing the degree of dissimilarity.

Further, when the reference partition refers to outside the image, or when there is no motion vector due to intra coding, the distance dist between the vectors cannot be calculated using the equations (1) and (2). Therefore, the similarity mvdif cannot be calculated. In this case, it is assumed that the correlation between the motion vectors is low, and the similarity mvdif is set to 0.

Further, as the similarity mvdif, an angle between motion vectors may be used as in the following equation (6).

Here, | A | is a symbol indicating the size of vector A, and <A · B> is an expression representing the inner product of vector A and vector B.

Assuming that the distance dist between the motion vectors is the angle θ formed by the two vectors, the similarity mvdif = cos θ, so that cos θ decreases monotonously when θ is between 0 degrees and 180 degrees. That is, when the motion vector is directed in the same direction (θ = 0 degree), mvdif takes the maximum value, and when the motion vector is directed in the reverse direction (θ = 180 degrees), mvdif takes the minimum value, so that cos θ is It can be used as similarity.

Hereinafter, the similarity mvdif will be described as being calculated using Equation (3).

Next, the similarity PMV candidate extraction unit 122 will be described.
The similarity degree PMV candidate extraction unit 122 extracts, for each target partition, a motion vector corresponding to the reference partition 1 described above from the encoded motion vectors stored in the buffer memory 12, and sets it as a prediction vector candidate. . That is, in the first embodiment, motion vectors of up to four types of partitions adjacent to the left, top, top right, and top left positions of the target partition are extracted.

Next, the similarity PMV determination unit 121 will be described.
The similarity PMV determination unit 121 inputs, for each target partition, the similarity calculated by the spatial direction similarity calculation unit 120 and the prediction vector candidate extracted by the similarity PMV candidate extraction unit 122. The similarities are compared, the direction having the highest similarity is specified, and the prediction vector candidate corresponding to the direction is determined and output as the prediction vector.
Hereinafter, the prediction vector calculated in this way is referred to as similarity PMV.

In the above similarity comparison, when there are a plurality of maximum similarities, the prediction vector is determined in the priority order of the left direction, the upward direction, the upper right direction, and the upper left direction.

For example, when the direction with the highest similarity is the upward direction or the upper left direction, the motion vector of the upward partition (partition B in FIG. 3) having a high priority is used as the prediction vector.

Here, the similarity PMV determination unit 121 employs the motion vector of the partition in the direction of the maximum similarity as a prediction vector. However, when the similarity is small in all directions, the reliability of the similarity itself is low. The following exception handling is performed.

When a threshold value TH is a preset value or a value calculated from an average constant multiple of the similarity of partitions already encoded, and the maximum similarity is smaller than the threshold TH, motion vectors of a plurality of neighboring partitions of the target partition With reference to the group, the following first prediction vector is used.
The threshold TH may be encoded as header information in units of sequences, frames, or slices.

(First prediction vector)
A prediction vector calculated with reference to a plurality of neighboring motion vector groups of the target partition is hereinafter referred to as a first prediction vector.
One of the methods for calculating the first prediction vector with reference to the motion vector group includes a left partition adjacent to the left side of the target partition, an upper partition adjacent to the upper side of the target partition, and a right side of the upper partition. There is a median value for each motion vector in the upper right partition.

Other methods for calculating the first prediction vector may be, for example, any of the following, but are not limited to these methods.

The average value of the motion vectors of the left partition, upper partition, and upper right partition of the target partition.
-The weighted average value of each motion vector of the left partition, the upper partition, and the upper right partition of the target partition.
The maximum value of each motion vector of the left partition, upper partition, and upper right partition of the target partition.
-The minimum value of each motion vector of the left partition, upper partition, and upper right partition of the target partition.
The average value of the motion vectors of the left partition and the upper partition of the target partition, the average value of the motion vectors of the upper partition and the upper right partition of the target partition, and the motion vectors of the left partition and upper right partition of the target partition The median of the average value of.
Among the motion vectors of the left partition, upper partition, and upper right partition of the target partition, the motion vector having the largest difference from the median value of the left partition, upper partition, and upper right partition of the target partition.

<Operation of Similarity PMV Deriving Unit 111>
The operation of the similarity PMV deriving unit 111 will be described with reference to the flowchart of FIG.

In order to calculate four types of similarity in the order of partition A, E (left), partition B, F (upper), partition C, G (upper right), partition D, H (upper left), step S1 to step S6 The process up to is repeated four times (step S1).

The motion vectors of reference partition 1 and reference partition 2 are extracted (step S2).
When the motion vectors of the reference partition 1 and the reference partition 2 can be extracted, the flag is set to zero. When at least one motion vector of the reference partition 1 and the reference partition 2 cannot be extracted, the flag is set to 1.

If the flag is zero, the similarity is calculated using equation (3), and if the flag is 1, the similarity is set to zero (step S3).

It is determined whether or not the calculated similarity is the maximum among the processes processed so far. If it is the maximum (YES in step S4), the current maximum similarity is updated (step S5).
On the other hand, if the calculated similarity is not the highest among the processes processed so far (NO in step S4), the process of step S6 is performed.

If the similarity calculation processing for partitions D and H (upper left) has been completed, the process returns to step S7, otherwise returns to step S1 to perform the process in the next direction (step S6).

If the calculated maximum similarity is greater than or equal to the threshold (YES in step S7), a motion vector corresponding to the direction of the maximum similarity is extracted and output as the similarity PMV (step S8).

On the other hand, if the calculated maximum similarity is equal to or smaller than the threshold (NO in step S7), the first prediction vector of the target partition is calculated and output (step S9). For example, motion vectors of the partition A (left), the partition B (upper), and the partition C (upper right) are extracted, and their median values are used as prediction vectors.

<Configuration of Encoded Bitstream Generated by Variable Length Encoding Unit of First Embodiment>
The encoded data generated by the variable-length encoding unit 17 is a bit stream generated by quantizing prediction residual data, a prediction mode, and a differential motion vector in a format standardized in the aforementioned Non-Patent Document 1. is there.

The bit stream in FIG. 6 shows a portion corresponding to the target block. Following the head block mode information, N pieces of index information Idxi (0 ≦ i <N) in the target block are numbered. After that, motion vector information MVi (0 ≦ i <N) is arranged.

The block mode information includes the prediction mode of the target block, partition division information (in the case of inter coding), and the like.
The index information Idxi includes information on the relative position of the reference image referenced by each partition, and is used for motion compensation.
The motion vector information MVi includes differential motion vector information of each partition in the block.

<Functional configuration of video decoding device>
Next, a functional configuration of the video decoding device according to the first embodiment will be described.
FIG. 7 is a block diagram illustrating a functional configuration of the moving picture decoding apparatus 2 according to the first embodiment. In the figure, the moving picture decoding apparatus 2 includes a variable length code decoding unit 10, a motion vector restoration unit 23, and a buffer memory. 12, an inter prediction image generation unit 13, an intra prediction image generation unit 20, a prediction method determination unit 25, an inverse quantization / inverse conversion unit 14, and an addition unit 21.

The variable length code decoding unit 10 performs variable length decoding on the input encoded data, stores the prediction mode in the buffer memory 12, and outputs the prediction mode to the prediction method determination unit 25, and outputs the difference motion vector to the motion vector restoration unit 23. The quantized prediction residual data is output to the inverse quantization / inverse transform unit 14.

The motion vector restoration unit 23 decodes the motion vector of the target partition using the prediction vector calculated from the difference motion vector and the motion vector stored in the buffer memory 12 and stores the decoded motion vector in the buffer memory 12. Output to the generator 13.

The buffer memory 12 stores local decoded images, motion vectors, and prediction modes for all decoded blocks (partitions) for the processing target frame.

The inter prediction image generation unit 13 generates a prediction image by motion compensation from the motion vector and the decoded image stored in the buffer memory 12, and outputs the prediction image to the prediction method determination unit 25.

The intra-predicted image generation unit 20 generates an intra-predicted image from the locally decoded image in the same image as the target block stored in the buffer memory 12 and outputs the intra-predicted image to the prediction method determining unit 25.

The prediction method determination unit 25 determines whether the intra prediction image that is the output of the intra prediction image generation unit 20 and the inter prediction image that is the output of the inter prediction image generation unit 13 according to the prediction mode decoded by the variable length code decoding unit 10. Either one is output to the adding unit 21 as a predicted image.

The inverse quantization / inverse transform unit 14 decodes the prediction residual image by performing inverse quantization and inverse DCT transform on the quantized prediction residual data, and outputs the decoded residual image to the addition unit 21.

The addition unit 21 generates a decoded image by adding the prediction residual image that is the output of the inverse quantization / inverse conversion unit 14 and the prediction image that is the output of the prediction method determination unit 25, and stores the decoded image in the buffer memory 12. At the same time, it is output to the outside of the video decoding device 2.

<Functional Configuration of Motion Vector Restoring Unit 23>
Next, a functional configuration of the motion vector restoration unit 23 that is a feature of the present invention will be described.
FIG. 8 is a block diagram showing a functional configuration of the motion vector restoration unit 23 that calculates the motion vector of the target partition. In FIG. 8, the motion vector restoration unit 23 includes a similarity PMV derivation unit 111 and an addition unit 112. Has been.

The similarity PMV deriving unit 111 is the same as the constituent elements of the video encoding device 1 shown in FIG.
The motion vector restoration unit 23 derives a prediction vector using the similarity calculated from the motion vector stored in the buffer memory 12 by the similarity PMV deriving unit 111, and outputs the output of the variable length code decoding unit 10 to the derived prediction vector. Are added by the adder 112 to calculate and output the motion vector of the target partition.

<Effect of prediction vector calculated by Embodiment 1>
The effect of using the prediction vector calculated by Embodiment 1 is demonstrated using FIG. 9 and FIG.

An example in which the rigid object indicated by the diagonal lines in FIG. 9 is horizontally long will be described.
Since each part in a rigid object generally has the same motion, it has a high correlation. Therefore, the motion vectors of the partitions inside the object tend to have a high correlation.
However, in a region other than the inside of a rigid object, the motion vector direction is generally different from the motion of the rigid object, and thus the direction and size of the motion vector cannot be said to be constant.

In addition, when the prediction partition is calculated using the median value of the neighboring partition of the target partition when the target partition is at the object boundary, the motion vector of the neighboring partition in the area outside the object is rarely obtained as an unexpected prediction vector. Absent.

For example, in FIG. 10, the prediction vector of the target partition (shaded area) located at the object boundary is calculated as a median from the motion vector A of the partition in the object inner area and the motion vectors B and C of the partition in the outer object area. It will be.

However, when the similarity according to the first embodiment is used, the similarity between the motion vectors in the direction belonging to the same object, here, the horizontal direction (left direction) increases, and thus the motion vector of the left partition is easily selected as the prediction vector. Become.
That is, the similarity PMV is effective in calculating a prediction vector when an object to which a part of a neighboring partition belongs and an object to which the target partition belongs are different.

As described above, according to the first embodiment, a motion vector in the spatial direction is used as a prediction vector candidate, and a prediction vector with a high similarity is selected from the motion vector candidates, so that the prediction vector prediction accuracy is improved.
In addition, since the selection of the prediction vector is narrowed down to one according to the degree of similarity, it is not necessary to encode information that distinguishes the method of deriving the prediction vector without reducing the number of substantial prediction vector candidates.

This makes it possible to select a prediction vector with high accuracy, and the difference motion vector and the prediction residual are reduced. As a result, the data amount of the encoded bit stream is reduced and the encoding efficiency is improved.

<< Embodiment 2 >>

<Functional Configuration of Video Encoding Device>
Next, a functional configuration of the video encoding apparatus according to the second embodiment will be described. Note that the video encoding apparatus according to the second embodiment is the same as that of the first embodiment shown in FIG. 1, and only the components of the motion vector redundancy reducing unit 24 are different. To do.

FIG. 11 is a block diagram illustrating a functional configuration of the motion vector redundancy reduction unit 24 of the video encoding device 1 according to the second embodiment. In FIG. 11, the motion vector redundancy reduction unit 24 includes the similarity PMV. The derivation unit 210 and the subtraction unit 191 are included.
The subtracting unit 191 is the same as the components having the same reference numerals in FIG.
In addition to the contents described in the first embodiment, the buffer memory 12 stores motion vectors of all blocks (partitions) for at least one frame in order to calculate the similarity in the time direction.

<Functional Configuration of Similarity PMV Deriving Unit 210>
As shown in FIG. 11, the similarity PMV deriving unit 210 includes a spatiotemporal direction similarity calculation unit 220, a similarity PMV determination unit 221, and a similarity PMV candidate extraction unit 222.

Furthermore, as shown in FIG. 11, the spatio-temporal direction similarity calculation unit 220 includes a spatial direction motion vector extraction unit 130, a temporal direction motion vector extraction unit 230, and a similarity calculation unit 231, and includes four types of spatial directions. In addition to the similarity, the temporal similarity is calculated and output.

The spatial direction motion vector extraction unit 130 extracts and outputs the motion vectors of the reference partition 1 and the reference partition 2 in the left, upper, upper right, and upper left directions in the same manner as in the first embodiment.

The temporal direction motion vector extraction unit 230 extracts a motion vector of a partition (hereinafter referred to as a collocated partition) that occupies the same position as the target partition in the encoded frame. The collocated partitions in different encoded frames are referred to as reference partition 1 and reference partition 2, respectively, and motion vectors of these reference partitions are extracted and output.
Next, a method of extracting the motion vectors of the reference partition 1 and the reference partition 2 in the time direction will be described with reference to FIGS.

(1) When the sizes of the target partition and the collocated partition are the same, the motion vectors of the reference partition 1 and the reference partition 2 are extracted as follows.
FIG. 12 shows an example of extracting the motion vector of the reference partition 1 frame before and 2 frames before the frame to which the target partition belongs. The extraction of the motion vector of the reference partition 1 The motion vector of the partition col ₁ (gray-dashed rectangular region) at the same position as the target partition (hatched solid line rectangular region) before the frame is extracted, and the motion vector of the reference partition 2 is extracted by two frames of the target frame N The motion vector of the previous partition col ₂ (colorless dashed rectangular area) at the same position as the target partition is extracted.

(2) When the sizes of the target partition and the collocated partition are different, the motion vectors of the reference partition 1 and the reference partition 2 are extracted as follows.
FIG. 13 shows an example in which the motion vector of the reference partition one frame before the frame to which the target partition belongs is extracted. For extracting the motion vector of the reference partition, as shown in FIG. 13, the motion vector of the partition on the reference image corresponding to the upper left pixel position of the target partition is selected.
For example, the reference partitions of frame (N−1) corresponding to partitions X, Y, and Z of frame N are colX, colY, and colZ, respectively.

(3) When the collocated partition belongs to an intra-coded block, the motion vectors of the reference partition 1 and the reference partition 2 are extracted as follows.
When a collocated partition belongs to an intra-coded block, a motion vector of an inter-coded partition existing at a position around the collocated partition is extracted.
For example, the reference partition 1 is an inter-coded partition one frame before that exists at the same position as the lower partition of the target partition, and the reference partition 2 is an inter-frame two frames before that exists at the same position as the lower partition of the target partition. Let it be an encoded partition.
Further, it is preferable that the reference partition 1 and the reference partition 2 are in the same position.
Further, the reference partition is not limited to the lower partition, and motion vectors may be extracted from the left, upper, and upper right partitions.
When the reference partition is intra-coded, the similarity may be 0 as described above, but the above method is more preferably used.

The similarity calculation unit 231 adds up to five types of similarities by adding up to four types of spatial directions output from the spatial direction motion vector extraction unit 130 and one type of temporal direction output from the temporal direction motion vector extraction unit 230. The degree is calculated by Expression (1), Expression (2), and Expression (3) and output.

It should be noted that the ease of selection of the motion vector located in the spatial direction and the motion vector located in the temporal direction may be adjusted using the weight w shown in Expression (3).

Similar to the similarity PMV candidate extraction unit 122 of the first embodiment, the similarity PMV candidate extraction unit 222 selects a maximum of four types of neighboring partitions from among the neighboring partitions in the spatial direction (left, upper, upper right, upper left) of the target partition. Extract and extract the motion vectors of those partitions and the motion vector of the collocated partition in the same manner as the reference partition 1 of the temporal direction motion vector extraction unit 230, and output a maximum of five types of motion vectors as prediction vector candidates To do.

The similarity PMV determination unit 221 compares up to five types of similarities input from the spatiotemporal direction similarity calculation unit 220 (that is, the similarity calculation unit 231), identifies the direction of the maximum similarity, and determines the similarity. The prediction vector candidate in the direction corresponding to the maximum similarity is selected from the prediction vector candidates extracted by the degree PMV candidate extraction unit 222 and is output as a prediction vector (similarity PMV).

In the determination of the similarity, when there are a plurality of directions of the maximum similarity, the prediction vector is determined in the priority order of the left direction, the time direction, the upward direction, the upper right direction, and the upper left direction.

Also, when the maximum similarity is smaller than a predetermined threshold, since the reliability of the similarity itself is low, an exception process for calculating the first prediction vector is performed. In this exceptional process, since it is determined that the degree of similarity is low, it is not necessary to add the motion vector in the time direction to the calculation target of the first prediction vector.

(Another configuration of the similarity calculation unit 231)
The time direction motion vector extraction unit 230 described above extracts the motion vector of the collocated partition for two frames and calculates the similarity in the time direction. However, another configuration may be used.
Specifically, a technique for calculating the temporal similarity by referring to a collocated partition for one frame and a neighboring partition of the collocated partition will be described.

For example, in the time direction, the distance dist between the motion vectors is calculated by calculating the difference between the motion vector of the neighboring partition and the motion vector of the neighboring partition of the collocated partition in one or a plurality of neighboring partitions. It is calculated by calculation such as average, weighted addition, median, minimum, maximum, and average excluding maximum and minimum values.

Similarly to the distance between the motion vectors in the time direction, the distance between the motion vectors in the spatial direction is, for example, the difference between the motion vector of the neighboring partition and the motion vector of the neighboring partition. It is calculated by the neighborhood partition, and is calculated by calculation such as the average of the calculated differences, weighted addition, median value, minimum value, maximum value, average excluding maximum value and minimum value.

Further, the difference between the motion vector of the neighboring partition and the median value of the motion vector of the neighboring partition is calculated in one or more neighboring partitions, and the average of the calculated differences, weighted addition, median value, minimum value, The distance between motion vectors in the spatial direction may be calculated by a calculation such as a maximum value, an average excluding the maximum value and the minimum value.

The spatial direction, the temporal direction, and the respective similarities are calculated from the distances between the motion vectors, and the similarity PMV is selected using the spatial direction similarity and the temporal direction similarity.

The operation of the similarity PMV deriving unit 210 adds one type of similarity in the time direction to the operation of the similarity PMV deriving unit 111 for up to four types of similarity in the spatial direction shown in FIG. Since the processes of S1 to S8 are merely replaced with the processes for five types of similarity, the description is omitted.

<Configuration of Encoded Bitstream Generated by Variable Length Encoding Unit of Embodiment 2>
The configuration of the bitstream generated in the second embodiment is the same as that described in the first embodiment.

<Functional configuration of video decoding device>
Next, a functional configuration of the video decoding device according to the second embodiment will be described. The moving picture decoding apparatus according to the second embodiment is the same as that of the first embodiment shown in FIG. 7, and only the components of the motion vector restoration unit 23 are different. Therefore, only the differences will be described.

FIG. 14 is a block diagram illustrating a functional configuration of the motion vector restoration unit 23 of the video decoding device 2 according to the second embodiment. In FIG. 14, the motion vector restoration unit 23 includes an addition unit 112 and a similarity PMV derivation. The unit 210 is configured.
The motion vector restoration unit 23 adds, for each target partition, the prediction vector determined by the similarity PMV deriving unit 210 and the difference motion vector decoded by the variable-length code decoding unit 10 as described in the first embodiment. The motion vector is decoded by adding in the unit 112.

<Effect of prediction vector calculated by Embodiment 2>
Next, the effect of using the prediction vector calculated by Embodiment 2 is demonstrated.
Since the similarity in the spatial direction is higher than the similarity in the temporal direction on average, it may be encoded by the method of the first embodiment. For example, in a region having no correlation in the spatial direction, the motion vector adjacent to the spatial direction is used. However, it is more effective to use a motion vector in the time direction as a prediction vector.
Therefore, if the temporal direction similarity is taken into account in addition to the spatial direction similarity as in the second embodiment, the temporal direction motion vector is easily calculated as a prediction vector in a region where there is no spatial direction correlation. Therefore, an optimal prediction vector can be selected.

As described above, according to the second embodiment, a motion vector in the spatial direction and the time direction is used as a prediction vector candidate, and a prediction vector having a high similarity is selected from the motion vector candidates, so that the prediction accuracy of the prediction vector is improved.
In addition, since the selection of the prediction vector is narrowed down to one according to the degree of similarity, it is not necessary to encode information that distinguishes the method of deriving the prediction vector without reducing the number of substantial prediction vector candidates.

This makes it possible to select a prediction vector with high accuracy, and the difference motion vector and the prediction residual are reduced. As a result, the data amount of the encoded bit stream is reduced and the encoding efficiency is improved.

<< Embodiment 3 >>

<Functional Configuration of Video Encoding Device>
Next, a functional configuration of the video encoding apparatus according to the third embodiment will be described. Note that the moving picture coding apparatus according to the third embodiment is the same as that of the first embodiment shown in FIG. 1, and only the components of the motion vector redundancy reduction unit 24 are different. To do.

FIG. 15 is a block diagram illustrating a functional configuration of the motion vector redundancy reducing unit 24 of the video encoding device 1 according to the third embodiment. In FIG. 15, the motion vector redundancy reducing unit 24 performs the first prediction. The vector derivation unit 190, the similarity PMV derivation unit 210, and the PMV determination unit 193 calculate a plurality of prediction vector candidates, select a prediction vector having a low coding cost from the candidates, and select the selected prediction vector. The PMV flag, which is information for distinguishing the method for deriving the, is output.

The first prediction vector deriving unit 190 derives a first prediction vector.

The similarity PMV deriving unit 210 determines the prediction vector of the target partition using the spatial and temporal similarity described in the second embodiment, and outputs it as the similarity PMV.

The PMV determination unit 193 determines a difference motion vector between the first prediction vector output from the first prediction vector derivation unit 190 and the motion vector of the target partition for the similarity PMV output from the similarity PMV derivation unit 210, respectively. And calculating a coding cost based on these difference motion vectors, selecting a prediction vector (first prediction vector or similarity PMV) that is the minimum coding cost, and calculating a difference motion vector calculated from the selected prediction vector. And the PMV flag is output.
In this embodiment, in order to explicitly indicate the encoding of the flag, the generation of the PMV flag is shown as a means different from the variable length encoding unit 17, but the variable length encoding unit 17 directly The selection information indicating the PMV flag may be encoded. In this case, the PMV flag generation process is included in the variable length encoding unit 17. Here, the PMV determination unit 193 and the variable length encoding unit 17 that encode the selection information indicating the PMV flag are simply referred to as “encoding unit”.

Here, as the coding cost, the code amount when the differential motion vector is encoded may be used, or the code amount when the differential motion vector is encoded including the PMV flag may be used.
The PMV flag is information for distinguishing a method for deriving a prediction vector from a plurality of prediction vector candidates. In the third embodiment, since the distinction represents whether the similarity PMV deriving unit 210 or the first predictive vector deriving unit 190 is selected, it is composed of 1 bit. For example, when it is zero, the similarity PMV deriving is performed. When the unit 210 is selected and the value is 1, it is defined that the first prediction vector deriving unit 190 is selected. When there are a plurality of motion vector candidates, the PMV flag is a flag of 2 bits or more.

<Configuration of Encoded Bitstream Generated by Variable Length Encoding Unit of Embodiment 3>
The encoded data generated by the variable-length encoding unit 17 of the moving image encoding apparatus 1 according to the third embodiment is the quantized prediction residual data, the prediction mode, and the difference motion vector, which are standardized in Non-Patent Document 1 described above. This is a bit stream generated in a normalized format.

The bit stream of FIG. 17 is a portion corresponding to the target block, and following the first block mode information, N pieces of the partition division number PMV flag information Flgi (0 ≦ i <N) in the target block is arranged, After that, index information Idxi (0 ≦ i <N) is arranged, and then motion vector information MVi (0 ≦ i <N) is arranged.
Since the configuration other than the PMV flag information Flgi is the same as that described in the first embodiment, a description thereof will be omitted.

The PMV flag information Flgi includes a PMV flag for each partition in the target block. That is, the information which distinguishes which prediction vector derivation method was used in the moving image encoder is shown.
By determining this PMV flag, the moving picture decoding apparatus can derive a prediction vector using the same processing as that of the moving picture encoding apparatus.

<Functional configuration of video decoding device>
Next, a functional configuration of the video decoding device according to the third embodiment will be described. The video decoding device can restore the same prediction vector as the prediction vector derived by the video encoding device based on the encoded data configured by the bit stream. Moreover, the moving image decoding apparatus can generate a decoded image based on such a prediction vector.
Note that the moving picture decoding apparatus according to the third embodiment is the same as that of the first embodiment shown in FIG. 7, and only the components of the motion vector restoration unit 23 are different. Therefore, only the differences will be described.

FIG. 18 is a block diagram illustrating a functional configuration of the motion vector restoration unit 23 of the video decoding device 2 according to the third embodiment. In the figure, the motion vector restoration unit 23 includes an addition unit 112 and a PMV determination unit 320. The prediction vector of the target partition decoded by the PMV determination unit 320 and the difference motion vector decoded by the variable length code decoding unit 10 are added by the adding unit 112 to decode the motion vector.
Further, the PMV determination unit 320 includes a similarity PMV deriving unit 210, a flag selecting unit 330, a first prediction vector deriving unit 190, and a switching unit 331.

As described in the second embodiment, the similarity PMV deriving unit 210 calculates the similarity PMV using the spatial similarity and the temporal similarity.
The first prediction vector deriving unit 190 derives a first prediction vector.

The flag selection unit 330 outputs a PMV flag for the target partition decoded by the variable length code decoding unit 10. In the above description, the variable length code decoding unit 10 is configured to decode the PMV flag. However, the variable length code decoding unit 10 directly decodes selection information that is information indicating which prediction vector candidate is selected. It doesn't matter.
The switching unit 331 switches the output value from the similarity PMV deriving unit 210 and the output value from the first prediction vector deriving unit 190 according to the PMV flag output from the flag selecting unit 330. For example, if the PMV flag is zero, the similarity PMV output by the similarity PMV deriving unit 210 is output as a prediction vector, and if the PMV flag is 1, the first prediction vector deriving unit 190 outputs One prediction vector is output.

<Operation of PMV Determination Unit 320>
The operation of the PMV determination unit 320 will be described with reference to the flowchart of FIG.

When the PMV flag of the target partition is determined and the PMV flag is zero (YES in step S10), the similarity PMV deriving unit 210 described in the second embodiment is performed to use the spatiotemporal similarity. The similarity PMV calculated in this way is output as a prediction vector (step S11).

On the other hand, if the PMV flag is 1 (NO in step S10), the first prediction vector calculated by the first prediction vector deriving unit 190 is output (step S12).

<Effect of prediction vector calculated by Embodiment 3>
In the conventional MV Completion method described in Non-Patent Document 2, as a prediction vector candidate, a motion vector using a median value, a spatial motion vector, a temporal motion vector, or a newly derived motion vector Etc. are used.
As described above, it is considered that the prediction vector prediction accuracy is improved by increasing the number of prediction vector candidates, but the code amount of information for identifying which candidate is selected increases.

On the other hand, in the third embodiment, the similarity PMV is used as one of the prediction vector candidates in the MV Completion method, so that there is an effect of reducing the code amount.
This is because the similarity PMV is narrowed down to one candidate from among a plurality of prediction vector candidates by using the similarity, so that it is not necessary to add a flag necessary for indicating a desired prediction vector, and the similarity PMV And only a flag for selecting one of the other prediction vectors need be encoded.

Therefore, if candidates are narrowed down using the same algorithm as described above for encoding and decoding, the flag as described above is only one bit, and the amount of code can be reduced as compared with the conventional method.

In the third embodiment, the similarity PMV (the motion vector having the maximum similarity among the spatial direction motion vector and the temporal direction motion vector) and the first prediction vector are used as the prediction vector candidates, and the one with the lower coding cost is selected. Although the prediction vector is used, the prediction vector candidate is not limited to this and may be other combinations.

For example, similarity PMV and temporal direction motion vector (colocated partition motion vector), or similarity PMV and spatial direction motion vector (for example, motion vector of a partition adjacent in the left direction or upward direction) are used as prediction vector candidates. Also good.

As described above, according to the third embodiment, the prediction vector is narrowed down to one of the method of selecting a prediction vector based on the similarity or the method of using the first prediction vector, and the prediction vector is selected. Since only the information for discriminating between the two is encoded, the discrimination information can be reduced without reducing the substantial number of prediction vector candidates.

This makes it possible to select a prediction vector with high accuracy, and the difference motion vector and the prediction residual are reduced. As a result, the data amount of the encoded bit stream is reduced and the encoding efficiency is improved.

<< Modification of Video Encoding Device >>
The present invention is not limited to the above embodiment. Below, the modification of the moving image encoder of this invention is demonstrated.

FIG. 20 is a block diagram showing a functional configuration of a motion vector redundancy reducing unit according to a modification of the video encoding device of the present invention. Note that the moving image encoding apparatus according to the modification of the present invention is the same as the moving image encoding apparatus 1 shown in FIG. 1, and only the components of the motion vector redundancy reduction unit 24 ′ are different. Only the differences will be described.

In FIG. 20, the motion vector redundancy reduction unit 24 ′ includes a first prediction vector derivation unit 190 that calculates a first prediction vector from motion vectors of neighboring partitions that are adjacent to the target partition on the processing target image, A second prediction vector deriving unit 192 that calculates a second prediction vector different from the first prediction vector from motion vectors of neighboring partitions that are adjacent to the target partition on the processing target image; A time-direction motion vector calculation unit 194 that calculates a time-direction motion vector from motion vectors of partitions located around the partition adjacent to the target partition at the same position as the target partition, and 1 corresponding to the time direction Use one or more pairs of motion vectors When the similarity in the time direction is greater than or equal to a predetermined value, the similarity calculation unit 231 ′ that obtains the similarity in the motion vector in the time direction is output from the first prediction vector deriving unit 190. The prediction vector and the temporal direction motion vector output from the temporal direction motion vector calculation unit 194 are used as prediction vector candidates, and otherwise output from the first prediction vector and the second prediction vector derivation unit 192. A prediction vector candidate selection unit 195 that uses the second prediction vector as a prediction vector candidate, and the encoding cost based on both prediction vectors, and the one with the lower encoding cost is set as the prediction vector of the target block, and the prediction vector Includes a flag indicating which of the predicted vector candidates and a PMV determination unit 193 ′ that outputs a differential motion vector. There.

More specifically, the first prediction vector uses the median value of the motion vectors of the partitions adjacent to the left, top, and right of the target partition, and the second prediction vector uses the motion of the partition adjacent to the left of the target partition. It is preferable to use a vector and use the motion vector of the partition at the same position as the target partition on the already processed image as the time direction motion vector, but it is not necessary to be limited to this.

The similarity in the time direction is obtained as follows, for example.
First, on the processing target image, the motion vector MV_U0 of the partition adjacent on the target partition and the motion vector MV_U1 of the block one frame before the processing target image at the same position as the partition position are calculated. DU, the motion vector MV_L0 of the partition adjacent to the left of the target partition on the processing target image, and the motion vector MV_L1 of the block one frame before the processing target image at the same position as the partition position The value dL calculated from the set of

dU = | MV_U0-MV_U1 |
dL = | MV_L0−MV_L1 |

Next, the minimum value of dU and dL is set as the distance dist, and the similarity is obtained by Expression (3).
In the above method, when the distance of the motion vector in the time direction is small in either the upper partition or the left partition of the target partition, the similarity of the motion vector in the time direction to be calculated increases. This method is suitable because it is considered that there is a spatial correlation in whether the motion vectors in the temporal direction are similar.

In the above, the minimum value of the distance is obtained from a set of a plurality of vectors. However, the distance may be calculated using an average value, a median value, or the like without being limited to the minimum value.

Note that the method of setting the similarity obtained from a plurality of sets of motion vectors as the similarity in a certain direction (here, the time direction) is not limited to the modification of the third embodiment, and the first embodiment, the second embodiment, and the like. You may use in.

As described above, the motion vector redundancy reducing unit 24 ′ can switch the prediction vector candidates based on the similarity in the time direction.
Specifically, since the prediction accuracy of the prediction vector in the time direction is considered to be high when the similarity in the time direction is equal to or greater than a predetermined value, the first prediction vector and the temporal direction motion vector are used as the prediction vector. If the temporal direction similarity is equal to or lower than a predetermined value, it is considered that the prediction accuracy of the prediction vector in the spatial direction is high. Therefore, the first prediction vector and the second prediction vector are regarded as prediction vector candidates. By doing so, a prediction vector with high prediction accuracy can be derived.

<< Modification of Video Decoding Device >>
A moving picture decoding apparatus for decoding an encoded bit stream created in a modification of the moving picture encoding apparatus of the present invention will be described.

FIG. 21 is a block diagram showing a functional configuration of a motion vector restoration unit according to a modification of the video decoding device of the present invention. Note that the moving picture decoding apparatus according to the modification of the present invention is the same as the moving picture decoding apparatus 2 shown in FIG. 7, and only the constituent elements of the motion vector restoration unit 23 ′ are different. To do. In addition, the same code | symbol is attached | subjected to the same thing as the component demonstrated in the modification of said moving image decoding apparatus.

In FIG. 21, the motion vector restoration unit 23 ′ includes a first prediction vector derivation unit 190, a second prediction vector derivation unit 192, a time direction motion vector calculation unit 194, a similarity calculation unit 231 ′, and a variable length code. A flag selection unit 330 that outputs a PMV flag for the target partition decoded by the decoding unit 10, a prediction vector output from the first prediction vector derivation unit 190, the second prediction vector derivation unit 192, and the temporal direction motion vector calculation unit 194 A switching unit 331 ′ is provided that switches candidates according to the similarity in the time direction output from the similarity calculation unit 231 ′ and the PMV flag output from the flag selection unit 330.

When the similarity in the time direction is equal to or greater than a predetermined value, the switching unit 331 ′ outputs the first prediction vector output by the first prediction vector deriving unit 190 if the PMV flag is zero, and the PMV flag is 1 If so, the time direction motion vector output by the time direction motion vector calculation unit 194 is output.
Conversely, when the similarity in the time direction is less than a predetermined value, if the PMV flag is zero, the first prediction vector output by the first prediction vector deriving unit 190 is output, and the PMV flag is 1. For example, the second prediction vector output unit 192 outputs the second prediction vector.

As described above, the motion vector restoration unit 23 ′ can switch the prediction vector candidates based on the similarity in the time direction. Therefore, encoded data with high encoding efficiency can be decoded by using the moving image decoding apparatus according to this modification.

In addition, the present invention is not limited to the above-described embodiment, and various modifications and corrections can be made without departing from the scope of the present invention.
For example, each unit (for example, transform / quantization unit, variable length coding unit, inverse quantization / inverse transform unit, motion vector estimation unit, inter prediction image generation unit, Intra prediction image generation unit, prediction scheme control unit, motion vector redundancy reduction unit, etc.) or each part of the video decoding device (for example, variable length code decoding unit, motion vector restoration unit, inter prediction image generation unit, intra prediction image generation) A program for causing a computer to function as a recording unit, a prediction method determination unit, an inverse quantization / inverse conversion unit, etc., and written in a recording medium in advance, and these programs recorded in the recording medium were provided in the computer The object of the present invention is achieved by storing the program in a memory or a storage device and executing the program by a CPU (Central Processing Unit) or the like. Needless to say, the.

In this case, the program itself read from the recording medium realizes the functions of the above-described embodiment, and the program and the recording medium recording the program also constitute the present invention.
In addition, the program includes a case where the functions of the above-described embodiment are realized by processing in cooperation with an operating system or another application program based on an instruction of the program.

The program for realizing the functions of the above-described embodiments includes a disk system (for example, magnetic disk, optical disk, etc.), a card system (for example, memory card, optical card, etc.), and a semiconductor memory system (for example, ROM, nonvolatile memory). Etc.) and a recording medium of any form such as a tape system (for example, magnetic tape, cassette tape, etc.). Or you may make it receive the said program stored in the memory | storage device via a network or a broadcast wave directly from a server computer. In this case, the storage device of this server computer is also included in the recording medium of the present invention.

Further, a part or all of the moving image encoding device and the moving image decoding device in the above-described embodiment may be realized as an LSI (Large Scale Integration) that is typically an integrated circuit. Each functional block of the moving image encoding device and the moving image decoding device may be individually chipped, or a part or all of them may be integrated into a chip. Further, the method of circuit integration is not limited to LSI, and may be realized with a dedicated circuit or a general-purpose processor. In addition, when an integrated circuit technology that replaces LSI appears due to progress in semiconductor technology, an integrated circuit based on the technology can also be used.

DESCRIPTION OF SYMBOLS 1 ... Moving image encoder, 12 ... Buffer memory, 13 ... Inter estimated image production | generation part, 14 ... Inverse quantization and inverse transformation part, 15 ... Transformation / quantization part, 16 ... Prediction scheme control part, 17 ... Variable length Encoding unit, 18 ... motion vector estimation unit, 20 ... intra prediction image generation unit, 21 ... addition unit, 22 ... subtraction unit, 24 ... motion vector redundancy reduction unit, 111 ... similarity PMV derivation unit, 191 ... subtraction unit , 120 ... spatial direction similarity calculation unit, 130 ... spatial direction motion vector extraction unit, 131 ... similarity calculation unit, 121 ... similarity PMV determination unit, 122 ... similarity PMV candidate extraction unit, 210 ... similarity PMV derivation unit , 220 ... spatiotemporal direction similarity calculation unit, 230 ... time direction motion vector extraction unit, 231 ... similarity calculation unit, 221 ... similarity PMV determination unit, 222 ... similarity PMV candidate extraction unit, 190 ... first prediction Vector derivation unit, 193 ... PMV determination unit, 24 '... motion vector redundancy reduction unit, 192 ... second prediction vector derivation unit, 193' ... PMV determination unit, 194 ... time direction motion vector calculation unit, 195 ... prediction vector candidate Selection unit, 231 '... similarity calculation unit, 2 ... video decoding device, 10 ... variable length code decoding unit, 23 ... motion vector restoration unit, 25 ... prediction method determination unit, 112 ... addition unit, 320 ... PMV determination unit , 330... Flag selection unit, 331... Switching unit, 23 ′... Motion vector restoration unit, 331 ′.

Claims (14)

In a video encoding device that generates a differential motion vector from a motion vector of a target block and a prediction vector and encodes a video,
One or a plurality of sets of motion vectors of two or more blocks corresponding to one or more directions in the time direction or the spatial direction are extracted, and the degree of similarity corresponding to the direction is calculated from the set of motion vectors. A similarity calculator;
A moving picture code comprising: a prediction vector deriving unit for deriving a motion vector calculated by a method corresponding to a direction in which the degree of similarity is maximum as a prediction vector of the target block or one of the prediction vectors Device. The similarity calculation unit is a set of motion vectors for calculating the similarity corresponding to the direction in the spatial direction, and is adjacent to at least two spatial directions in the neighboring blocks of the target block on the processing target image. 2. The moving image according to claim 1, wherein a set of two motion vectors for each spatial direction, which includes a motion vector of a neighboring block and a motion vector of a neighboring block that exists away from the target block in the direction, is used. Encoding device. The similarity calculation unit is present at the same position as the processing target block on the already processed first image as a set of motion vectors for calculating the similarity corresponding to the direction in the time direction, or the processing A motion vector of a block existing at a position around the target block and a second image that has already been processed, different from the first image, are present at the same position as the processing target block, or the processing target block 2. The moving picture coding apparatus according to claim 1, wherein a pair with a motion vector of a block existing at a position around is used. The similarity calculation unit, as a set of motion vectors for calculating the similarity corresponding to the direction of the time direction, on the encoding target image, on the block adjacent to the processing target block, on the already processed image, One or more pairs of motion vectors of blocks existing at the same position as the block adjacent to the processing target block or existing at positions around the block adjacent to the processing target block are used. The moving picture encoding apparatus according to claim 1. A motion vector redundancy reducing unit that generates a differential motion vector from a difference between the prediction vector and the motion vector of the target block; and a variable length encoding unit that generates encoded data using the differential motion vector,
The similarity calculation unit includes a motion vector of adjacent neighboring blocks existing in at least two spatial directions among neighboring blocks of the target block on the processing target image, and a neighborhood existing away from the target block in the direction. Extracting a set of two motion vectors for each spatial direction composed of motion vectors of blocks, and calculating a similarity based on a distance between the motion vectors calculated from the set of motion vectors;
The video encoding apparatus according to claim 2, wherein the prediction vector deriving unit derives a motion vector of a block having the maximum similarity as a prediction vector of the target block. A first prediction vector derivation unit that calculates a first prediction vector from motion vectors of neighboring blocks adjacent to the target block on the processing target image;
The prediction vector derivation unit derives the first prediction vector as a prediction vector when the similarity calculated by the similarity calculation unit is a predetermined threshold value or less. The moving image encoding apparatus described. A first prediction vector deriving unit that calculates a first prediction vector from motion vectors of neighboring blocks that are adjacent to the target block on the processing target image;
The maximum motion vector of the similarity calculated by the similarity calculation unit and the first prediction vector derived by the first prediction vector deriving unit are candidates, and the encoding costs based on both prediction vectors are compared. An encoding unit that sets and encodes a flag indicating which of the candidates the prediction vector is the prediction vector of the target block,
The moving picture coding apparatus according to claim 1, further comprising: A first prediction vector deriving unit that calculates a first prediction vector from motion vectors of neighboring blocks that are adjacent to the target block on the processing target image;
A second prediction vector derivation unit that calculates a second prediction vector different from the first prediction vector from motion vectors of neighboring blocks existing adjacent to the target block on the processing target image;
A time direction for calculating a time direction motion vector from a motion vector of a block located at the same position as the block adjacent to the processing target block or around the block adjacent to the processing target block on an already processed image A motion vector calculation unit;
When the similarity calculation unit calculates the similarity of the motion vector in the time direction using one or a plurality of motion vector pairs corresponding to the time direction, the similarity in the time direction is equal to or greater than a predetermined value. In some cases, the first prediction vector and the temporal motion vector are used as prediction vector candidates. In other cases, the prediction vector candidate selection is performed using the first prediction vector and the second prediction vector as prediction vector candidates. And
Compare the coding costs based on both prediction vectors, set the one with the lower coding cost as the prediction vector of the target block, and set and encode a flag indicating which of the prediction vector candidates the prediction vector is An encoding unit;
The moving picture coding apparatus according to claim 1, further comprising: A prediction vector of a target block is generated from a motion vector of a block obtained by decoding encoded data obtained by encoding a moving image to be processed using motion compensation prediction, and a motion vector is generated from the difference motion vector of the target block and the prediction vector. In the moving picture decoding apparatus that restores and decodes the encoded data,
One or a plurality of sets of motion vectors of two or more blocks corresponding to one or more directions in the time direction or the spatial direction are extracted, and the degree of similarity corresponding to the direction is calculated from the set of motion vectors. A similarity calculator;
And a prediction vector deriving unit that derives a motion vector calculated by a method corresponding to a direction in which the similarity is maximized as one of the prediction vector of the target block or one of the prediction vectors. Image decoding device. The similarity calculation unit is a set of motion vectors for calculating the similarity corresponding to the direction in the spatial direction, and is adjacent to at least two spatial directions in the neighboring blocks of the target block on the processing target image. The moving image according to claim 9, wherein a set of two motion vectors for each spatial direction, which includes a motion vector of a neighboring block and a motion vector of a neighboring block that exists away from the target block in the direction, is used. Decoding device. The similarity calculation unit is present at the same position as the processing target block on the already processed first image as a set of motion vectors for calculating the similarity corresponding to the direction in the time direction, or the processing A motion vector of a block existing at a position around the target block and a second image that has already been processed, different from the first image, are present at the same position as the processing target block, or the processing target block The moving image decoding apparatus according to claim 9, wherein a pair of motion vectors of blocks existing at positions around the image is used. The similarity calculation unit, as a set of motion vectors for calculating the similarity corresponding to the direction of the time direction, on the encoding target image, on the block adjacent to the processing target block, on the already processed image, One or more pairs of motion vectors of blocks existing at the same position as the block adjacent to the processing target block or existing at positions around the block adjacent to the processing target block are used. The moving picture decoding apparatus according to claim 9. A motion vector restoration unit that restores a motion vector from the prediction vector and the difference motion vector of the target block;
The similarity calculation unit includes a motion vector of adjacent neighboring blocks existing in at least two spatial directions among neighboring blocks of the target block on the processing target image, and a neighborhood existing away from the target block in the direction. Extracting a set of motion vectors for each spatial direction composed of block motion vectors, and calculating a similarity based on a distance between the motion vectors calculated from the set of motion vectors;
The moving picture decoding apparatus according to claim 10, wherein the prediction vector deriving unit derives a motion vector of a block having the maximum similarity as a prediction vector of the target block. A first prediction vector deriving unit that calculates a first prediction vector from motion vectors of neighboring blocks that are adjacent to the target block on the processing target image;
A second prediction vector derivation unit that calculates a second prediction vector different from the first prediction vector from motion vectors of neighboring blocks that are adjacent to the target block on the processing target image;
A time direction for calculating a time direction motion vector from a motion vector of a block located in the same position as the block adjacent to the processing target block or around the block adjacent to the processing target block on an already processed image A motion vector calculation unit;
When the similarity calculation unit calculates the similarity of the motion vector in the time direction using one or a plurality of motion vector pairs corresponding to the time direction, the similarity in the time direction is equal to or greater than a predetermined value. In some cases, the first prediction vector and the temporal motion vector are used as prediction vector candidates. In other cases, the prediction vector candidate selection is performed using the first prediction vector and the second prediction vector as prediction vector candidates. And
A flag selection unit that outputs a PMV flag for the target block decoded by the variable-length code decoding unit;
A switching unit that switches the prediction vector candidate output from the prediction vector candidate selection unit according to the temporal direction similarity output from the similarity calculation unit and the PMV flag output from the flag selection unit; The moving picture decoding apparatus according to claim 9, further comprising:

Images