JPH0630395A - Method for predicting motion compensation - Google Patents

Abstract

PURPOSE:To improve the accuracy of prediction by calculating motion compensation of an input picture at an optional setting time based on an input picture motion vector and a reference picture motion vector. CONSTITUTION:Let the magnitude of vertical component of a motion vector (MV) detected be 1, then a motion compensation predicted value of a (M,1) is a picture element at a position of (M-2,2). Then a time position is corrected so that a position of a reference picture in a (M-1) th field is a position of a picture in a (M-2)th field. When the position of the (M-1)th field is corrected to be the position of the (M-2)th field, the position a (M-1,2) is corrected to be the position (M-2,2.5). Then the position (M-2,2) being a predicted value of motion compensation of the a (M,1) is obtained from a picture element in the (M-2)th field and in the (M-1)th field subjected to time position correction. In this case, a picture element at the position (M-2,2), that is, the predicted value of motion compensation of the a (M,1) is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion-compensated prediction method for a moving image used in a device requiring the prediction of a moving image such as image transmission or image compression.

[0002]

2. Description of the Related Art In recent years, with the progress of semiconductor technology, motion compensation prediction methods used for image transmission and image compression have been used in many fields. As a conventional motion compensation prediction method, there is a method of performing motion compensation from a certain reference image.

FIG. 6 is a diagram showing the concept of a conventional image motion compensation prediction method. In FIG. 6, the moving image signal is a set of images sampled at equal intervals t0 on the time axis. For example, in the NTSC signal, each field is sampled every 1/60 second, and in the PAL signal, each field is sampled every 1/50 second. Here, for example, when a certain subject is moving, the spatial position of the subject A ′ in the (M−1) th image and the spatial position of the subject A in the Mth image are between t0. It will be displaced by the amount of movement. Here, consider a case where the M-th image is predicted from the (M-1) -th image. In this case, in order to compensate for the motion between the input image and the reference image during the time difference t0 and perform accurate prediction, the M-th image is divided into blocks including one or more pixels, and (M-1 ) The motion is detected between the second image and the pixel value at the position shifted by this motion is used as the predicted value. This will be described with reference to FIG. 6. As the predicted value of the pixel X of the Mth image, the (M−1) th pixel is calculated.
Pixel X ″, which is obtained by shifting the motion MV detected in block units from X ′, which is a pixel spatially at the same position as the pixel X in the th image, is used as the predicted value of pixel X. However, FIG.
In, the block size is 3 × 3.

When the signal is an interlaced signal, there are some cases such as the case where the image is a frame, the case where it is a field, and the case where a reference image is a frame and an input image is a field. The basic idea is that explained in FIG. An example of this is the CMTT (Com
mission Mixte CCIR / CCITT pour les Transmissio
Reco standardized by ns Televisuelleset Sonores 3)
mmendation 723, "Transmission of component-cod
ed digital television signals for contributio
n-quality at the third hierarchical level of
CCITT Recommendation G.702 "is included. In this recommendation, interframe motion-compensated prediction and interfield prediction are adaptively switched. Since the motion is compensated for the prediction, the motion image including the motion can be accurately predicted.

[0005]

However, in the above-described conventional motion compensation prediction method, the pixel density of the image to be referred to becomes the pixel density of the reference image even when the motion compensation prediction cannot be performed correctly, or even when the motion compensation prediction can be performed correctly. There was a problem that it could not make a good prediction.

For example, when an interlaced signal is treated as a frame and a block is generated from the frame and motion compensation prediction is performed, the difference between the temporal sampling positions of two fields in the frame is ignored and the frames are combined to perform motion compensation. ing. Therefore, when the sampling positions of the correct fields are considered, the motions compensated in the first field and the second field may not match. Such an example is shown in FIG. In FIG. 7, the input signal is an interlaced signal (FIG. 7 (a)), which is combined into a frame for motion compensation prediction. The vertical component of the detected motion is 1
7B, the first field of the Mth frame is predicted from the second field of the (M−1) th frame, and the second field of the Mth frame is (M−1) th frame. ) It will be predicted from the first field of the frame. FIG. 7C shows this operation at the correct field position. As is clear from FIG. 7C, the compensated motions do not match in the first field and the second field of the Mth frame. In this way, when interlaced images are treated as frames and motion compensation is performed,
Since the motions to be compensated for in the first field and the second field may be different, there is a problem that the accuracy of prediction deteriorates in a vector in which such a phenomenon occurs.

Next, consider the case where the motion compensation prediction is performed as an image at the correct position without ignoring the sampling time difference between certain images as described above. As such an example, there is a case where a block is generated from a field for an interlaced signal and motion compensation prediction is performed, or a case where motion compensation prediction is performed for a non-interlaced signal. In this case, since the motion-compensated prediction is performed using the image at the temporally correct position, there is no problem as in the case where the block is generated from the frame of the interlaced signal and the motion-compensated prediction is performed. However, in this case, the prediction is performed from one reference image, and therefore the pixel density of the image to be referred to is the pixel density of the reference image, and there is a limit to performing more accurate prediction. FIG. 8 shows a case where a block is generated from a field for motion compensation by inputting an interlaced signal. In this case, since the field image is used as the reference image for prediction, when the motion vector is 0, there is no sampling point at the position required for prediction on the reference image, It is necessary to calculate a pixel value, that is, a predicted value by interpolation.
Comparing this with the case where a block is generated with pixel values in a frame and motion compensation is performed, the pixel density in the vertical direction when performing motion compensation in the field is half that when performing motion compensation in a frame. There are limits to trying to make compensation predictions. This is the same even when a non-interlaced signal is input, and the pixel density of the image to be referred to becomes the pixel density of the reference image, and there is a problem that there is a limit in attempting more accurate motion compensation prediction. .

The present invention solves such a conventional problem, and provides an excellent motion-compensated prediction method capable of performing highly accurate prediction by using a plurality of reference images. The purpose is to

[0009]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention uses a reference image obtained by sampling motion compensation of an input image at a first set time and a motion vector between the reference image and the reference image. In the method of predicting, a unit for calculating an input image motion vector in which a block unit consisting of a plurality of pixels, which is a part of the input image, is moved in a second set time, and a plurality of pixels which are a part of the reference image. And a unit for calculating a reference image motion vector moved in a first set time, and performing motion compensation of the input image at an arbitrary set time from the input image motion vector and the reference image motion vector. It is to be calculated.

A means for calculating an input image motion vector in which a block unit consisting of a plurality of pixels, which is a part of the input image, is moved in a second set time, and a first means using the input image motion vector A reference image motion vector for a set time is calculated, and motion compensation of the input image for an arbitrary set time is calculated from the input image motion vector and the reference image motion vector.

In a method of predicting motion compensation of a plurality of input images from a reference image sampled at a first set time and a motion vector between the reference images, a plurality of input images which are a part of these input images are predicted. And a unit for calculating an input image motion vector in which the block unit made up of pixels has moved in the second set time, and a block unit made up of a plurality of pixels that is part of the reference image moved in the first set time. And a means for calculating a reference image motion vector,
These input image motion vectors are the same, and motion compensation of the input image at an arbitrary set time is calculated from the input image motion vector and the reference image motion vector.

[0012]

According to the present invention, therefore, a plurality of reference images having different sampled times are input from the input image in the time interval according to the motion detected in the time interval in the block unit including one or more pixels. In order to correct the time position of the reference image by using a certain motion vector so that the images are separated by a certain time, it is necessary to obtain a plurality of images at positions separated by the time interval from the input image. You can A reference image having a high pixel density is obtained by combining the plurality of images, and a pixel value at a position compensated for the detected motion amount is calculated using the reference image having a high pixel density, and this is used as a predicted value. Therefore, very accurate motion compensation prediction can be performed.

Further, according to the present invention, the vector for correcting the time position of the reference image can be calculated from the motion detected at a certain time interval, and the motion vector for time correction is detected again. It is possible to perform motion compensation with high accuracy without the need for

Further, since an interlaced signal is used as an input signal and a reference image is two fields of a frame, the above motion compensation prediction can be applied to a frame image. On the other hand, it is possible to accurately predict.

Further, as a motion detected at a certain time interval in a block unit including one or more pixels, among the blocks of a plurality of input images, all or some of the spatial positions occupied by the blocks overlap each other. Since the same value is used for the blocks of the input image, it is not necessary to perform the motion detection for several blocks of the plurality of input images a plurality of times, and the motion-compensated prediction can be performed with high accuracy.

[0016]

FIG. 1 is a diagram for explaining a first embodiment of the present invention. FIG. 1 is premised on a motion-compensated prediction based on a field image in which a block is generated from pixels in a field with an input signal as an interlaced signal. Here, the input image is the Mth field, and the reference image is the (M-
1) field and (M-2) th field.
Now, in FIG. 1, it is assumed that a motion vector (MV) for performing motion compensation prediction of a block is detected at an interval of two fields, that is, between the Mth field and the (M-2) th field. For convenience of explanation, only the vertical movement is considered among the detected movements, and the pixel value is represented as a (x, y). Here, x represents a field number and y represents a line number. The line numbers are numbered as 1, 2, ... The vertical position of each pixel is always expressed in units of frame lines.

Now, let us consider obtaining the predicted value of a (M, 1). If the vertical component of the detected MV is 1, then a (M,
The motion compensation prediction value of 1) is the pixel value at the position of (M-2, 2). Next, the time position is corrected so that the position of the reference image in the (M-1) th field becomes the image in the position of the (M-2) th field. The vector for correcting this time position is MVadj. When calculating MVadj from MV, assuming that the movement between the (M-2) th field and the Mth field is constant, the following relationship can be easily derived.

MVadj = -MV / 2 Therefore, if the vertical component of MV is 1, the vertical component of MVadj is -0.5. As shown in FIG.
1) When the field is corrected to the position of the (M-2) th field, a (M-1,2) is corrected to the position of (M-2,2.5). After the above operation, the position of (M-2, 2) which is the motion compensation prediction value of a (M, 1) is changed to the (M-
2) Obtained from the pixel values of the field and the (M-1) th field whose time position has been corrected. At this time, if the required pixel value is obtained by a weighted average that is inversely proportional to the distance from the neighboring pixel value, the pixel value at the position of (M−2,2), that is, the motion compensation prediction value of a (M, 1) Is calculated by the following formula.

A (M−2,1) / 3 + 2 * a (M−1,2) / 3 The above is a description when only the vertical component is considered, but the same is true when both vertical and horizontal components are included. Shall be performed.

As described above, according to the first embodiment,
A plurality of reference images having different sampled times are imaged at a time distant from the input image by the time interval in accordance with a motion detected at a time interval in a block unit including one or more pixels. In order to correct the time position of the reference image by using a certain motion vector as necessary, it is possible to obtain a plurality of images at positions separated from the input image by the time interval. A reference image having a high pixel density is obtained by combining the plurality of images, and a pixel value at a position compensated by the detected motion amount is calculated using the reference image having a high pixel density, and this is used as a predicted value. Therefore, there is an effect that very accurate motion compensation prediction can be performed.

In the first embodiment, the reference image is assumed to be two images, but this number may be any number as long as it is plural, and the positions of the plural reference images are assumed to be the previous image and the previous image. However, these positions are arbitrary and MVadj rather than MV
The calculation was performed on the assumption that the motion is constant when calculating, but this assumption may be any assumption according to a certain law, and the pixel values in the neighborhood may be used to obtain the pixel value at the required position. Was obtained by a weighted average inversely proportional to the distance, but this method is not limited to the weighted average and may be obtained using, for example, the coefficient of a low-pass filter, and the required pixel value can be obtained by interpolation. However, this method is not limited to interpolation but may be extrapolation, for example.

Further, in the first embodiment, the vector MVadj for correcting the positions of the plurality of reference images is calculated from the motion vector MV detected in block units, but this is calculated in the (M-1) th field. And (M-2) th field can be detected independently of MV. In this case, by detecting independently, the time correction can be performed with a more accurate motion, so that there is an effect that more accurate motion compensation prediction can be performed.

Further, in the first embodiment, the input signal is used as an interlaced signal and a field image is assumed as each image. However, even if this is used as a non-interlaced image, prediction can be performed accurately by the same explanation. Has the effect.

As a second embodiment, a method will be shown in which an input signal is an interlaced signal and motion compensation prediction is performed in frame units. FIG. 2 is a diagram for explaining the second embodiment of the present invention. In FIG. 2, the reference image is the two fields of the previous frame, that is, the (M-1) th and (M-2) th fields, and the input image is the two fields of the current frame, that is, the Mth field and the (M + 1) th field. And
In addition, the motion vector for performing motion compensation prediction of a block in FIG. 2 is obtained separately for each of the two fields of the current frame in the input image and the reference frame and between the fields having the same phase as the input image. Be present. A motion vector for motion compensation prediction of the Mth field is represented as MV (M), and a motion vector for the (M + 1) th field is represented as MV (M + 1). The method of expressing the pixel value at each pixel position is in accordance with the first embodiment. Further, in FIG. 2, for the sake of convenience of explanation, only the vertical movement will be considered.

In FIG. 2, prediction of pixels in the Mth field is performed using the images of the (M-1) th field and the (M-2) th field. The operation at this time is exactly the same as that of the first embodiment. Therefore, assuming that the vertical component of MV (M) is 1.5, the predicted value of a (M, 1) is (M-2,2.
It becomes the pixel value at the position of 5), and this value is obtained by the following formula.

A (M−2,1) / 7 + 6 * a (M−1,2) / 7 Further, the prediction value of the pixel in the (M + 1) th field is the same as the prediction value of the pixel in the Mth field. Similarly, it is predicted from the two fields (M-1) and (M-2) of the reference frame. The idea at this time is similar to the prediction of the pixel in the Mth field, but in this case,
-2) The field must be corrected to the position of the (M-1) th field. The vector for correcting this time position is MVadj (M + 1). MVadj (M
When calculating +1) from MV (M + 1), assuming that the motion between the (M-2) th field and the (M + 1) th field is constant, the following relationship can be easily derived.

MVadj (M + 1) = MV (M + 1) / 2 Therefore, if the vertical component of MV (M + 1) is 1, then MVadj
The vertical component of j (M + 1) is 0.5. As shown in FIG. 2, when the position correction is performed from the (M-2) th field to the position of the (M-1) th field, a (M-2,3) becomes (M-
1, 2.5) position is corrected. After the above operation a
The position of (M-1,3), which is the predicted value of (M + 1,2), is the (M-1) th field and the (M-1) th position is time-position corrected.
-2) Obtained from the pixel value of the field. At this time, if the required pixel value is obtained by, for example, a weighted average inversely proportional to the distance from the neighboring pixel values, the pixel value at the position of (M-1,3), that is, the motion compensation prediction value of a (M + 1,2) Is calculated by the following formula.

A (M-1,4) / 3 + 2 * a (M-2,3) / 3 The above is a description when only the vertical component is considered, but the same applies to the case where both vertical and horizontal components are included. Shall be performed.

As described above, according to the second embodiment,
By using an interlaced signal as an input signal and using a reference image as two fields of a frame, the above motion-compensated prediction can be applied to a frame image. It has the effect of being able to make predictions.

In the second embodiment, as in the first embodiment, the number of reference frames, the position of the reference frame, MV (M) or MV (M + 1) is used as MVadj.
Assumptions for deriving (M) or MVadj (M + 1), a calculation method for obtaining a pixel value at a required position, and interpolation or extrapolation can be arbitrarily selected. Further, in the present embodiment, the motion vector for motion compensation prediction is found between the input image and the reference frame and between the input image and the in-phase field. The same effect can be obtained with the same operation even if required by. Further, similarly to the case of the first embodiment, by obtaining the position correction vector independently of the detected motion vector, there is an effect that more accurate motion compensation prediction can be performed.

As a third embodiment, another method of performing motion compensation on a frame-by-frame basis for an interlaced input signal will be described. 3 and 4 are diagrams illustrating a third embodiment of the present invention. In FIG. 3, the reference image is the (N-1) th frame, that is, the (M-2) th and (M-1) th field, and the input image is the Nth frame, that is, the Mth and (M +) th frame.
1) It is a field. Now, it is assumed that the block for motion compensation is generated from the frame. That is, the Nth
For each block (N +) generated from the pixels of the frame
1) It is assumed that the motion vector MV is obtained between the frame and the frame. The state of the block at this time is shown in FIG. If we reconsider this as a field-based prediction method, we can think as follows. The reference image is two fields of the (N-1) th frame, and the input image is two fields of the Nth frame. The MV detection interval is two field intervals. However, the pixels included in the block have the same motion vector MV regardless of whether the pixel is in the Mth field or in the (M + 1) th field.

That is, in this case, the motion vector used in the pixel in the block generated from the above frame in the second embodiment is independent of whether the pixel belongs to the Mth field or the (M + 1) th field. Without taking the same value. Other operations are similar to those of the second embodiment. FIG. 4 shows an example in which the vertical component of MV is 1. The above is a description when only the vertical component is taken into consideration, but the same operation is performed when both vertical and horizontal components are included.

As described above, according to the third embodiment, the same motion vector is applied to the pixels of two input fields located in a predetermined spatial area such as in a block generated from a frame. Since it is used, there is an effect that it is not necessary to detect motion for each field, and it is possible to perform accurate motion compensation prediction.

In the third embodiment, as in the case of the second embodiment, the number of reference frames, the position of the reference frame, the assumption for deriving MVadj from MV, and the calculation of the pixel value at the required position The calculation method and interpolation or extrapolation can be arbitrarily selected. In this embodiment, the description has been given based on the motion-compensated prediction in units of frames. However, this may be performed based on the field as shown in the first embodiment or the non-interlaced image. The effect does not change. In addition, as a method of determining a block that uses the same value as a motion vector among blocks of a plurality of input images, the effect is also obtained as a block of each input image in which all or some of the spatial positions occupied by each block overlap. does not change. Further, similarly to the case of the second embodiment, by obtaining the position correction vector independently of the detected motion vector, there is an effect that more accurate motion compensation prediction can be performed.

FIG. 5 is a diagram for explaining the fourth embodiment of the present invention. The premise of the fourth embodiment is the same as that of the first embodiment, the input signal is an interlaced signal, and the input image is the M-th image.
The fields and reference images are the (M-1) th field and the (M-2) th field. Now, referring to FIG. 5, a motion vector (M
V) is to be detected between two field intervals, that is, between the Mth field and the (M-2) th field.
For convenience of explanation, only the movement in the vertical direction among the detected movements will be considered, and the way of expressing the pixel value at each pixel position is the same as in FIG.

Now, let us consider obtaining the predicted value of a (M, 1). If the vertical component of the detected MV is 3, then a (M,
The motion compensation prediction value of 1) becomes the pixel value at the position of (M-2, 4). First, this pixel value is obtained from the pixel value in the (M-2) th field. For example, if the weighted average is inversely proportional to the distance from the neighboring pixel values, (M-2,
The pixel value at the position of 4) is obtained by the following formula.

A (M−2,3) / 2 + a (M−2,5) / 2 Next, from the MV, the input image (Mth field) and the (Mth field)
-1) Calculate the movement with the field. The time difference between the Mth field and the (M-1) th field is 1/2 of the time difference between the Mth field and the (M-2) th field. Therefore, this motion vector can be considered as MV / 2. Now, since the vertical component of MV is 3, the vertical component of MV / 2 is 1.5. Therefore, the (M-
1) When the motion compensation prediction value of a (M, 1) is obtained from the image of the field, the pixel value at the position of (M-1,2.5) is obtained. This pixel value is obtained from the pixel value in the (M-1) th field. For example, if the weighted average is inversely proportional to the distance from the neighboring pixel values, the pixel value at the position of (M-1, 2.5) is calculated by the following formula.

3 * a (M-1,2) / 4 + a (M-1,4) / 4 For example, the average of the two predicted values obtained above is calculated as a
Let it be the predicted value of (M, 1). The above is a description when only the vertical component is taken into consideration, but the same operation is performed when both vertical and horizontal components are included.

As described above, according to the fourth embodiment,
For each block unit including one or more pixels, the movement between a plurality of reference images and sampled images having different sampled times is calculated according to the movement detected at a certain time interval from the detected movement. However, it is possible to obtain a plurality of motion compensation prediction values from a plurality of reference images in order to calculate a pixel value at a position that is compensated for the calculated motion amount in each reference image. Since the predicted value of the input image is calculated from the plurality of predicted values, it is possible to remove this noise even when the predicted value includes noise, and it is possible to perform accurate motion compensation prediction. Have.

In the fourth embodiment, as in the case of the first embodiment, the number of reference images, the positions of the reference images, the calculation method for obtaining the pixel values at the required positions in each reference image, and the calculation method It is possible to arbitrarily select whether to insert or extrapolate. Regarding the method of calculating the predicted value from the multiple pixel values found in each reference image, there are methods other than simple averaging, such as weighting, and methods using low pass filter coefficients. Can be considered. In the present embodiment, the description has been given based on the motion compensation in units of fields of the interlaced signal. However, this is also based on the frame as shown in the second and third embodiments, and also non-interlaced. The effect does not change even if the image is used as a base.

[0041]

As is apparent from the above embodiment, the present invention refers to a plurality of reference samples having different sampling times according to the motion detected at a time interval in a block unit including one or more pixels. A position separated by the time interval from the input image in order to correct the time position of the reference image by using a certain motion vector so that the image becomes an image at the time separated by the time interval from the input image. You can get multiple images of. A reference image having a high pixel density is obtained by combining the plurality of images, and a pixel value at a position compensated for the detected motion amount is calculated using the reference image having a high pixel density, and this is used as a predicted value. Therefore, there is an effect that very accurate motion compensation prediction can be performed.

Furthermore, according to the present invention, the vector for correcting the time position of the reference image can be calculated from the motion detected at a certain time interval, and the motion vector for time correction is detected again. There is an effect that motion compensation can be performed with high precision without the need for In addition, since an interlaced signal is used as an input signal and a reference image is two fields of a frame, the above motion compensation prediction can be applied to a frame image. This has the effect that it is possible to make accurate predictions.

Further, as a motion detected at a time interval in a block unit including one or more pixels, among the blocks of a plurality of input images, all or some of the spatial positions occupied by the blocks overlap each other. Since the same value is used for the blocks of the input image, it is not necessary to perform motion detection multiple times for some blocks of multiple input images, and it is possible to perform accurate motion compensation prediction. Have.

[Brief description of drawings]

FIG. 1 is an explanatory diagram of a first embodiment of the present invention.

FIG. 2 is an explanatory diagram of a second embodiment of the present invention.

FIG. 3 is an explanatory diagram of blocks in frame-based motion-compensated prediction.

FIG. 4 is an explanatory diagram of a third embodiment of the present invention.

FIG. 5 is an explanatory diagram of a fourth embodiment of the present invention.

FIG. 6 is a conceptual diagram of a conventional motion compensation prediction method.

FIG. 7 is an explanatory diagram of problems in the conventional inter-frame motion compensation prediction method.

FIG. 8 is an explanatory diagram of a conventional inter-field motion compensation prediction method.

Claims (8)

[Claims] 1. A method for predicting motion compensation of an input image from a reference image sampled at a first set time and a motion vector between the reference image and a plurality of pixels which are part of the input image. Means for calculating an input image motion vector in which a block unit made up of is moved in a second set time, and a reference image movement made in a block unit made up of a plurality of pixels that is a part of the reference image in a first set time A motion compensation prediction method comprising: calculating a vector, and calculating motion compensation of the input image at an arbitrary set time from the input image motion vector and the reference image motion vector. 2. The reference image motion vector calculating means sets the sampling time of the reference image to at least two or more, and calculates by reference block movement data of the reference image by these sampling times. Item 1. The motion compensation prediction method according to Item 1. 3. A method of predicting motion compensation of an input image from a reference image sampled at a first set time and a motion vector between the reference image and a plurality of pixels which are part of the input image. The block unit consisting of consists of means for calculating an input image motion vector moved in a second set time, and means for calculating a reference image motion vector in the first set time using the input image motion vector,
A motion compensation prediction method, wherein motion compensation of the input image at an arbitrary set time is calculated from the input image motion vector and the reference image motion vector. 4. The reference image motion vector calculating means sets a sampling time of the reference image to at least 2 or more, and calculates the reference image movement data in block units according to these sampling times. 1. The motion compensation prediction method described in 1. 5. A method of predicting motion compensation of a plurality of input images from a reference image sampled at a first set time and a motion vector between the reference images, wherein a part of these input images is used. Means for calculating an input image motion vector in which a block unit made up of a plurality of pixels has moved at a second set time, and a block unit made up of a plurality of pixels that is a part of the reference image at a first set time And a means for calculating the moved reference image motion vector, and these input image motion vectors are the same,
A motion compensation prediction method, wherein motion compensation of the input image at an arbitrary set time is calculated from the input image motion vector and the reference image motion vector. 6. The reference image motion vector calculating means sets the sampling time of the reference image to at least two or more, and calculates by the movement data of the reference image in block units according to these sampling times. Item 5. The motion compensation prediction method according to Item 5. 7. A method of predicting motion compensation of a plurality of input images from a reference image sampled at a first set time and a motion vector between the reference images, wherein a part of these input images is used. Yes, means for calculating an input image motion vector in which a block unit composed of a plurality of pixels has moved in a second set time, and means for calculating a reference image motion vector in a first set time using the input image motion vector Motion compensation prediction, characterized in that these input image motion vectors are the same, and motion compensation of the input image at an arbitrary set time is calculated from the input image motion vector and the reference image motion vector. Method. 8. The reference image motion vector calculating means sets a sampling time of the reference image to at least 2 or more, and calculates the reference image movement data in block units according to the sampling time. 7. The motion compensation prediction method described in 7.