JPH08161524A - Inter-frame interpolating device and animation preparing device - Google Patents

Abstract

PURPOSE: To generate interpolation data by using a time-series filter in the case of a probablistic event. CONSTITUTION: Vector data of a frame are inputted from a vector data input part 11 and saved in a vector data memory part 13. An affine transformation matrix calculation part 14 reads vector data out of the vector data memory part 13 and calculates and affine matrix for linear interpolation. An affine transformation matrix coefficient correction part 15 corrects the affine transformation matrix with parameters inputted from a correction parameter input part 12 and saves it in a transformation matrix memory pat 16. A time-series filter part 17 calculates an interpolation matrix by performing a filtering process using the affine transformation matrix saved in the transformation matrix memory part 16 as an input signal value and vector data as an observation value. An affine transformation execution part 18 multiplies the original vector by the interpolation matrix and the result is outputted from an output part 19.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an interframe interpolating apparatus for moving images and an animation creating apparatus using the interframe interpolating apparatus, and more particularly to an apparatus suitable for interframe interpolating image data handling time series events. Is.

[0002]

2. Description of the Related Art Conventionally, techniques in such a field include:
For example, some documents were described in the following documents. Reference: Japanese Patent Application Laid-Open No. 03-109677 When given moving image data, it is necessary to construct a method of performing interframe interpolation between data of a frame f _n and a frame f _{n + 1} at an arbitrary time. Is an essential technique for dealing with. FIG. 2 is a functional block diagram of the conventional inter-frame interpolation device for moving images described in the above-mentioned document. The above-mentioned document proposes an interframe interpolating device as shown in FIG. 2, assuming that interpolating can be performed without a problem even if the position or shape of an object in an image changes significantly between frames f _n and f _{n + 1} . ing. That is, the moving image frame input from the input unit 1 is stored in the image memory 3. The moment calculation unit 4 obtains the primary and secondary moments of the interpolation target image frames f _n and f _{n + 1} and stores the primary and secondary moments in the moment memory 5. The affine transformation coefficient calculation unit 6 calculates the affine coefficients of the first and second moments stored in the moment memory 5. The coefficient correction unit 7 inputs a parameter from the correction parameter input unit 2 and multiplies the affine coefficient by a constant of the parameter by proportional calculation to perform an appropriate correction, and then the affine transformation execution unit 8 applies the correction to the original image. Perform affine transformation. The image synthesis unit 9 synthesizes the affine-transformed images and outputs a linearly interpolated image.

[0003]

However, the conventional inter-frame interpolation device has the following problems. In the conventional inter-frame interpolator, in the case of time-series data in which there is no causal relation in motion between the frames of the original image or there is very little (for example, a stochastic event such as Brownian motion of particle / Markov process). For example, when the noise component included in the data becomes too large, the solution cannot be estimated simply by performing the linear interpolation, and it is considered that the solution is far from the true value due to the noise amplification. Therefore, there is a problem in that it is not possible to obtain data that is operationally (physically) meaningful. Moreover, the width of deviation from the actual data increases as the number of interpolation frames increases.

[0004]

In order to solve the above-mentioned problems, a first invention is based on data that defines temporal changes in frames of first and second moving images that are temporally continuous.
An affine transformation matrix calculating means for calculating an affine transformation matrix that transforms into data defining a temporal change of an interpolated frame between frames of the first and second moving images by regarding the moving image as a non-stochastic event; , Input observing values of data of a plurality of moving image frames before and after the time to be interpolated, and the affine transformation matrix based on data of each two frames preceding and succeeding in time with respect to the frame data of the plurality of moving images A matrix based on the affine transformation matrix calculated by the calculating means is used as an input signal value, a time series filter for creating an interpolation matrix by filtering, an interpolation matrix created by the time series filtering process, and frame data of the moving image. And interpolation data creating means for creating interpolation data based on In the second invention, the time-series filter of the first invention is composed of a Kalman filter or a Wiener Healthstrom filter. In the third invention, the time-series filter of the first invention is configured to handle a noise component due to a moving image being a stochastic event as Gaussian white observation noise having an average value of 0. In a fourth aspect of the invention, the affine transformation matrix calculating means of the first aspect of the invention is configured to linearly interpolate by proportional distribution according to the number of interpolation frames between the frames of the first and second moving images.

In the fifth invention, the data defining the temporal change of the first and second frames is an affine transformation matrix. A sixth aspect of the present invention is based on vector data that defines temporal changes in frames of the first and second moving images that are temporally before and after each other, and based on vector data, a time of an interpolation frame between the frames of the first and second moving images. An affine transformation matrix calculating means for calculating an affine transformation matrix for transforming into data defining the change by linearly interpolating the moving image as a non-stochastic event, and an affine transformation calculated by the affine transformation matrix calculating means. And an affine transformation matrix coefficient correcting means for correcting the matrix. Then, when the moving image is a probabilistic event, the data of the frames of a plurality of moving images before and after the time to be interpolated are input observation values, and the data of the frame data of the plurality of moving images are two before and after in time. A time-series filter that creates an interpolation matrix by filtering using a matrix based on the affine transformation matrix calculated by the affine transformation matrix coefficient correcting means based on the data of one frame as an input signal value is provided. Further, when the moving image is a non-stochastic event, an interpolation vector is calculated by the affine transformation matrix corrected by the affine transformation matrix correcting means, and when the moving image is a stochastic event, it is obtained by the time series filter. Affine transformation executing means for calculating an interpolation vector by the interpolation matrix is provided.

According to a seventh aspect of the present invention, shape storage means for storing the shape of a moving object, operation storage means for storing the position of the moving object at each time, and position for each time stored in the operation storage means. Is used as vector data, and an interframe interpolating device of the sixth invention for interpolating the position of a moving object between preceding and following times, and a rewriting process for writing the position of the shape interpolated by the interframe interpolating device into the motion storage means. Means and a synthesis processing means for obtaining the shape of the object at each time and the interpolated time based on the position of the shape stored in the motion storage means and the shape of the object stored by the shape storage means. Is prepared.

[0007]

According to the first aspect of the invention, since the interframe interpolating device is configured as described above, the affine transformation matrix calculating means can be used to calculate the temporal change of the frames of the first and second moving images that are temporally preceding and following each other. Based on the specified data, the moving image calculates an affine transformation matrix that converts the temporal change of the interpolated frames between the frames of the first and second moving images into data that defines the non-stochastic event. When the moving image is a probabilistic event, the data converted by the affine transformation matrix calculated by the affine transformation matrix calculation means due to the noise component deviates from the actually observed data. Therefore, a matrix based on the affine transformation matrix calculated by the affine transformation matrix calculation means (the affine transformation matrix itself or the coefficient corrected by some correction means) may be used as the input signal value and the temporal change of the frame of the moving image. An interpolation matrix is calculated by a time series filter using the specified data as the input observation value. The interpolation data creating means creates the interpolation data based on the interpolation matrix calculated by the time series filter and the frame data of the moving image. Therefore, the above problem can be solved.

[0008]

1 is a functional block diagram of an interframe interpolation device according to an embodiment of the present invention. The difference between the interframe interpolating apparatus of the present embodiment and the conventional interframe interpolating apparatus is that the time series filter unit 17 is provided, and when the moving image is a stochastic event, the affine transformation matrix calculating unit 14 calculates it by linear interpolation. The affine transformation matrix corrected by the affine transformation matrix coefficient correcting unit 15 is used as the input signal value, and the original vector that defines the temporal change of the frame of the moving image is used as the input observation value to obtain the interpolation matrix. That is. This inter-frame interpolating device has a vector data input unit 11 for inputting vector data defining a temporal change of a frame of a moving image and a correction parameter input unit 12 for correcting an affine transformation coefficient. A vector data memory unit 13 for storing vector data is connected to the output side of the vector data input unit 11. On the output side of the vector data memory unit 13, an affine transformation matrix calculation unit 14 that calculates an affine transformation matrix by linear interpolation based on vector data, a time series filter unit 17, and an affine transformation execution unit 18 that calculates an interpolation vector. It is connected.

An affine transformation matrix coefficient correction unit 15 is connected to the correction parameter input unit 12 for inputting parameters for correcting the affine transformation coefficient and the output side of the affine transformation matrix calculation unit 14. The output side of the affine transformation matrix coefficient correction unit 15 is connected to the transformation matrix memory unit 16 that stores the corrected affine transformation matrix and the affine transformation execution unit 18. The time series filter unit 17 is connected to the output side of the conversion matrix memory unit 16. An affine transformation execution unit 18 is provided on the output side of the time series filter unit 17.
Is connected. An output unit 19 is connected to the output side of the affine transformation executing unit 18.

The time-series filter unit 17 uses the corrected affine transformation matrix stored in the transformation matrix memory unit 16 as an input signal value and the frame stored in the vector data memory unit 13 when the moving image is a stochastic event. This is a means for obtaining an interpolation matrix by filtering using vector data of as an input observation value. The operation of the interframe interpolation device of FIG. 1 will be described below. FIG. 3 is a diagram showing a specific example of affine transformation in a two-dimensional homogeneous coordinate system.

As shown in FIG. 3, paying attention to the vector VA at the point A, the vector VB at the point B is scaled n times, and the vector VA is further rotated by θ degrees around the origin (Z axis). The vector VC of C is obtained. This process is expressed as a matrix as follows.

[Equation 1] FIG. 4 is a diagram showing a specific example of affine transformation in a two-dimensional homogeneous coordinate system showing intermediate interpolation.

If the same number of frames is required for the scaling and the rotation processing at this time, (VA, VB), (V
The intermediate interpolation for B, VC) is as follows.

[0013]

[Equation 2] Here, VB 'and VC' are interpolation vectors.

In addition,

(Equation 3) Are conversion matrices based on the vectors VA and VB, respectively.

FIG. 5 is a flowchart showing the non-stochastic event linear interpolation processing. Hereinafter, the operation of the interframe interpolation device of FIG. 1 in the case of a non-stochastic event will be described with reference to FIG. In step S1, the vector data input unit 11 stores the frame number f of the frame required for interpolation.
_n and f _{n + 1} and the vector data VA and VB are input (hereinafter, this data set is represented by (VA, f
_n )). In step S2, the vector data VA and VB are stored in the vector data memory unit 13 as they are. In step S3, the affine transformation matrix calculation unit 14 causes the vector data memory unit 13
From (VA, f _n ) and (VB, f _{n + 1} ) are read out, and the conversion matrix T to the vector VB based on the vector VA
_{Find VAB} . The transformation matrix T _VAB is the vector VA and VB
Depending on the magnitude of the vector and the angle formed by the vector, scaling and rotation are combined. FIG. 5 shows a case of scaling as an example.

In step S4, the interpolation matrix T of each interpolation frame is proportionally distributed according to the number of interpolation frames.
Calculate _{VAB '} . FIG. 5 shows, as an example, the case of intermediate interpolation in which the number of interpolation frames is 1. The affine transformation matrix coefficient correcting unit 15 corrects each coefficient of the interpolation matrix T _{VAB ′} by a constant multiple according to the correction parameter value input from the correction parameter input unit 12. In step S5,
The affine transformation execution unit 18 reads the base vector VA from the vector data memory unit 13 to obtain the vector V
A is multiplied by the corrected interpolation matrix to obtain an interpolation vector, and the original vector VA and the interpolation vector are output to the output unit 1.
Output to 9. On the other hand, when dealing with probabilistic events, the entire interframe interpolator can be regarded as equivalent to the system [measurement system + observation system + filter].

FIG. 6 is a diagram showing a linear discrete-time stochastic system equivalent to the embodiment of the present invention. As shown in FIG.
In the linear discrete time stochastic system, the signal Vx is measured by the measurement system 31.
Is input together with noise, and these signals Vx and noise (noise newly generated in the measurement system is ignored) are observed as an observation value Vy by the observation system. This is filter 32
By using the observed value Vy as the input observed value and the signal Vx as the input signal value, the estimated value Vx ^* is obtained. The measurement system may be macroscopically non-linear, but in that case, if the local linearization method is used, the system itself can be treated as being linear. Further, the fact that the vector value of the shape data has no causality is understood as the whiteness of the system and observation noise. In the embodiment of the present invention, as the input signal value Vx, an affine transformation matrix of the original vector is regarded as a non-stochastic process and linearly interpolated by the proportional distribution method. If the signal and noise are regarded as stochastic processes in this way, the estimated solution is obtained by a method that minimizes the statistical expected value of the difference from the true value. That is, in this embodiment, the observed value vector Vy is the original vector and the estimated value Vx is
^The interpolation matrix is obtained by filtering by associating ^* with the interpolation matrix.

FIG. 7 is a flowchart of the interpolation process in the case of a stochastic event. Hereinafter, the interpolation process in the case of a stochastic event will be described with reference to FIG. 7. First, step S1
1, the number of original vectors (hereinafter referred to as the number of frames N) used when obtaining the interpolation matrix is designated. The number of frames N designated here is designated by, for example, the user according to the magnitude of variation between the original vectors, the frame memory size, and the like. The vector data memory unit 13 stores the vector data input unit 1 in step S12.
First, vector data (V _l , f _l ) of, for example, the frame immediately before the frame to be interpolated from 1 (hereinafter referred to as current) is input. In step S13, subsequently, vector data (V _{1 + m} , f _{1 + m} ) (m =) of (N−1) frames before and after the current frame.
± 1, ..., ± (N-1) / 2) are input.

FIG. 8 is a diagram showing a frame reading method for interpolation. After the first N frames have been read, when the vector data of the read frame for the current frame number n is n−N / 2 to n + N / 2, as shown in FIG. Then, it is assumed that the current frame number (n + 1) is slid one frame at a time such as (n + 1) -N / 2 to (n + 1) + N / 2. Further, the read frame overlaps with the interpolation process in the current frame and the interpolation process in the next frame, but the interpolation process in the current frame only interpolates the frame between the frame numbers n and (n + 1).

In step S14, the affine transformation matrix calculating unit 14 in FIG. 1 calculates an affine transformation matrix between the preceding and succeeding frames for the N frames. The affine transformation matrix calculated here is calculated by the number (N-1) between frames. Then, s affine transformation matrices are calculated by linear interpolation between each frame. In FIG. 7, the case where the intermediate interpolation (s = 1) is described, and for simplicity, N−1 intermediate interpolation matrices (Vx _{l + m} , f _{l + m} ) (m = 1, ... , N-1). In step S15, according to the correction parameter value input from the correction parameter input unit 12 in FIG. 1, the affine transformation matrix coefficient correction unit 15 calculates each coefficient of the affine transformation matrix (Vx _{l + m} , f _{l + m} ). A correction of a constant multiple is performed to obtain an affine transformation matrix (Vx ′ _{l + m} , f _{l + m} ). In step S16, the corrected affine transformation matrix (Vx ′ _{l + m} , f _{l + m} ) is temporarily stored in the transformation matrix memory unit 16.

The processing of the time series filter unit 17 will be described below. This is a method of calculating an interpolation matrix from an affine transformation matrix. At this time, the problem is that Vy _t = H _t * V 'x _t + v _t t = 0,1, ..., N as fixed interval smoothing.
-1 is formulated as (1). In the equation (1), t indicates time, and t = 0 is 0, which is the smallest frame number ( _fl ) actually read.
v _t is a noise component, and this noise component represents the difference between the vector V ′ x _t obtained by linear interpolation and the observation vector Vy _t , regarding the system as a non-stochastic process, and is a Gaussian white observation noise with an average value of 0. Treat as. H _t is an observation matrix. In each dimension of the equation (1), V ′ x _t is a p × p dimensional matrix, Vy _t
Is a 1 × p-dimensional observation vector, and H _t is a 1 × p-dimensional observation matrix.

FIG. 9 is a diagram showing an example of observation of time series data. FIG. 9 shows the observation result of the stochastic process of Brownian motion, and the coordinate of the starting point at time 0 is 0.
The horizontal axis represents time t, and the vertical axis represents the observation result y _t . Gaussian white noise is seen in this figure. Equation (1) is
When the input observation values y ₀ ..., y _N of a certain time section [0, N] are given, the optimum estimated value x ′ _{t / N} of the state x _i at any time point of the time section [0, N] is obtained. Although it is a problem, here, the Kalman filter is applied. The method is well known and will not be described in detail.

FIG. 10 is a diagram showing the principle of interpolation processing in the case of a stochastic event. In the figure, circles observations (corresponding to vector data), and the broken line the signal value of the non-stochastic process linear interpolation value of the observed value (corresponding to _{H t * V 'x t)} , the interpolation value solid is finally obtained Is. The t-axis corresponds to the frame, and the interpolation value is calculated by the method that minimizes the statistical expected value of the difference from the true value so that the average of the observation noise is 0 with respect to the broken line indicating the non-stochastic process linear interpolation value. Ask. In step S17 in FIG. 7, the linear interpolation affine transformation matrix V stored in the transformation matrix memory 16 by the time series filter unit 17 is used.
_{x'l + m} (m = ± 1, ..., ± (N−1) / 2) is an input signal value, and N vector data V _{l + m} (m = 0, ± 1,
, ± (N-1) / 2) is used as an input observation value and a Kalman filter is used to calculate an interpolation matrix of a frame between the current frame and the next frame. The affine transformation execution unit 18 multiplies the calculated interpolation matrix and the current vector data to obtain an interpolation vector. Then, the original vector data of the current frame and the interpolation vector are output from the output unit 19. In step S18, it is checked whether the frame is the final frame, and if it is not the final frame, the process returns to step S11 to repeat the interpolation process. If it is the last frame, the interpolation process is terminated. As described above, according to the present embodiment, the time-series filter unit 17 is provided to interpolate between frames in consideration of the noise component when the moving image is a stochastic event, so that more accurate interframe There is an advantage that interpolation can be performed.

FIG. 11 is a functional block diagram of the animation creating apparatus. With this animation creation device,
It has an input unit 41. The output side of the input unit 41 includes a shape management unit 42, a motion storage unit 43, and a control unit 44. The shape storage unit 45 is provided on the output side of the shape management unit 42.
Is connected. The shape storage unit 45, the motion storage unit 43,
The synthesis processing unit 46 is connected to the output side of the control unit 44, and the drawing unit 47 is connected to the output side of the synthesis processing unit 46. The operation of the animation device shown in FIG. 11 will be described below. The input unit 41 receives inputs such as the type of shape to be displayed, time change of the same shape, animation environment data, and operation instruction. In the shape management unit 42, the input unit 4
The shape input from 1 is decomposed into a plurality of constituent elements and registered in the shape storage unit 45. Data indicating the time change of the shape input from the input unit 41 is stored in the operation storage unit 43. In the synthesis processing unit 46, based on an instruction from the control unit 44, each component of the shape is extracted from the shape storage unit 45, and the position of the shape at each time is extracted from the operation storage unit 43 to change the relative position of the component. By doing so, the entire shape is deformed. The drawing unit 47 draws the data output by the composition processing unit 46. The control unit 44 controls all the above processes.

FIG. 12 is a functional block diagram showing a configuration example of an animation creating apparatus using the interframe interpolating apparatus of FIG. 1, and elements common to those of FIG. 3 are designated by common reference numerals. In this animation creation device,
The interpolation device 52 is provided on the output side of the operation storage unit 43. The interpolation device 52 is connected to the output side of the interpolation processing control unit 51. A rewrite processing section 53 is connected to the output sides of the interpolation processing control section 51 and the interpolation device 52, and an operation storage section 43 is connected to the output side of the rewrite processing section 53. The operation of the animation creating apparatus shown in FIG. 12 will be described below. The interpolation processing control unit 51 gives an interpolation instruction to the interpolation device 52 as necessary. The interpolation device 52 reads the data indicating the change in shape over time from the operation storage unit 43 in accordance with the interpolation instruction from the interpolation processing control unit 52, and performs the above-described interpolation processing. The interpolation processing control unit 51 instructs the rewriting processing unit 53 to rewrite the interpolation data. The rewriting processing unit 53 rewrites the data interpolated by the interpolation device 52 into the operation storage unit 43.
The synthesizing processing unit 46 takes out the respective constituent elements of the shape from the shape storage unit 45 and the position of the shape at each time from the operation storage unit 43, and changes the relative positions of the constituent elements to deform the entire shape.

As described above, according to this embodiment,
In the animation of stochastic events, there is an advantage that interpolation between frames can be performed more accurately. The present invention is not limited to the above embodiment, and various modifications can be made. The following are examples of such modifications.

(1) In the present embodiment, the non-stochastic process interpolated value is obtained by linearly interpolating the affine transformation matrix of the original Beccurt data by the proportional distribution method, but it can be obtained by the least square method. (2) In this embodiment, the Kalman filter is used as the time series filter, but a Wiener Helstromstrom filter may be used. Here, the Wiener Helstromstrom filter is the observation matrix H of the equation (1) on the Fourier plane.
_When the frequency filter corresponding to _t is H (ω) and the noise frequency spectrum is N (ω), H (ω) / (| H
It is a filter represented by (ω) | ² + | N (ω) | ² . (3) When the image in FIG. 12 is given in the form of [image constituent basic elements + affine transformation matrix that defines its temporal change], the interframe interpolation apparatus may use only the affine transformation matrix as an interpolation target. Good.

[0028]

As described in detail above, the first to seventh aspects
According to the invention, when the moving image is a stochastic event, the data defining the temporal change are input observation values, the affine transformation matrix calculated as a non-stochastic event is used as the input signal value, and the interpolation matrix is calculated by the time series filter. Since it is determined, inter-frame interpolation can be performed more accurately in the case of a stochastic event.

[Brief description of drawings]

FIG. 1 is a functional block diagram of an interframe interpolation device showing an embodiment of the present invention.

FIG. 2 is a functional block diagram of a conventional interframe interpolation device.

FIG. 3 is a diagram showing a specific example of affine transformation in a two-dimensional homogeneous coordinate system.

FIG. 4 is a diagram showing a specific example of affine transformation in a two-dimensional homogeneous coordinate system.

FIG. 5 is a flowchart showing non-stochastic linear interpolation processing.

FIG. 6 is a diagram showing a linear discrete-time stochastic system equivalent to the embodiment of the present invention.

FIG. 7 is a flowchart showing interpolation processing of a probabilistic event.

FIG. 8 is a diagram showing a method of reading a frame at the time of interpolation.

FIG. 9 is a diagram illustrating an observation example of time series data.

FIG. 10 is a diagram showing a principle of interpolation processing of a probabilistic event.

FIG. 11 is a functional block diagram of the animation creating device.

12 is a functional block diagram of an animation creating device using the interframe interpolation device of FIG. 1. FIG.

[Explanation of symbols]

 11 vector data input unit 12 correction parameter input unit 13 vector data memory unit 14 affine transformation matrix calculation unit 15 affine transformation matrix coefficient correction unit 16 transformation matrix memory unit 17 time series filter unit 18 affine transformation execution unit 41 input unit 42 shape management unit 43 operation storage unit 44 control unit 45 shape storage unit 51 interpolation processing control unit 52 interpolation device 53 rewriting processing unit

Claims (7)

[Claims] 1. A time change of an interpolated frame between frames of the first and second moving images is calculated based on data defining a time change of frames of the first and second moving images which are temporally continuous. Affine transformation matrix calculating means for calculating the affine transformation matrix to convert into the specified data by considering that the moving image is a non-probability event, and data of a plurality of moving image frames before and after the time to be interpolated. The input observation value, the matrix based on the affine transformation matrix calculated by the affine transformation matrix calculation means based on the data of each of the two frames preceding and succeeding temporally with respect to the frame data of the plurality of moving images is filtered as an input signal value. A time series filter for creating an interpolation matrix, and the interpolation matrix created by the time series filtering process and the frame of the moving image. Based on the data, the interpolation data producing means for producing interpolation data, inter-frame interpolation device characterized by comprising. 2. The interframe interpolation apparatus according to claim 1, wherein the time series filter is a Kalman filter or a Wiener Helstromstrom filter. 3. The interframe interpolation according to claim 1, wherein the time series filter is configured to handle a noise component due to a stochastic event in a moving image as Gaussian white observation noise having an average value of 0. apparatus. 4. The affine transformation matrix calculating means is configured to linearly interpolate by proportional distribution according to the number of interpolation frames between frames of the first and second moving images. Interframe interpolator. 5. The interframe interpolation device according to claim 1, wherein the data defining the temporal change of the first and second frames is an affine transformation matrix. 6. A temporal change of an interpolated frame between frames of the first and second moving images based on vector data defining temporal changes of frames of the first and second moving images which are temporally continuous. And an affine transformation matrix calculated by the affine transformation matrix calculation means for calculating the affine transformation matrix for converting the affine transformation matrix into linear data by regarding the moving image as a non-stochastic event. An affine transformation matrix coefficient correcting means for correcting the input image data of a plurality of moving image frames before and after the time to be interpolated when the moving image is a probabilistic event, the plurality of moving image frames Affine transformation row calculated by the affine transformation matrix coefficient correcting means on the basis of the data of each two frames that are temporally before and after the data. When the moving image is a non-stochastic event, an interpolation vector is corrected by the affine transformation matrix corrected by the affine transformation matrix correction means when the moving image is a non-stochastic event. An interframe interpolating device, comprising: affine transformation executing means for calculating and calculating an interpolation vector by an interpolation matrix obtained by the time series filter when the moving image is a stochastic event. 7. A shape storage unit that stores the shape of a moving object, an operation storage unit that stores the position of the moving object at each time, and a position at each time stored in the operation storage unit as vector data. 7. The interframe interpolating device according to claim 6, which interpolates the position of a moving object between preceding and following times, and a rewriting processing device that writes the position of the shape interpolated by the interframe interpolating device into the motion storage device, Synthesis processing means for obtaining the shape of the object at each time and at the interpolated time based on the position of the shape stored in the motion storage means and the shape of the object stored by the shape storage means. Animation creation device characterized by.