JP4245587B2 - Motion compensation prediction method - Google Patents

Description

  The present invention relates to a motion-compensated prediction method, and more specifically, in the generation of a prediction image, a first prediction mode in which a prediction image is generated by affine transformation or motion vector interpolation, and a prediction image is generated by parallel movement. The present invention relates to a motion-compensated prediction method in which two prediction modes are obtained, and any of these prediction modes is selected for each block in accordance with prediction mode information to improve prediction efficiency. For example, the present invention is applied to high-efficiency encoding of image data in digital image processing.

  As a method of inter-frame prediction, a prediction method using affine transformation has been studied conventionally. For example, in the methods described in Non-Patent Documents 1 and 2, as shown in FIG. 12, the motion of an image is expressed by lattice deformation, and a reference image (usually locally decoded by an encoding device and stored in a frame memory). The predicted image is created by geometric conversion from the image that has already been decoded by the decoding device and stored in the frame memory. Here, ● represents a lattice point (vertex of a rectangular block).

FIG. 13 is a configuration diagram of a predicted image generation circuit in a conventional image encoding device, in which 32 is a rectangular block dividing unit, 33 is a lattice point motion estimation unit, and 34 is a motion compensation prediction unit.
In the rectangular block dividing unit 32, the reference image is divided into squares. When dividing, for example, a method of dividing by a uniform grid such as a square grid regardless of the reference image is used. Alternatively, a method of dividing by a deformation grid in accordance with the edge on the reference image may be used. The lattice point motion estimation unit 33 estimates the motion of the lattice points of the reference image and obtains a motion vector of each lattice point. A new deformation grid is defined on the predicted image by this motion vector.

  The method shown in FIG. 13 is called “backward motion estimation”. First, a grid on the reference image is determined, and then a grid on the prediction target image is determined. On the other hand, a technique called “forward motion estimation” may be used. First, a lattice (usually a uniform lattice) is defined on the prediction target image, the motion of the lattice point is estimated, and a deformed lattice is newly defined on the reference image by a motion vector of the lattice point. Both the conventional example and the present invention can be used in both cases, but for the sake of simplicity, the description will be made with “backward motion estimation” as an example.

Next, a case where it is estimated at which position the grid point on the reference image has moved on the original image (in the case of “backward motion estimation”) will be described.
As a method of motion estimation, a method of taking an area consisting of a lattice point on the reference image and its neighboring pixels and checking which area on the original image matches this area (referred to as “block matching”) is used. It is done. Specifically, considering an area of M pixels × N pixels centering on a lattice point on the reference image, the degree of coincidence with an area of the same size on the original image is examined, and the center of the region with the best degree of coincidence is The destination of the point. A vector representing the movement at this time is called a motion vector. As the degree of coincidence of regions, an absolute value sum of errors of pixel values in the region or an absolute value weighted sum of errors is used. Note that the motion vector of the lattice point obtained by the lattice point motion estimation unit 33 is encoded by an encoding unit (not shown), and is incorporated into encoded data for transmission or accumulation.

  The motion compensated prediction unit 34 obtains a predicted image using the lattice points on the reference image, its motion vector, and the reference image. As a method for obtaining a predicted image, there are an affine transformation using a triangular block, a motion vector interpolation in a rectangular block, and the like. These are described in detail in Non-Patent Document 1.

  In the encoding device, difference data between the predicted image and the original image obtained by the motion compensation prediction unit 34 is encoded by an encoding unit (not shown), and is transmitted or stored by being incorporated into the encoded data. Also, the encoded data is decoded by a decoding unit (not shown) and added to the predicted image to obtain a decoded image. The decoded image is stored in the frame memory and used as a reference image for subsequent encoding.

Next, a conventional image decoding apparatus will be described.
FIG. 14 is a configuration diagram of a predicted image generation circuit in a conventional image decoding apparatus, in which 41 is a lattice point motion vector decoding unit, 42 is a motion compensation prediction unit, and 43 is a division unit into rectangular blocks.
The lattice point motion vector decoding unit 41 is a part that decodes the code of the motion vector of the lattice point incorporated in the encoded data. The output is a motion vector of lattice points, and is supplied to the motion compensation prediction unit 42. Since the rectangular block dividing unit 43 and the motion compensation prediction unit 42 function in the same manner as the encoding apparatus in FIG. 13, description thereof is omitted.

In the decoding apparatus, the same reference image as that in the encoding apparatus is obtained, and the motion vector obtained in the lattice point motion vector decoding unit 41 is the same as that obtained in the encoding apparatus. Therefore, the prediction image obtained by the motion compensation prediction unit 42 is also the same as that obtained by the encoding device. A predicted image obtained by the motion compensation prediction unit 34 and a decoded value of difference data obtained by a decoding unit (not shown) are added to obtain a decoded image. The decoded image is displayed on a display or the like and stored in a frame memory (not shown), and is used as a reference image for subsequent decoding.
“Very Low Bitrate Video Coder using Warping Prediction” (1993 Image Coding Symposium 8-7, pp.167-168) “A Novel Video Coding Scheme Based on Temporal Prediction Using Digital Image Warping” (IEEE International Conference on Consumer Electronics, 1993)

FIGS. 15A and 15B and FIGS. 16A to 16C are diagrams for explaining conventional problems.
FIG. 15A represents a part of the reference image, and FIG. 15B represents a part of the prediction target image. As shown in FIG. 15, a rectangular object moves to the left while rotating counterclockwise against an oblique stripe pattern as a background. ● represents a grid point. Here, when attention is paid to two blocks (rectangle abdc and rectangle befd) formed from the lattice points a, b, c, d, e, and f on the reference image, the rectangle abdc on the stationary background is greatly deformed, It is a square ABDC. These two blocks are taken out and shown in FIGS.

  16A shows the two blocks of interest in FIG. 15A, and FIG. 16B shows the two blocks of interest in FIG. 15B. FIG. 16C shows a case where a predicted image is created by the conventional method in such a case. It can be seen that the diagonal pattern of the left block is deformed by geometric transformation and is different from the prediction target image (original image) shown in FIG.

  In general, the shape of a block on a background image in contact with an animal body changes with the movement of the animal body. According to the conventional method, although the background image is stationary or moved in parallel, the predicted image is created by geometric transformation such as deformation and rotation. For this reason, there existed a problem that prediction efficiency fell. Here, the parallel movement means that the movement of each pixel is parallel.

FIGS. 17A to 17C are diagrams for explaining other conventional problems.
17A represents a part of the reference image, and FIG. 17B represents a part of the prediction target image. As shown in FIG. 17, the object 1 is moving in the lower left direction and the object 2 is moving in the upper right direction. Here, attention is focused on two blocks formed from lattice points a, b, c, d, e, and f on the reference image. Among these lattice points, c and e move to C and E on the prediction target image due to the movement of the object, but a, b, d, and f do not move.
In such a case, according to the conventional method, the result is as shown in FIG. Although the object is moving in parallel, it is distorted due to geometric transformation. Generally, when a plurality of objects on adjacent blocks are translated in different directions, the geometric distortion as described above occurs.

  The present invention has been made in view of such a situation, and relates to a motion compensation prediction method. More specifically, in the creation of a prediction image, the first prediction for creating a prediction image by affine transformation or motion vector interpolation. Prediction efficiency is obtained by obtaining a mode and a second prediction mode for creating a prediction image by parallel movement, and selecting the prediction mode for each block according to prediction mode information indicating which of these prediction modes is selected. For example, it is suitable for high-efficiency encoding of image data in digital image processing.

The present invention is, motion compensation prediction and a prediction step of obtaining a selection step of selecting one of the prediction mode used in the motion compensated prediction, the predicted image by motion compensated prediction for each block obtained by dividing an image In the method, the selection step selects the prediction mode for each block according to prediction mode information indicating which of the prediction modes is selected, and the prediction step includes a prediction image by affine transformation or motion vector interpolation. a first prediction mode for creating, in which was characterized by have a second prediction mode for creating prediction image by translational movement.

  In the prediction step of obtaining a predicted image by motion compensation prediction for each block obtained by dividing an image, a first prediction mode for creating a predicted image by affine transformation or motion vector interpolation, and a predicted image by parallel movement Since the second prediction mode to be created is provided and which one of these prediction modes is selected is selected for each block in accordance with the prediction mode information, the prediction efficiency can be improved.

Embodiments will be described below with reference to the drawings.
First, the concept of the present invention will be described with reference to FIG. FIG. 11 shows how a predicted image (for one block) of a block of interest (rectangular ABDC) on the predicted image is obtained by the method of the present invention. Similar to the conventional method, the quadrangle abdc on the reference image is deformed using geometric transformation by deformation of the lattice, and a predicted image such as P1 is obtained. On the other hand, another predicted image P2 is obtained by parallel movement of the rectangle a′b′d′c ′ on the reference image. In the present invention, the plurality of predicted images are adaptively selected for each block, and one of them is used as an actual predicted image. Here, the parallel movement means that all the pixels in the block move in parallel, and means that only the position moves without changing the shape of the block.

FIG. 16D is an example of a predicted image created according to the present invention. As a prediction method for the left block, there are a method of geometrically transforming the quadrangle abdc of FIG. 16A and a method of moving in parallel the portion surrounded by the dotted line of FIG. In this case, the latter is adaptively selected to obtain the predicted image of FIG. As can be seen from the figure, the geometric distortion that was a problem of the conventional method does not occur.
Moreover, FIG.17 (d) is another example of the estimated image produced by this invention. By moving in parallel the portion surrounded by the dotted line in FIG. 17A as the predicted image of the left block, and by moving in parallel the portion surrounded by the dashed line in FIG. 17A as the predicted image of the right block. A predicted image is obtained. As can be seen from the figure, the geometric distortion (see FIG. 17C), which was a problem of the conventional method, does not occur.

FIG. 1 is a block diagram for explaining an example of an image encoding apparatus to which the present invention is applied. In the figure, 1 is a dividing unit into quadrangular blocks, 2 is a lattice point motion estimating unit, and 3 is parallel motion. An estimation unit, 4 is a first motion compensation prediction unit, 5 is a second motion compensation prediction unit, and 6 is a predicted image selection unit.
The quadrangular block dividing unit 1 and the lattice point motion estimating unit 2 are units that divide the reference image into quadrilaterals and obtain the motion vectors of the lattice points by the same operation as described in FIG. The motion vector of the lattice point is supplied to the first motion compensation prediction unit 4. In addition, although the division | segmentation part 1 to a rectangular block is used also in the Example of another encoding apparatus, it is not shown in subsequent FIGS. 4-6, and abbreviate | omits description.

  The parallel movement motion estimation unit 3 is a part that obtains the movement direction of the block on the reference image by a technique such as block matching. The block matching described above is to check which region of the same shape on the prediction target image matches the region around the lattice point. However, here, the block itself is regarded as a region, and it is examined which region of the same shape on the prediction target image matches this region. Thereby, a motion vector (parallel movement vector) when the block itself moves in parallel is obtained. The motion vector of the parallel movement is supplied to the second motion compensation prediction unit 5.

The first motion compensation prediction unit 4 performs motion compensation prediction by geometric transformation by the same function as the motion compensation prediction unit 34 described with reference to FIG. The output predicted image is supplied to the predicted image selection unit 6. The second motion compensation prediction unit 5 performs motion compensation prediction by parallel movement. The output predicted image is supplied to the predicted image selection unit 6.
The predicted image selection unit 6 adaptively selects the predicted image for each block by the method of the present invention, and outputs the predicted image and the prediction mode information. The prediction mode information is encoded by an encoding unit (not shown), and is transmitted or stored by being incorporated in encoded data. As a method for selecting a predicted image, a method for minimizing an error from the original image or a method for minimizing a weighted error from the original image can be considered. Hereinafter, these will be described.

FIG. 2 is a configuration diagram illustrating an example of the predicted image selection unit in FIG. 1, in which 7 is a first error calculation unit, 8 is a second error calculation unit, and 9 is a selection unit.
Here, there are two types of predicted images, but of course this is not restrictive. The first error calculation unit 7 is a part that inputs the data of the original image and the predicted image for one block and calculates the prediction error. The prediction error is calculated using the pixel value f (n) of the original image and the pixel value f (n) ′ of the prediction image,

Calculated by Here, n is the number of pixels in the block, and N is the number of pixels in the block. Although the prediction error due to the square error is shown here, an absolute value error may be used. The second error calculation unit 8 is a part that calculates a prediction error by the same function as the first error calculation unit 7. The selection unit 9 compares each prediction error and outputs a prediction image that minimizes the prediction error and prediction mode information. For example, if the error of the second error calculation unit 8 is minimum, the predicted image (2) is selected and output. Moreover, the information showing a prediction image (2) is output as prediction mode information. When there are two types of prediction methods, for example, if the prediction mode information is represented by a fixed-length code, one bit is required per block.

FIG. 3 is a configuration diagram illustrating another example of the predicted image selection unit in FIG. 1, in which 10 is a first orthogonal transform unit, 11 is a second orthogonal transform unit, and 12 is a third orthogonal transform unit. The other parts having the same functions as those in FIG.
The first orthogonal transformation unit 10 to the third orthogonal transformation unit 12 perform two-dimensional orthogonal transformation on the input pixel value for one block and convert it to a frequency component. The first error calculator 7 and the second error calculator 8 in FIG. 3 are weighted errors (prediction errors) between the orthogonal transform coefficient g (k) of the original image and the orthogonal transform coefficient g (k) ′ of the predicted image. ) Is calculated as follows.

  Here, k is the number of the orthogonal transform coefficient, N is the number of pixels in the block, and w (k) is a predetermined weight. Although the prediction error due to the square error is shown here, an absolute value error may be used. For example, by setting the weight w (k) at a low frequency and small at a high frequency, a prediction error that matches human visual characteristics is calculated. The selection unit 9 in FIG. 3 selects a prediction image by the same operation as the selection unit 9 in FIG. 2, and outputs a prediction image and prediction mode information.

  As described above, a plurality of motion compensation prediction methods can be switched by the encoding apparatus shown in FIG. That is, the motion prediction method using geometric transformation and the motion prediction method using parallel movement can be adaptively switched for each block. As a result, the prediction efficiency can be improved and the encoding efficiency can be improved.

FIG. 4 is a block diagram for explaining another example of an image coding apparatus to which the present invention is applied. In the figure, 13 is a third motion compensation prediction unit, 14 is a fourth motion compensation prediction unit, Reference numeral 15 denotes a fifth motion compensation prediction unit, and other parts having the same functions as those in FIG. 1 are denoted by the same reference numerals.
The lattice point motion estimation unit 2 is a part for obtaining a motion vector of a lattice point by the same function as described in FIG. The motion vector of the lattice point is supplied to the first motion compensation prediction unit 4 to the fifth motion compensation prediction unit 15. The first motion compensation prediction unit 4 performs motion compensation prediction by geometric transformation by the same function as the motion compensation prediction unit 34 described with reference to FIG. The output predicted image is supplied to the predicted image selection unit 6.

  The second motion compensation prediction unit 5 to the fifth motion compensation prediction unit 15 perform motion compensation prediction by parallel movement. The motion vector used for prediction is a motion vector of four lattice points of the block of interest. Assuming that the motion vectors at the respective grid points are V1, V2, V3, and V4, the second motion compensation prediction unit 5 uses the motion vector V1, the third motion compensation prediction unit 13 uses the motion vector V2, and the fourth motion. The compensation prediction unit 14 uses the motion vector V3, and the fifth motion compensation prediction unit 15 uses the motion vector V4 to perform motion compensation prediction by parallel movement. The created predicted images are respectively supplied to the predicted image selection unit 6. The predicted image selection unit 6 adaptively selects a predicted image by the method of the present invention, and outputs a predicted image and prediction mode information. Specifically, the method already described (FIGS. 2 and 3) is used. However, here, there are five types of predicted images.

  As described above, a plurality of motion compensation prediction methods can be switched by the encoding apparatus shown in FIG. That is, the motion prediction method using geometric transformation and the four motion prediction methods using parallel movement can be adaptively switched for each block. As a result, prediction efficiency can be improved and encoding efficiency can be improved. In addition, since it is not necessary to encode a motion vector indicating parallel movement of blocks, the amount of encoded data can be reduced.

  FIG. 5 is a block diagram for explaining another embodiment of the image coding apparatus to which the present invention is applied. In FIG. 5, reference numeral 16 denotes a parallel movement vector determination unit, and the other parts having the same operation as FIG. Are given the same reference numerals. The difference from FIG. 4 is a portion (parallel movement vector determination unit 16 and second motion compensation prediction unit 5) that performs motion compensation prediction by parallel movement.

In the parallel movement vector determination unit 16, a motion vector representing the parallel movement of the block itself based on the motion vectors of the four lattice points belonging to one block and the reference image (both are information obtained by the decoding device). To decide. The parallel movement vector obtained here is supplied to the second motion compensation prediction unit 5. For example, a parallel movement vector for obtaining an average value of four lattice point motion vectors is used. In this case, since no reference image information is used, no reference image is input to the parallel movement vector determination unit 16.
As in the embodiment of FIG. 4, the embodiment of FIG. 5 also uses a motion prediction method using geometric transformation (motion compensation prediction unit (1)) and a motion prediction method using parallel movement (motion compensation prediction unit (2)). Can be adaptively switched for each block. As a result, prediction efficiency can be improved and encoding efficiency can be improved. In addition, since it is not necessary to encode a motion vector indicating parallel movement of blocks, the amount of encoded data can be reduced.
Furthermore, by using the average value of the four motion vectors used for geometric conversion as the motion vector that represents the parallel movement of the block itself, the motion vector that represents the parallel movement can be predicted more accurately, and the amount of code can be reduced. It becomes possible to do.
In addition, since a motion vector representing parallel movement is obtained without newly performing block matching processing, the processing amount of the encoding device can be reduced.

Alternatively, the following methods can be considered as other parallel movement vector determination methods. That is, the parallel movement vector of the block of interest from the reference image one frame before and the reference image two frames before is searched by block matching as described in the description of the embodiment shown in FIG. The motion vector obtained here is a parallel movement vector to be obtained. Alternatively, the four lattice point motion vectors and the average vector thereof may be compared with the searched motion vector to obtain a parallel motion vector for obtaining the closest one.
In this case, since a vector close to a vector actually searched by block matching can be used as a parallel movement vector, prediction efficiency can be improved and encoding efficiency can be improved. Further, although information for indicating the nearest vector needs to be transmitted as additional information, it is not necessary to directly transmit the parallel movement vector itself, so that the amount of encoded data can be reduced.

  The second motion compensation prediction unit 5 in FIG. 5 performs motion compensation prediction by parallel movement based on the input motion vector. The predicted image selection unit 6 selects one of two types of predicted images with a small prediction error, and outputs the selected predicted image and prediction mode information. The selection method is the same as that shown in FIGS. According to the encoding apparatus of the present embodiment, the prediction mode information requires only 1 bit per block, so that the amount of data related to the prediction mode information can be reduced.

  FIG. 6 is a block diagram for explaining another embodiment of the image coding apparatus to which the present invention is applied. In the figure, 17 is a control unit, 18 is a prediction mode selection unit, and others are the same as FIG. Parts that act are given the same reference numerals. The difference from FIG. 5 is that the prediction image selection unit 6 is eliminated, and a prediction mode selection unit 18, a control unit 17, a switch (1), and a switch (2) are provided instead.

  The prediction mode selection unit 18 determines a prediction mode based on information obtained by the decoding device such as a motion vector and a reference image. It is assumed that mode 0 represents motion compensation prediction by geometric transformation, and mode 1 represents motion compensation prediction by parallel movement. The determined prediction mode is supplied to the control unit 17.

  For example, it is possible to know the change in the size of the block of interest from the motion vector. When a block changes greatly, it is often caused by the movement of an adjacent block, and the block of interest itself is often stationary or in parallel, so a mode for motion compensation prediction by parallel movement is selected. In other words, the change in the number of pixels included in the block is examined, and if this change is greater than the threshold value, the motion compensation prediction mode by parallel movement is selected, otherwise the motion compensation prediction mode by geometric transformation is selected. To do.

Alternatively, the following may be considered as another prediction mode selection method.
That is, the motion vector of the block of interest is searched by block matching from the reference image one frame before and the reference image two frames before. If the minimum error of block matching at this time is smaller than a threshold value, a motion compensation prediction mode by parallel movement is selected, and a motion compensation prediction mode by geometric transformation is selected in other cases. Note that the threshold value described here is determined theoretically or empirically. These may be determined in advance, or may be changed as appropriate manually or automatically.

  The control unit 17 controls the switch (1) and the switch (2) based on the prediction mode determined as described above. That is, when the prediction mode is 0, the motion compensation prediction by geometric transformation is performed with the switch (1) and the switch (2) on the first motion compensation prediction unit 4 side. When the prediction mode is 1, motion compensation prediction by parallel movement is performed with the switch (1) and the switch (2) on the second motion compensation prediction unit 5 side. Thus, according to the encoding apparatus of the present embodiment, the prediction mode is determined based on the information obtained also by the decoding apparatus, so that it is not necessary to encode the prediction mode information, and the data amount can be reduced.

  FIG. 7 is a block diagram for explaining an example of an image decoding apparatus to which the present invention is applied. In the figure, 21 is a control unit, 22 is a lattice motion vector decoding unit, 23 is a parallel movement vector decoding unit, 24 Is a rectangular block dividing unit, 25 is a first motion compensation prediction unit, and 26 is a second motion compensation prediction unit.

  The lattice point motion vector decoding unit 22 is a part that decodes a lattice point motion vector from the encoded data by the same function as described in FIG. The parallel movement vector decoding unit 23 is a part that decodes the parallel movement vector from the encoded data. The rectangular block dividing unit 24 is a part that divides the reference image into squares by the same function as described in FIG. Note that the rectangular block dividing unit 24 is also used in other embodiments of the decoding device, but is not shown in FIGS. 8 to 10 and will not be described.

  The control unit 21 controls the switch based on the prediction mode information in the encoded data. That is, when the prediction mode information represents motion compensation prediction by geometric transformation, the switch is on the first motion compensation prediction unit 25 side, and when the prediction mode information represents motion compensation prediction by parallel movement, the switch is set to the second motion compensation prediction. Switch to the motion compensation prediction unit 26 side. The motions of the first motion compensation prediction unit 25 and the second motion compensation prediction unit 26 are exactly the same as those of the encoding device shown in FIG.

  As described above, in the decoding apparatus shown in FIG. 7, the same prediction mode as that of the encoding apparatus is selected, and the same predicted image is obtained. The predicted image and the decoded value of the difference data obtained by a decoding unit (not shown) are added to obtain a decoded image. The decoded image is displayed on a display or the like and stored in a frame memory (not shown), and is used as a reference image for subsequent decoding.

  FIG. 8 is a block diagram for explaining another example of an image decoding apparatus to which the present invention is applied. In the figure, 27 is a third motion compensation prediction unit, 28 is a fourth motion compensation prediction unit, 29 Is a fifth motion compensation prediction unit, and other parts having the same functions as those in FIG. 7 are denoted by the same reference numerals. The present embodiment is a decoding apparatus that decodes encoded data created by the encoding apparatus shown in FIG.

  The lattice point motion vector decoding unit 22 is a part that decodes a lattice point motion vector from the encoded data by the same function as described in FIG. The control unit 21 controls the switch (1) and the switch (2) based on the prediction mode information in the encoded data by the method of the present invention. That is, when the prediction mode information represents motion compensation prediction by geometric transformation, the switch (1) and the switch (2) are on the first motion compensation prediction unit 25 side, and the prediction mode information uses the motion vector V1. When representing motion compensation prediction by parallel movement, the switches are switched such that the switches (1) and (2) are on the second motion compensation prediction unit 26 side. Switching from the third motion compensation prediction unit 27 to the fifth motion compensation prediction unit 29 is performed in the same manner.

  The functions of the first motion compensation prediction unit 25 to the fifth motion compensation prediction unit 29 are exactly the same as those of the encoding apparatus shown in FIG. As described above, in the decoding apparatus shown in FIG. 8, the same prediction mode as that of the encoding apparatus is selected, and the same predicted image is obtained. The subsequent operation is the same as in the example of FIG.

  FIG. 9 is a block diagram for explaining still another example of the image decoding apparatus to which the present invention is applied. In FIG. 9, reference numeral 30 denotes a parallel movement vector determination unit, and other parts that perform the same operations as in FIG. The same reference numerals are given. This example is a decoding apparatus that decodes encoded data created by the encoding apparatus shown in FIG.

  The lattice point motion vector decoding unit 22 is a part that decodes a lattice point motion vector from the encoded data by the same function as described in FIG. The control unit 21 controls the switch (1) and the switch (2) based on the prediction mode information in the encoded data by the method of the present invention. That is, when the prediction mode information represents motion compensation prediction by geometric transformation, the switch (1) and the switch (2) are on the first motion compensation prediction unit 25 side, and the prediction mode information is motion compensated prediction by parallel movement. , The switch (1) and the switch (2) are on the second motion compensation prediction unit 26 side.

The parallel movement vector determination unit 30 determines the parallel movement vector used in the second motion compensation prediction unit 26 by the same function as that of FIG. The functions of the first motion compensation prediction unit 25 and the second motion compensation prediction unit 26 are exactly the same as those of the encoding apparatus shown in FIG.
As described above, in the decoding apparatus shown in FIG. 9, the same prediction mode as that of the encoding apparatus is selected, and the same predicted image is obtained. The subsequent operation is the same as in the example of FIG.

  FIG. 10 is a block diagram for explaining still another example of an image decoding apparatus to which the present invention is applied. In the figure, 31 is a prediction mode selection unit, and the other parts having the same functions as those in FIG. The code | symbol is attached | subjected. This example is a decoding apparatus that decodes encoded data created by the encoding apparatus shown in FIG.

  The lattice point motion vector decoding unit 22 is a part that decodes a lattice point motion vector from the encoded data by the same function as described in FIG. The prediction mode selection unit 31 determines a prediction mode based on information such as a motion vector and a reference image by the same operation as that of FIG. The determined prediction mode is supplied to the control unit 21.

The control unit 21 controls the switch (1) and the switch (2) based on the prediction mode information in the encoded data by the same operation as that of FIG. The parallel movement vector determination unit 30 determines the parallel movement vector used by the second motion compensation prediction unit 26 by the same function as that of FIG. The functions of the first motion compensation prediction unit 25 and the second motion compensation prediction unit 26 are exactly the same as those of the encoding apparatus shown in FIG.
As described above, in the decoding apparatus shown in FIG. 10, the same prediction mode as that of the encoding apparatus is selected, and the same predicted image is obtained. The subsequent operation is the same as in the example of FIG.

  The example of the encoding device and the decoding device described here mainly obtains a motion vector in only one direction in time, and thereby performs motion compensation prediction. Motion compensation prediction may be performed by obtaining a motion vector (from previous and subsequent reference images). In the prediction by geometric transformation, the problem as shown in FIG. 16C is not solved even if bidirectional motion compensation prediction is used. By adaptively switching between motion compensation by parallel movement and motion compensation by geometric transformation using the present invention, even when multiple objects with different motions overlap or when functions such as rotation and deformation are included A good predicted image can be obtained.

It is a block diagram for demonstrating an example of the image coding apparatus with which this invention is applied. It is a block diagram which shows an example of the estimated image selection part in FIG. It is a block diagram which shows the other example of the estimated image selection part in FIG. It is a block diagram for demonstrating the other example of the image coding apparatus with which this invention is applied. It is a block diagram for demonstrating the other example of the image coding apparatus with which this invention is applied. It is a block diagram for demonstrating the other example of the image coding apparatus with which this invention is applied. It is a block diagram for demonstrating an example of the image decoding apparatus with which this invention is applied. It is a block diagram for demonstrating the other example of the image decoding apparatus with which this invention is applied. It is a block diagram for demonstrating the further another example of the image decoding apparatus with which this invention is applied. It is a block diagram for demonstrating the further another example of the image decoding apparatus with which this invention is applied. It is a figure for demonstrating motion compensation prediction. It is a figure for demonstrating the motion compensation prediction by the conventional geometric transformation. It is a block diagram of the estimated image production circuit in the conventional image coding apparatus. It is a block diagram of the estimated image production circuit in the conventional image decoding apparatus. It is a figure which shows the example of a reference image and a prediction object image. It is a figure which shows the comparative example of the prediction image by the conventional this invention. It is a figure which shows the other comparative example of the prediction image by the past and this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Dividing part into square blocks, 2 ... Lattice point motion estimation part, 3 ... Parallel movement motion estimation part, 4 ... 1st motion compensation prediction part, 5 ... 2nd motion compensation prediction part, 6 ... Predictive image selection , 7... First error calculation unit, 8... Second error calculation unit, 9... Selection unit, 10... First orthogonal transformation unit, 11. , 13 ... third motion compensation prediction unit, 14 ... fourth motion compensation prediction unit, 15 ... fifth motion compensation prediction unit, 16 ... parallel motion vector determination unit, 17 ... control unit, 18 ... prediction mode selection , 21 ... control unit, 22 ... lattice motion vector decoding unit, 23 ... parallel motion vector decoding unit, 24 ... dividing unit into quadrangular blocks, 25 ... first motion compensation prediction unit, 26 ... second motion compensation Prediction unit, 27 ... third motion compensation prediction unit, 28 ... fourth motion compensation prediction unit, 29 ... fifth motion compensation Measuring unit, 30 ... parallel movement vector determination unit, 31 ... prediction mode selection unit.

Claims (1)

There the selection step and the motion compensated prediction method having a prediction step of obtaining a predicted image by motion compensated prediction for each block obtained by dividing an image to select one of the prediction modes used in the motion compensated prediction And
The selection step selects the prediction mode for each block according to prediction mode information indicating which of the prediction modes is selected, and the prediction step creates a prediction image by affine transformation or motion vector interpolation. prediction mode and the motion compensation prediction method characterized by have a second prediction mode for creating prediction image by parallel movement of the.

Images