- in a case that a predicted value derivation mode corresponding to any first block is a first derivation mode, performing inter prediction on the first block by using the first derivation mode;
- in a case that a predicted value derivation mode corresponding to any first block is a second derivation mode, performing inter prediction on the first block by using the second derivation mode; and
- in a case that a predicted value derivation mode corresponding to any first block is a third derivation mode, performing inter prediction on the first block by using the third derivation mode.

In this embodiment, the predicted value derivation mode corresponding to each first block in the target picture frame may be determined based on only the predicted value derivation mode corresponding to each first block in the target picture frame, that is, the second enable identifier.

In this embodiment, in a case that a predicted value derivation mode corresponding to a first block is a first derivation mode, it is determined that the predicted value derivation mode corresponding to the first block is the first derivation mode; in a case that a predicted value derivation mode corresponding to a first block is a second derivation mode, it is determined that the predicted value derivation mode corresponding to the first block is the second derivation mode; and in a case that a predicted value derivation mode corresponding to a first block is a third derivation mode, it is determined that the predicted value derivation mode corresponding to the first block is the third derivation mode.

In this embodiment, after the predicted value derivation mode corresponding to each first block is determined, inter prediction is performed on the first block by using the predicted value derivation mode.

For example, the following is set: The second enable identifier being 0 indicates the first derivation mode, the second enable identifier being 1 indicates the second derivation mode, and the second enable identifier being 2 indicates the third derivation mode. In this case, if a second enable identifier corresponding to a first block is 0, it is determined that a predicted value derivation mode corresponding to the first block is the first derivation mode; if a second enable identifier corresponding to a first block is 1, it is determined that a predicted value derivation mode corresponding to the first block is the second derivation mode; or if a second enable identifier corresponding to a first block is 2, it is determined that a predicted value derivation mode corresponding to the first block is the third derivation mode.

For example, the following is set: The second enable identifier being 0 indicates the third derivation mode, and the second enable identifier being 1 indicates the first derivation mode. In this case, if a second enable identifier corresponding to a first block is 0, it is determined that a predicted value derivation mode corresponding to the first block is the third derivation mode; or if a second enable identifier corresponding to a first block is 1, it is determined that a predicted value derivation mode corresponding to the first block is the first derivation mode.

For example, the following is set: The second enable identifier being 0 indicates the first derivation mode, and the second enable identifier being 1 indicates the second derivation mode. In this case, if a second enable identifier corresponding to a first block is 0, it is determined that a predicted value derivation mode corresponding to the first block is the first derivation mode; or if a second enable identifier corresponding to a first block is 1, it is determined that a predicted value derivation mode corresponding to the first block is the third derivation mode.

In this embodiment, the predicted value derivation mode corresponding to each first block is obtained, and then inter prediction is performed on each first block based on the predicted value derivation mode corresponding to each first block. In this way, for all first blocks included in a block, inter prediction may be performed by using a plurality of predicted value derivation modes, so that an inter prediction mode for a block is more flexible and accuracy of a predicted value of a block is improved.

Optionally, the performing inter prediction on each first block in the target picture frame based on the target information includes:

- in a case that the predicted value derivation mode corresponding to the target picture frame is a third derivation mode, performing inter prediction on each first block by using the third derivation mode; or
- in a case that the predicted value derivation mode corresponding to the target picture frame is not a third derivation mode, performing inter prediction on each first block based on the predicted value derivation mode corresponding to each first block in the target picture frame.

In this embodiment, the predicted value derivation mode corresponding to each first block may be determined based on the predicted value derivation mode corresponding to the target picture frame and the predicted value derivation mode corresponding to each first block in the target picture frame. To be specific, the predicted value derivation mode corresponding to each first block in the target picture frame is determined based on the first enable identifier and the second enable identifier.

In this embodiment, in a case that the predicted value derivation mode corresponding to the target picture frame is the third derivation mode, it is determined that the predicted value derivation mode corresponding to each first block in the target picture frame is the third derivation mode; or in a case that the predicted value derivation mode corresponding to the target picture frame is not the third derivation mode, the predicted value derivation mode corresponding to each first block in the target picture frame is obtained, to determine the predicted value derivation mode corresponding to each first block.

In this embodiment, after the predicted value derivation mode corresponding to each first block is determined, inter prediction is performed on the first block by using the predicted value derivation mode.

For example, the following is set: The first enable identifier being 0 indicates the third derivation mode, the first enable identifier being 1 indicates a non-third derivation mode, the second enable identifier being 0 indicates the first derivation mode, the second enable identifier being 1 indicates the second derivation mode, and the second enable identifier being 2 indicates the third derivation mode. In this case, if the first enable identifier is 0, it is determined that the predicted value derivation mode corresponding to each first block in the target picture frame is the third derivation mode; if the first enable identifier is 1 and the second enable identifier is 0, it is determined that the predicted value derivation mode corresponding to each first block is the first derivation mode; or if the first enable identifier is 1 and the second enable identifier is 1, it is determined that the predicted value derivation mode corresponding to each first block is the second derivation mode.

For example, the following is set: The first enable identifier being 0 indicates the third derivation mode, the first enable identifier being 1 indicates a non-third derivation mode, the second enable identifier being 0 indicates the third derivation mode, and the second enable identifier being 1 indicates the first derivation mode. In this case, if the first enable identifier is 0, it is determined that the predicted value derivation mode corresponding to each first block in the target picture frame is the third derivation mode; if the first enable identifier is 1 and the second enable identifier is 0, it is determined that the predicted value derivation mode corresponding to each first block is the third derivation mode; or if the first enable identifier is 1 and the second enable identifier is 1, it is determined that the predicted value derivation mode corresponding to each first block is the first derivation mode.

In this embodiment, the predicted value derivation mode corresponding to the target picture frame is first determined, and in a case that the predicted value derivation mode corresponding to the target picture frame is not the third derivation mode, the second enable identifier corresponding to each first block is further obtained. In this way, bitstreams during inter prediction are reduced.

Optionally, in a case that the predicted value derivation mode corresponding to each first block in the target picture frame is a first derivation mode, the performing inter prediction on each first block based on the target information includes:

- obtaining first motion information of the first block and second motion information of a second block;
- determining, based on position information of the first block and the first motion information, at least one first pixel region associated with the first block;
- determining a predicted value of each pixel in the first pixel region based on the first motion information and the second motion information; and
- determining, based on the predicted value of each pixel, a target predicted value of each pixel in a second pixel region of the first block.

The first block is a to-be-encoded block, and the second block is an encoded block adjacent to the first block; or the first block is a to-be-decoded block, and the second block is a decoded block adjacent to the first block. The first block and the second block meet any one of the following conditions:

- 1. A prediction direction of the first block is different from a prediction direction of the second block.
- 2. A prediction direction of the first block is the same as a prediction direction of the second block, but the prediction directions point to different reference frames.
- 3. A prediction direction of the first block is the same as a prediction direction of the second block, and the prediction directions point to a same reference frame, but a motion vector of the first block is different from a motion vector of the second block.

In this step, in a case that the first block and the second block meet the foregoing condition, the first motion information of the first block and the second motion information of the second block are obtained.

The motion information includes a prediction direction and a motion vector. In this step, prediction may be performed by using the motion vector and the prediction direction based on a position of the first block to obtain at least one first pixel region associated with the first block. As shown in FIG. 2, the first pixel region may be a rectangular region at the top in the first block. Alternatively, as shown in FIG. 3, the first pixel region may be a rectangular region adjacent to a top edge of the first block. Alternatively, as shown in FIG. 4, the first pixel region may be a rectangular region adjacent to a left edge of the first block. Alternatively, the first pixel region may be a rectangular region on the left in the first block.

In this step, after the first pixel region is determined, the predicted value of each pixel in the first pixel region is determined based on the first motion information and the second motion information. For a specific technical solution of determining the predicted value of each pixel in the first pixel region, refer to subsequent embodiments. It should be understood that each pixel in the first pixel region includes at least two predicted values, where one predicted value is determined based on the first motion information, and the other predicted value is determined based on the second motion information.

The second pixel region is a partial pixel region of the first block, and each pixel in the second pixel region is also referred to as a boundary pixel. For case of understanding, refer to FIG. 5 and FIG. 6. FIG. 5 shows a position of the second pixel region in a case that the first pixel region is a rectangular region at the top in the first block or the first pixel region is a rectangular region adjacent to the top edge of the first block. FIG. 6 shows a position of the second pixel region in a case that the first pixel region is a rectangular region on the left in the first block or the first pixel region is a rectangular region adjacent to the left edge of the first block.

In this step, after the predicted value of each pixel in the first pixel region is determined, the target predicted value of each pixel in the second pixel region is determined based on the predicted value of each pixel. For a specific technical solution, refer to subsequent embodiments.

In this embodiment of this application, the first motion information of the first block and the second motion information of the second block are obtained; the at least one first pixel region associated with the first block is determined based on the position information of the first block and the first motion information; the predicted value of each pixel in the first pixel region is determined based on the first motion information and the second motion information; and the target predicted value of each pixel in the second pixel region of the first block is determined based on the predicted value of each pixel. In this embodiment of this application, the predicted value of each pixel in the first pixel region is determined based on the first motion information and the second motion information. In this way, a motion difference between the first block and the second block is fully considered during refinement of a predicted value of a boundary pixel, to improve accuracy of a refined predicted value of the boundary pixel and therefore improve efficiency of video encoding and decoding.

Optionally, the determining a predicted value of each pixel in the first pixel region based on the first motion information and the second motion information includes:

- for any first pixel in a first pixel sub-region, determining a reconstructed value of a first reference pixel that is in a first reference frame and that corresponds to the first pixel as a first predicted value of the first pixel; and
- determining a reconstructed value of a second reference pixel that is in a second reference frame and that corresponds to the first pixel as a second predicted value of the first pixel.

In this embodiment, the first pixel region includes the first pixel sub-region. The first pixel sub-region is a rectangular region at the top in the first block or a rectangular region on the left in the first block, or the first pixel sub-region is a rectangular region adjacent to a top edge of the first block, or the first pixel sub-region is a rectangular region adjacent to a left edge of the first block, or the first pixel sub-region is a rectangular region including a rectangular region adjacent to a top edge of the first block and a rectangular region at the top in the first block, or the first pixel sub-region is a rectangular region including a rectangular region adjacent to a left edge of the first block and a rectangular region on the left in the first block. In this embodiment, a pixel included in the first pixel sub-region is referred to as a first sub-pixel.

The motion information includes a prediction direction, reference frame information, and a motion vector. In this embodiment, the first reference frame and the first reference pixel may be determined based on a first motion vector and first reference frame information, and the first reference pixel is a reconstructed pixel that is located in the first reference frame and that has a same position as the first pixel sub-region. Then a reconstructed value of a pixel in the first reference frame to which the first motion vector points is determined as the first predicted value based on the first reference pixel according to a first prediction direction.

The second reference frame and the second reference pixel may be determined based on a second motion vector and second reference frame information, and the second reference pixel is a reconstructed pixel that is located in the second reference frame and that has a same position as the first pixel sub-region. Then a reconstructed value of a pixel in the second reference frame to which the second motion vector points is determined as the second predicted value based on the second reference pixel according to a second prediction direction.

Optionally, the determining, based on the predicted value of each pixel, a target predicted value of each pixel in a second pixel region of the first block includes:

- calculating a first difference between a first predicted value and a second predicted value that correspond to each first pixel in the first pixel sub-region; and
- in a case that the first difference is greater than a first preset threshold, determining the target predicted value of each pixel in the second pixel region based on position information of each pixel in the second pixel region and the first motion information; or
- in a case that the first difference is less than or equal to the first preset threshold, performing weighted summation on a third predicted value and a fourth predicted value of each pixel in the second pixel region based on a preset first weight value combination to obtain the target predicted value of each pixel in the second pixel region.

In this embodiment, after the first predicted value and the second predicted value of each first pixel are obtained, the first difference between the first predicted value and the second predicted value is calculated. Optionally, the first difference may be an absolute value of a difference between the first predicted value and the second predicted value, or a mean between the first predicted value and the second predicted value, or a variance between the first predicted value and the second predicted value, or a mean square error between the first predicted value and the second predicted value. This is not specifically limited herein.

In this embodiment, each pixel in the second pixel region includes the third predicted value and the fourth predicted value, where the third predicted value is determined based on the first motion information, and the fourth predicted value is determined based on the second motion information. It should be understood that a manner of determining the third predicted value is the same as the manner of determining the first predicted value in the foregoing embodiments, and a manner of determining the fourth predicted value is the same as the manner of determining the second predicted value in the foregoing embodiments. Details are not described herein again.

In this embodiment, the first preset threshold is further set in advance. After the first difference is obtained, for each first pixel, a magnitude relationship between the first difference and the first preset threshold is determined through comparison. If the first difference is greater than the first preset threshold, for each pixel in the second pixel region, a target predicted value of the pixel is determined based on position information of the pixel and the first motion information. Specifically, the pixel is used as a motion starting point, the motion starting point is offset based on the first motion vector and the first prediction direction in the first motion information, and a pixel obtained through offsetting is determined, where the pixel obtained through offsetting is located in the first reference frame. Further, a reconstructed value of the pixel obtained through offsetting is determined as the target predicted value of the pixel. It should be understood that, in a case that the first pixel region is located in the first block, a first predicted value of each first pixel in the first pixel region is determined as a target predicted value of a pixel that is in the second pixel region and that corresponds to the first pixel.

If the first difference is less than or equal to the first preset threshold, weighted summation is performed on the third predicted value and the fourth predicted value of each pixel in the second pixel region by using the first weight value combination to obtain the target predicted value of each pixel in the second pixel region. It should be understood that the first weight value combination includes at least one weight group, the weight group includes a first weight value and a second weight value, the first weight value corresponds to the third predicted value, and the second weight value corresponds to the fourth predicted value.

Specifically, the target predicted value of each pixel in the second pixel region may be calculated by using the following formula:

shift
=

log
2

(

w
⁢

1
1

+

w
⁢

1
2

)

offset
=

(

w
⁢

1
1

+

w
⁢

1
2

)

/
2

Pixel
(

i
,
j

)

=

(

w
⁢

1
1

×
Pixel
⁢
3
⁢

(

i
,
j

)

+

w
⁢

1
2

×
Pixel
⁢
4
⁢

(

i
,
j

)

+
offset

)

≫
shift

,
where

Pixel represents the target predicted value, w1₁represents the first weight value, w1₂represents the second weight value, Pixel3 represents the third predicted value, and Pixel4 represents the fourth predicted value.

It should be understood that the first block may be a luminance block or a chrominance block, and a first weight value combination corresponding to the first block as a luminance block may be different from a first weight value combination corresponding to the first block as a chrominance block.

Optionally, the determining a predicted value of each pixel in the first pixel region based on the first motion information and the second motion information includes:

- for any second pixel in a second pixel sub-region, determining a reconstructed value of a third reference pixel that is in a third reference frame and that corresponds to the second pixel as a fifth predicted value of the second pixel;
- determining a reconstructed value of a fourth reference pixel that is in a fourth reference frame and that corresponds to the second pixel as a sixth predicted value of the second pixel;
- for any third pixel in a third pixel sub-region, determining a reconstructed value of a fifth reference pixel that is in a fifth reference frame and that corresponds to the third pixel as a seventh predicted value of the third pixel; and
- determining a reconstructed value of a sixth reference pixel that is in a sixth reference frame and that corresponds to the third pixel as an eighth predicted value of the third pixel.

In this embodiment, the first pixel region includes the second pixel sub-region and the third pixel sub-region. The second pixel sub-region is a rectangular region at the top in the first block, or the second pixel sub-region is a rectangular region on the left in the first block, or the third pixel sub-region is a rectangular region adjacent to a top edge of the first block, or the third pixel sub-region is a rectangular region adjacent to a left edge of the first block. In this embodiment, a pixel included in the second pixel sub-region is referred to as a second pixel, and a pixel included in the third pixel sub-region is referred to as a third pixel.

As described above, the first motion information includes the first prediction direction, the first reference frame information, and the first motion vector. In this embodiment, the third reference frame and the third reference pixel may be determined based on the first motion information, and then a reconstructed value of a pixel in the third reference frame to which the first motion vector points is determined as the fifth predicted value based on the third reference pixel according to the first prediction direction. It should be understood that a specific technical solution of determining the fifth predicted value of the second pixel in this embodiment is the same as the foregoing technical solution of determining the first predicted value of the first pixel. Details are not described herein again.

As described above, the second motion information includes the second prediction direction, the second reference frame information, and the second motion vector. In this embodiment, the fourth reference frame and the fourth reference pixel may be determined based on the second motion information, and then a reconstructed value of a pixel in the fourth reference frame to which the second motion vector points is determined as the sixth predicted value based on the fourth reference pixel according to the second prediction direction. It should be understood that a specific technical solution of determining the sixth predicted value of the second pixel in this embodiment is the same as the foregoing technical solution of determining the second predicted value of the second pixel. Details are not described herein again.

In this embodiment, the seventh predicted value of the third pixel may be determined by using a technical solution same as the foregoing technical solution of determining the fifth predicted value of the second pixel. Details are not described herein again.

In this embodiment, the eighth predicted value of the third pixel may be determined by using a technical solution same as the foregoing technical solution of determining the sixth predicted value of the second pixel. Details are not described herein again.

Optionally, the determining, based on the predicted value of each pixel, a target predicted value of each pixel in a second pixel region of the first block includes:

- determining a second difference and a third difference of each target pixel based on the fifth predicted value, the sixth predicted value, the seventh predicted value, and the eighth predicted value; and
- in a case that the second difference and the third difference meet a preset condition, determining the target predicted value of each pixel in the second pixel region based on position information of each pixel in the second pixel region and the first motion information; or
- in a case that the second difference and the third difference do not meet the preset condition, performing weighted summation on a ninth predicted value and a tenth predicted value of each pixel in the second pixel region based on a preset second weight value combination to obtain the target predicted value of each pixel in the second pixel region.

In this embodiment, after a fifth predicted value and a sixth predicted value of each second pixel and a seventh predicted value and an eighth predicted value of each third pixel are obtained, a second difference and a third difference of a target pixel may be determined, where the target pixel includes a second pixel and a third pixel. To be specific, a second difference and a third difference of the second pixel and a second difference and a third difference of the third pixel may be determined. For a specific technical solution of determining a second difference and a third difference of a target pixel, refer to subsequent embodiments.

It should be understood that each pixel in the second pixel region includes a ninth predicted value and a tenth predicted value, where the ninth predicted value is determined based on the first motion information, and the tenth predicted value is determined based on the second motion information. It should be understood that a manner of determining the ninth predicted value is the same as the manner of determining the first predicted value in the foregoing embodiments, and a manner of determining the tenth predicted value is the same as the manner of determining the second predicted value in the foregoing embodiments. Details are not described herein again.

In this embodiment, the preset condition is set in advance. If the second difference and the third difference meet the preset condition, it indicates that a motion mode of a boundary pixel in the first block is more consistent with that of the first block. In this case, for each pixel in the second pixel region, a target predicted value of the pixel is determined based on position information of the pixel and the first motion information. For a specific implementation, refer to the foregoing embodiments. Details are not described herein again.

If the second difference and the third difference do not meet the preset condition, weighted summation may be performed on the ninth predicted value and the tenth predicted value of each pixel in the second pixel region by using the second weight value combination to obtain the target predicted value of each pixel in the second pixel region. It should be understood that the second weight value combination includes at least one weight group, the weight group includes a third weight value and a fourth weight value, the third weight value corresponds to the ninth predicted value, the fourth weight value corresponds to the tenth predicted value. Optionally, the second weight value combination may be the same as the first weight value combination.

A specific implementation of weighted summation is consistent with the foregoing implementation of performing weighted summation on the third predicted value and the fourth predicted value by using the first weight value combination. Details are not described herein again.

Optionally, the determining a second difference and a third difference based on the fifth predicted value, the sixth predicted value, the seventh predicted value, and the eighth predicted value includes:

- determining a difference between the fifth predicted value and the sixth predicted value as the second difference, and determining a difference between the seventh predicted value and the eighth predicted value as the third difference; or
- determining a difference between the fifth predicted value and the seventh predicted value as the second difference, and determining a difference between the sixth predicted value and the eighth predicted value as the third difference.

In an optional implementation, the difference between the fifth predicted value and the sixth predicted value is determined as the second difference, and the difference between the seventh predicted value and the eighth predicted value is determined as the third difference.

In this implementation, the second difference may be an absolute value of a difference between the fifth predicted value and the sixth predicted value, or a mean between the fifth predicted value and the sixth predicted value, or a variance between the fifth predicted value and the sixth predicted value, or a mean square error between the fifth predicted value and the sixth predicted value. This is not specifically limited herein.

The third difference may be an absolute value of a difference between the seventh predicted value and the eighth predicted value, or a mean between the seventh predicted value and the eighth predicted value, or a variance between the seventh predicted value and the eighth predicted value, or a mean square error between the seventh predicted value and the eighth predicted value. This is not specifically limited herein.

In another optional implementation, the difference between the fifth predicted value and the seventh predicted value is determined as the second difference, and the difference between the sixth predicted value and the eighth predicted value is determined as the third difference.

In this implementation, the second difference may be an absolute value of a difference between the fifth predicted value and the seventh predicted value, or a mean between the fifth predicted value and the seventh predicted value, or a variance between the fifth predicted value and the seventh predicted value, or a mean square error between the fifth predicted value and the seventh predicted value. This is not specifically limited herein.

The third difference may be an absolute value of a difference between the sixth predicted value and the eighth predicted value, or a mean between the sixth predicted value and the eighth predicted value, or a variance between the sixth predicted value and the eighth predicted value, or a mean square error between the sixth predicted value and the eighth predicted value. This is not specifically limited herein.

Optionally, the preset condition includes any one of the following:

- the second difference is less than the third difference;
- a ratio of the second difference to the third difference is less than a second preset threshold;
- a ratio of the third difference to the second difference is less than the second preset threshold;
- the second difference is greater than a third preset threshold, and a ratio of the second difference to the third difference is less than the second preset threshold; or
- the second difference is greater than the third preset threshold, and a ratio of the third difference to the second difference is less than the second preset threshold.

In an implementation of this embodiment, a magnitude relationship between the second difference and the third preset threshold is first determined through comparison, and in a case that the second difference is greater than the third preset threshold, a magnitude relationship between the ratio of the second difference to the third difference and the second preset threshold is determined through comparison, or a magnitude relationship between the ratio of the third difference to the second difference and the second preset threshold is determined through comparison. The magnitude relationship between the second difference and the third preset threshold is first determined through comparison, so that a motion difference between the first block and the second block is more fully considered during refinement of a predicted value of a boundary pixel.

Optionally, the first pixel region meets at least one of the following:

- the first pixel region is an encoded or decoded pixel region that includes M1 rows and N1 columns of pixels adjacent to a top edge of the first block;
- the first pixel region is an encoded or decoded pixel region that includes M2 rows and N2 columns of pixels adjacent to a left edge of the first block;
- the first pixel region is an unencoded or undecoded pixel region that includes M3 rows and N3 columns of pixels located at the top in the first block;
- the first pixel region is an unencoded or undecoded pixel region that includes M4 rows and N4 columns of pixels located on the left in the first block;
- the first pixel region is an M5-row N5-column pixel region that includes an encoded or decoded pixel region adjacent to a top edge of the first block and an unencoded or undecoded pixel region at the top in the first block; and
- the first pixel region is an M6-row N6-column pixel region that includes an encoded or decoded pixel region adjacent to a left edge of the first block and an unencoded or undecoded pixel region on the left in the first block, where
- M1, M2, M3, M4, M5, M6, N1, N2, N3, N4, N5, and N6 are all positive integers.

In an optional implementation, the first pixel region is an unencoded or undecoded pixel region that includes M3 rows and N3 columns at the top in the first block. For ease of understanding, refer to FIG. 2. In a scenario shown in FIG. 2, the first pixel region is a rectangular region that includes one row and three columns of pixels at the top in the first block.

In another optional implementation, the first pixel region is an encoded or decoded pixel region that includes M1 rows and N1 columns adjacent to the top edge of the first block. For case of understanding, refer to FIG. 3. In a scenario shown in FIG. 3, the first pixel region is a rectangular region that includes one row and three columns of pixels adjacent to the top edge of the first block.

In another optional implementation, the first pixel region is an encoded or decoded pixel region that includes M2 rows and N2 columns adjacent to the left edge of the first block. For case of understanding, refer to FIG. 4. In a scenario shown in FIG. 4, the first pixel region is a rectangular region that includes three rows and one column adjacent to the left edge of the first block.

In another optional implementation, the first pixel region may be an encoded or decoded pixel region that includes some pixels adjacent to the top edge of the first block, and an encoded or decoded pixel region that includes some adjacent pixels adjacent to the left edge of the first block. In this case, the first pixel region is in an “L” shape.

In this embodiment, a position relationship between the first pixel region and the first block is limited, to fully consider a motion difference between the first block and each block adjacent to the first block, and improve accuracy of a refined predicted value of a boundary pixel.

The embodiments of this application further provide an inter prediction method. The following describes in detail the inter prediction method provided in the embodiments of this application with reference to the accompanying drawings and by using some embodiments and application scenarios thereof. For case of understanding, the following describes some content included in the embodiments of this application.

In a case that a boundary of a block does not fit a contour of a current block, a motion mode of a boundary pixel of the current block may be consistent with that of the current block, or may be consistent with that of an adjacent block. A predicted value, determined based on motion information of the current block, of the boundary pixel may be greatly different from a real predicted value, reducing efficiency of video encoding and decoding. The current block may be a to-be-encoded block, and the adjacent block is an encoded block; or the current block may be a to-be-decoded block, and the adjacent block is a decoded block.

Currently, the predicted value of the boundary pixel of the current block may be refined by using an OBMC technology, to resolve the foregoing technical problem. The OBMC technology is an inter prediction method. The OBMC technology is specifically described below.

Case 1: Inter Prediction Modes for Pixels in a Current Block are the Same.

In this case, motion information of an adjacent block is obtained in a case that the adjacent block is in an inter prediction mode rather than an intra block copy mode and a motion mode of the adjacent block is inconsistent with a motion mode of the current block. Refer to FIG. 7. The adjacent block may be a block adjacent to a top edge of the current block, or a block adjacent to a left edge of the current block.

In a case that any one of the following conditions is met, it can be determined that the motion mode of the adjacent block is inconsistent with the motion mode of the current block.

- 1. A prediction direction of the adjacent block is different from a prediction direction of the current block.
- 2. A prediction direction of the adjacent block is the same as a prediction direction of the current block, but the prediction directions point to different reference frames.
- 3. A prediction direction of the adjacent block is the same as a prediction direction of the current block, and the prediction directions point to a same reference frame, but a motion vector of the adjacent block is different from a motion vector of the current block.

After the motion information of the adjacent block is obtained, a first predicted value is obtained based on motion information of the current block, and a second predicted value is obtained based on the motion information of the adjacent block. A predicted value of a boundary pixel of the current block is refined by using the first predicted value and the second predicted value.

Specifically, if the current block is a luminance subblock, weighted summation may be performed on the first predicted value and the second predicted value by using the following formula, to obtain a refined predicted value of the boundary pixel:

NewPixel
⁡
(

i
,
j

)

=

(

26
×
Pixel
⁢
1
⁢

(

i
,
j

)

+

6
×
Pixel
⁢
2
⁢

(

i
,
j

)

+
16

)

≫
5

NewPixel
⁡
(

i
,
j

)

=

(

7
×
Pixel
⁢
1
⁢

(

i
,
j

)

+

Pixel
⁢
2
⁢

(

i
,
j

)

+
4

)

≫
3

NewPixel
⁡
(

i
,
j

)

=

(

15
×
Pixel
⁢
1
⁢

(

i
,
j

)

+

Pixel
⁢
2
⁢

(

i
,
j

)

+
8

)

≫
4

NewPixel
⁡
(

i
,
j

)

=

(

31
×
Pixel
⁢
1
⁢

(

i
,
j

)

+

Pixel
⁢
2
⁢

(

i
,
j

)

+
16

)

≫
5

,
where

- i represents a column coordinate of the boundary pixel in the current block, j represents a row coordinate of the boundary pixel in the current block, Pixel1 represents the first predicted value of the boundary pixel, Pixel2 represents the second predicted value of the boundary pixel, and NewPixel represents the refined predicted value of the boundary pixel.

If the current block is a chrominance subblock, weighted summation may be performed on the first predicted value and the second predicted value by using the following formula, to obtain a refined predicted value of the boundary pixel:

NewPixel
⁡
(

i
,
j

)

=

(

26
×
Pixel
⁢
1
⁢

(

i
,
j

)

+

6
×
Pixel
⁢
2
⁢

(

i
,
j

)

+
16

)

≫
5

,
where

- NewPixel represents the refined predicted value of the boundary pixel.

It should be understood that an application scenario corresponding to the foregoing formula is that a pixel region of the boundary pixel includes four rows or four columns, and in other application scenarios, a pixel region of the boundary pixel is not specifically limited.

Case 2: The current block is an encoded block, and an inter prediction mode is an affine mode; or the current block is a decoded block, and an inter prediction mode is a motion vector refinement mode.

In this case, motion information of four adjacent blocks that are adjacent to a top edge, a bottom edge, a left edge, and a right edge of the current block is obtained. Refer to FIG. 8. FIG. 8 shows a position relationship between the adjacent blocks and the current block in the foregoing case.

A first predicted value is obtained based on motion information of the adjacent block, and second predicted values are obtained based on the motion information of the adjacent blocks in a case that the current block and the adjacent blocks meet any one of the following conditions.

- 1. Prediction directions of the adjacent blocks are different from a prediction direction of the current block.
- 2. Prediction directions of the adjacent blocks are the same as a prediction direction of the current block, but the prediction directions point to different reference frames.
- 3. Prediction directions of the adjacent blocks are the same as a prediction direction of the current block, and the prediction directions point to a same reference frame, but absolute values of differences between motion vectors of the adjacent blocks and a motion vector of the current block are greater than a preset threshold.

A predicted value of a boundary pixel of the current block is refined by using the first predicted value and the second predicted values. Specifically, weighted summation may be performed on the first predicted value and the second predicted values by using the following formula, to obtain a refined predicted value of the boundary pixel:

rem_w
⁢

(

i
,
j

)

=

(

32
-

w
⁡
(
i
)

-

w
⁡
(

width
-
i

)

-

w
⁡
(
j
)

-

w
⁡
(

height
-
j

)

)

subNewPixel
⁡
(

i
,
j

)

=

(

subPixel
⁢

2
L

⁢

(

i
,
j

)

×

w
⁡
(
i
)

+

subPixel
⁢

2
R

⁢

(

i
,
j

)

×

w
⁡
(

width
-
1
-
i

)

+

subPixel
⁢

2
T

⁢

(

i
,
j

)

×

w
⁡
(
j
)

+

subPixel
⁢

2
B

⁢

(

i
,
j

)

×

w
⁡
(

height
-
1
-
j

)

+

subPixel
⁢
1
×
rem_w
⁢

(

i
,
j

)

+
16

)

≫
5

,
where

- i represents a column coordinate of the boundary pixel in the current block, j represents a row coordinate of the boundary pixel in the current block, subNewPixel represents the refined predicted value of the boundary pixel, subPixel2_L, subPixel2_R, subPixel2_T, and subPixel2_Brepresent the second predicted values determined based on the motion information of the adjacent blocks, width represents the number of columns of the adjacent block, height represents the number of rows of the adjacent block, w represents a preset weight combination, and a weight combination corresponding to the current block as a luminance block is different from a weight combination corresponding to the current block as a chrominance block.

It should be understood that an application scenario corresponding to the foregoing formula is that a pixel region of the boundary pixel includes four rows or four columns, and in other application scenarios, a pixel region of the boundary pixel is not specifically limited.

In the foregoing process of refining a predicted value of a boundary pixel by using the OBMC technology, a difference between a motion mode of a current block and a motion mode of an adjacent block is not considered. Consequently, a refined predicted value of the boundary pixel is inaccurate, reducing efficiency of video encoding and decoding.

Based on the foregoing cases, how to improve accuracy of a refined predicted value of a boundary pixel and further improve efficiency of video encoding and decoding is a technical problem to be resolved.

To resolve the foregoing possible technical problem, the embodiments of this application provide an inter prediction method. Refer to FIG. 9. FIG. 9 is a flowchart of another inter prediction method provided in this application. The another inter prediction method provided in this embodiment includes the following steps.

S901: Obtain first motion information of a first block and second motion information of a second block.

S902: Determine, based on position information of the first block and the first motion information, at least one first pixel region associated with the first block.

S903: Determine a predicted value of each pixel in the first pixel region based on the first motion information and the second motion information.

S904: Determine, based on the predicted value of each pixel, a target predicted value of each pixel in a second pixel region of the first block.

It should be understood that, in other embodiments, the inter prediction method provided in this embodiment of this application may alternatively be used to generate a predicted value of a boundary pixel of each subblock in an encoded block or a decoded block. In this implementation, refer to FIG. 10. A subblock shown in FIG. 10 may be understood as the first block in this embodiment, and the subblock is located at the lower right of an encoded block or a decoded block. Refer to FIG. 11. A subblock shown in FIG. 11 may be understood as the first block in this embodiment, and the subblock is located in the middle of an encoded block or a decoded block.

In this embodiment of this application, the first motion information of the first block and the second motion information of the second block are obtained; the at least one first pixel region associated with the first block is determined based on the position information of the first block and the first motion information; the predicted value of each pixel in the first pixel region is determined based on the first motion information and the second motion information; and the target predicted value of each pixel in the second pixel region of the first block is determined based on the predicted value of each pixel. In this embodiment of this application, the predicted value of each pixel in the first pixel region is determined based on the first motion information and the second motion information. In this way, a motion difference between the first block and the second block is fully considered during refinement of a predicted value of a boundary pixel, to improve accuracy of a refined predicted value of the boundary pixel and therefore improve efficiency of video encoding and decoding.

Optionally, the first pixel region includes a first pixel sub-region, and the first pixel sub-region is a partial pixel region of the first block, or the first pixel sub-region is a partial pixel region of the second block, or the first pixel sub-region is a region including a partial pixel region of the first block and a partial pixel region of the second block; and

- the determining a predicted value of each pixel in the first pixel region based on the first motion information and the second motion information includes:
- for any first pixel in the first pixel sub-region, determining a reconstructed value of a first reference pixel that is in a first reference frame and that corresponds to the first pixel as a first predicted value of the first pixel; and
- determining a reconstructed value of a second reference pixel that is in a second reference frame and that corresponds to the first pixel as a second predicted value of the first pixel, where
- the first reference frame is determined based on the first motion information, and the second reference frame is determined based on the second motion information.

Optionally, the determining, based on the predicted value of each pixel, a target predicted value of each pixel in a second pixel region of the first block includes:

- calculating a first difference between a first predicted value and a second predicted value that correspond to each first pixel in the first pixel sub-region; and
- in a case that the first difference is greater than a first preset threshold, determining the target predicted value of each pixel in the second pixel region based on position information of each pixel in the second pixel region and the first motion information; or
- in a case that the first difference is less than or equal to the first preset threshold, performing weighted summation on a third predicted value and a fourth predicted value of each pixel in the second pixel region based on a preset first weight value combination to obtain the target predicted value of each pixel in the second pixel region, where
- the first weight value combination includes at least one weight group, the weight group includes a first weight value and a second weight value, the first weight value corresponds to the third predicted value, the second weight value corresponds to the fourth predicted value, the third predicted value is determined based on the first motion information, and the fourth predicted value is determined based on the second motion information.

Optionally, the first pixel region includes a second pixel sub-region and a third pixel sub-region, the second pixel sub-region is a partial pixel region of the first block, and the third pixel sub-region is a partial pixel region of the second block; and

- the determining a predicted value of each pixel in the first pixel region based on the first motion information and the second motion information includes:
- for any second pixel in the second pixel sub-region, determining a reconstructed value of a third reference pixel that is in a third reference frame and that corresponds to the second pixel as a fifth predicted value of the second pixel;
- determining a reconstructed value of a fourth reference pixel that is in a fourth reference frame and that corresponds to the second pixel as a sixth predicted value of the second pixel;
- for any third pixel in the third pixel sub-region, determining a reconstructed value of a fifth reference pixel that is in a fifth reference frame and that corresponds to the third pixel as a seventh predicted value of the third pixel; and
- determining a reconstructed value of a sixth reference pixel that is in a sixth reference frame and that corresponds to the third pixel as an eighth predicted value of the third pixel, where
- the third reference frame and the fifth reference frame are determined based on the first motion information, and the fourth reference frame and the sixth reference frame are determined based on the second motion information.

Optionally, the determining, based on the predicted value of each pixel, a target predicted value of each pixel in a second pixel region of the first block includes:

- determining a second difference and a third difference of each target pixel based on the fifth predicted value, the sixth predicted value, the seventh predicted value, and the eighth predicted value, where the target pixel includes a second pixel and a third pixel; and
- in a case that the second difference and the third difference meet a preset condition, determining the target predicted value of each pixel in the second pixel region based on position information of each pixel in the second pixel region and the first motion information; or
- in a case that the second difference and the third difference do not meet the preset condition, performing weighted summation on a ninth predicted value and a tenth predicted value of each pixel in the second pixel region based on a preset second weight value combination to obtain the target predicted value of each pixel in the second pixel region, where
- the second weight value combination includes at least one weight group, the weight group includes a third weight value and a fourth weight value, the third weight value corresponds to the ninth predicted value, the fourth weight value corresponds to the tenth predicted value, the ninth predicted value is determined based on the first motion information, and the tenth predicted value is determined based on the second motion information.

Optionally, the determining a second difference and a third difference based on the fifth predicted value, the sixth predicted value, the seventh predicted value, and the eighth predicted value includes:

- determining a difference between the fifth predicted value and the sixth predicted value as the second difference, and determining a difference between the seventh predicted value and the eighth predicted value as the third difference; or
- determining a difference between the fifth predicted value and the seventh predicted value as the second difference, and determining a difference between the sixth predicted value and the eighth predicted value as the third difference.

The inter prediction method provided in the embodiments of this application may be performed by an inter prediction apparatus. In the embodiments of this application, an inter prediction apparatus provided in the embodiments of this application is described by using an example in which the inter prediction apparatus performs the inter prediction method.

As shown in FIG. 12, an inter prediction apparatus 1200 includes:

- an obtaining module 1201, configured to obtain target information; and
- a processing module 1202, configured to perform inter prediction on each first block based on the target information.

Optionally, the processing module 1202 is specifically configured to:

- in a case that the predicted value derivation mode corresponding to the target picture frame is a first derivation mode, determine that the predicted value derivation mode corresponding to each first block in the target picture frame is the first derivation mode, and performing inter prediction on each first block by using the first derivation mode;
- in a case that the predicted value derivation mode corresponding to the target picture frame is a second derivation mode, determine that the predicted value derivation mode for each first block in the target picture frame is the second derivation mode, and performing inter prediction on each first block by using the second derivation mode; and
- in a case that the predicted value derivation mode corresponding to the target picture frame is a third derivation mode, determine that the predicted value derivation mode for each first block in the target picture frame is the third derivation mode, and performing inter prediction on each first block by using the third derivation mode.