JP2006157531A - Apparatus for coding and decoding moving image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus for coding and decoding a moving image capable of minimizing an increase in a code amount of a motion vector while inserting an interpolation pixel among actual pixels to improve the accuracy of motion compensation. <P>SOLUTION: Motion vectors are divided into motion vectors with integer pixel accuracy that can be represented by an integer and motion vectors with fractional accuracy that has to be represented by a fraction. In the case of a motion vector with integer accuracy, the motion vector is converted into a value obtained by multiplying it by prescribed value to encode the value. In the case of a motion vector with fractional accuracy, the motion vector is converted into a value obtained by multiplying an integer part by a prescribed number and adding a prescribed odd number value to encode the value, while a fractional part is regarded as motion vector accuracy information and encoded as different information. In decoding, in the case of integer accuracy, a division is performed by the prescribed value to calculate a motion vector, and in the case of fractional accuracy, subtraction is performed by a prescribed odd number value, and a value divided by the prescribed value is added to the motion vector accuracy information to calculate the motion vector. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a moving picture coding / decoding apparatus.

  In motion picture compression coding represented by MPEG, motion compensation predictive coding (hereinafter simply referred to as motion compensation) in which a code amount is compressed using a correlation between successive pictures is often used. Motion compensation is performed by dividing an image into blocks of 16 × 16 pixels, for example, and sequentially setting each block as an encoding target image block.

  In motion compensation, a motion vector that indicates how much the corresponding block region in the reference image used for encoding the image block to be encoded exists at a shifted position is used. In the reference image, when the motion vector that minimizes the error evaluation value with respect to the encoding target image block is detected and used for motion compensation, the encoding efficiency becomes the best. The error evaluation value expresses the degree of difference between corresponding blocks, and is obtained by block matching.

  Block matching is a technique for matching an encoding target image block with an image within a detection range in a reference image at every position in units of pixels, calculating a matching evaluation function, and obtaining an error evaluation value. As the matching evaluation function, for example, a sum of absolute values of pixel value differences (absolute value difference sum) for all pixels in the block is used.

  In such motion compensation, a technique that improves the accuracy of motion compensation by inserting interpolated pixels between actual pixels of the reference image so that fractional motion can be captured has been used. It has been.

  FIG. 5 is a diagram showing a state in which interpolation with 1/2 pixel accuracy is performed by inserting one interpolation pixel between actual pixels. The calculation of the interpolated pixel value with 1/2 pixel accuracy includes a method of averaging the adjacent actual pixel values.

  FIG. 6 is a diagram showing a state in which interpolation with 1/4 pixel accuracy in which three interpolation pixels are inserted between actual pixels is performed. Motion compensation with 1/4 pixel accuracy is 1/2 pixel accuracy. Therefore, the accuracy of motion compensation is further improved than that of motion compensation. The calculation of the interpolated pixel value with the ¼ pixel accuracy includes a method of calculating the average with the adjacent actual pixel or the interpolated pixel with the ½ pixel accuracy after calculating the interpolated pixel with the ½ pixel accuracy. .

However, as the number of interpolated pixels increases between actual pixels in order to improve the accuracy of motion compensation in this way, the amount of code of the motion vector increases.
For example, when motion compensation is performed using only actual pixels without using interpolated pixels, the resolution of a motion vector for representing the motion of one pixel can be represented by 1, but interpolated pixels with 1/2 pixel accuracy are used. In the case of motion compensation, since one interpolation pixel is inserted between actual pixels, 2 is necessary as the resolution of the motion vector in order to represent the motion of one pixel of the actual pixel. Similarly, in motion compensation using interpolated pixels with ¼ pixel accuracy, there are three interpolated pixels between actual pixels. Therefore, in order to represent the motion of one actual pixel, the resolution of the motion vector 4 is required. Therefore, the magnitude of the motion vector representing the movement of one pixel of the actual pixel increases as the number of interpolation pixels increases.

  Usually, as a motion vector expression method, a motion vector indicating a position up to an actual pixel is expressed as an integer, and a motion vector indicating a position up to an interpolation pixel is expressed as an integer and a fraction, or an integer and a decimal. For example, in the case of motion compensation using an interpolation pixel with 1/2 pixel accuracy, the motion vector indicating the position separated by two pixels in the actual pixel is 2, and further to the next interpolation pixel. The motion vector indicating the position of is represented by 5/2, or 2 and 1/2, or 2.5.

  Hereinafter, in this description, as described above, a motion vector indicating a position up to an actual pixel that can be represented only by an integer is referred to as an integer pixel precision motion vector, up to an interpolation pixel that must be represented by an integer and a fraction, or an integer and a decimal. A motion vector that indicates the position of the position will be referred to as a fractional pixel precision motion vector for distinction.

Various proposals have heretofore been made in order to suppress an increase in the code amount of a motion vector while improving the accuracy of motion compensation using interpolated pixels. For example, in Patent Document 1, a moving image coding decoding that suppresses an increase in the amount of code of a motion vector by adaptively selecting a motion vector of a current image block to be predicted from motion vectors of surrounding blocks. A device has been proposed.
[Conventional encoding device]
Hereinafter, the configuration of this conventional moving image encoding apparatus will be described with reference to FIG. A conventional moving image coding apparatus includes an image memory 301, an image interpolation unit 302, a motion vector detection unit 303, a motion compensation unit 304, an orthogonal transform unit 305, a quantization unit 306, and an inverse quantization unit 307. , An inverse orthogonal transform unit 308, a motion vector memory 309, a difference vector calculation unit 310, and an entropy encoding unit 311.

The image memory 301 stores a decoded image obtained by encoding and re-decoding each block of the encoding target image as a reference image to be used for encoding each block of the next encoding target image.
The image interpolating unit 302 generates a reference interpolation image obtained by interpolating between actual pixels of the reference image with interpolation pixels.

  The motion vector detection unit 303 detects the motion vector of the encoding target image block using the reference interpolation image generated by the image interpolation unit 302. Then, the detected motion vector is supplied to the motion compensation unit 304 and the difference vector calculation unit 310 and stored in the motion vector memory 309.

  The motion compensation unit 304 performs motion compensation using the motion vector detected by the motion vector detection unit 303 and the reference interpolation image generated by the image interpolation unit 302 to generate a predicted image block.

The orthogonal transform unit 305 performs orthogonal transform on the difference signal between the encoding target image block and the prediction image block, and supplies the converted difference signal to the quantization unit 306.
The quantization unit 306 quantizes the difference signal transformed by the orthogonal transformation unit 305 and supplies the quantized difference signal to the inverse quantization unit 307 and the entropy encoding unit 311.

The inverse quantization unit 307 inversely quantizes the difference signal quantized by the quantization unit 306 and supplies the inversely quantized difference signal to the inverse orthogonal transform unit 308.
The inverse orthogonal transform unit 308 performs inverse orthogonal transform on the differential signal inversely quantized by the inverse quantization unit 307.

The motion vector memory 309 stores the motion vector detected by the motion vector detection unit 303.
The difference vector calculation unit 310 calculates a difference (hereinafter referred to as a difference vector) between the motion vector detected by the motion vector detection unit 303 and the motion vector of the peripheral block of the encoding target image block stored in the motion vector memory 309. calculate. Then, the difference vector having the smallest difference and the surrounding block selection information are supplied to the entropy encoding means 311.

Here, the operation of the difference vector calculation means 310 will be described using the flowchart of FIG. 7 and FIG.
First, as shown in FIG. 8, a difference vector between the motion vector (MV1) of the left block of the encoding target image block and the motion vector of the encoding target image block is calculated (S21). It is determined whether or not the calculated difference vector is the smallest difference vector (hereinafter referred to as the minimum difference vector) (S22). If the calculated difference vector is the minimum difference vector, the difference vector is updated by replacing the determined difference vector with the determined difference vector. (S23). Here, the difference vector for MV1 to be determined first is always determined to be the minimum difference vector. The minimum difference vector is determined by the sum of the x component and y component of the difference vector.

  The above processing is repeated for the motion vector (MV2) of the upper block of the encoding target image block and the motion vector (MV3) of the upper right block of the encoding target image block (S24). Then, the determined minimum difference vector and block position information corresponding to the minimum difference vector are output (S25).

  Returning to FIG. 3, the description will be continued. The entropy encoding unit 311 entropy-encodes the difference signal quantized by the quantization unit 306, the difference vector selected by the difference vector calculation unit 310, and the surrounding block selection information, and outputs the encoded data to the outside.

Encoding is performed by the above processing.
[Conventional Decoding Device]
Next, the operation of the conventional video decoding device will be described with reference to FIG.

  As shown in FIG. 4, the conventional moving image decoding apparatus includes an entropy decoding unit 401, an image memory 402, an image interpolation unit 403, a motion vector calculation unit 404, a motion vector memory 405, and a motion compensation unit 406. And an inverse quantization means 407 and an inverse orthogonal transform means 408.

  The entropy decoding unit 401 performs entropy decoding on the input code data. Then, the decoded difference vector and the neighboring block selection information are supplied to the motion vector calculation unit 404 and the decoded difference signal is supplied to the inverse quantization unit 407.

The image memory 402 stores a decoded image for which all blocks have already been decoded as a reference image to be used for decoding each block of the next decoding target image.
The image interpolation unit 403 generates a reference interpolation image obtained by interpolating the pixels of the reference image.

  The motion vector calculation unit 404 selects a prediction vector from the motion vectors of the peripheral blocks of the decoding target image block stored in the motion vector memory 405 according to the peripheral block selection information decoded by the entropy decoding unit 401. select. Then, a complete motion vector of the decoding target image block is calculated by adding the prediction vector of the selected peripheral block and the decoded difference vector of the decoding target image block. The calculated motion vector of the decoding target image block is supplied to the motion vector memory 405 and the motion compensation unit 406.

The motion vector memory 405 stores the motion vector calculated by the motion vector calculation unit 404.
The motion compensation unit 406 performs motion compensation using the motion vector of the decoding target image block calculated by the motion vector calculation unit 404 and the reference interpolation image generated by the image interpolation unit 403 to generate a predicted image.

The inverse quantization means 407 inversely quantizes the difference signal decoded by the entropy decoding means 401 and supplies the inversely quantized difference signal to the inverse orthogonal transform means 408.
The inverse orthogonal transform unit 408 performs inverse orthogonal transform on the difference signal inversely quantized by the inverse quantization unit 407.

Decoding is performed by the above processing.
JP-A-11-112994

  According to the above-described moving picture coding apparatus disclosed in Patent Document 1, a motion vector code is obtained by using a minimum difference vector having the smallest difference between a motion vector of an encoding target image block and a motion vector of a peripheral block as a prediction vector. The amount can be reduced.

  However, it is necessary to encode peripheral block selection information indicating which peripheral block motion vector is selected as a prediction vector for the encoding target image block. Therefore, compared with a general method in which the median (median value) of motion vectors of surrounding blocks is used as a prediction vector, the effect of suppressing the increase in the amount of code of motion vectors is not so large, and the generated codes necessary for motion vector information transmission It is difficult to say that the effect of suppressing the increase in the amount is large.

  In addition, in the conventional video encoding / decoding device, in order to improve the accuracy of motion compensation, the number of interpolated pixels between actual pixels is increased, and the more accurate the pixel is, the more the motion vector code is encoded. The amount increases. Therefore, once the accuracy of motion compensation is set to a predetermined level, a motion vector is generated based on the set motion compensation accuracy even when motion compensation should be performed with an accuracy lower than the set accuracy. Therefore, since the accuracy is originally lowered, it is desired to keep the code amount of the motion vector low. However, the code amount is not so small as compared with the case where the motion compensation accuracy is set low from the beginning.

The present invention has been made in view of the above problems.
The present invention is a moving image that can suppress an increase in the amount of code of a generated motion vector to a minimum while inserting interpolation pixels between actual pixels of an encoding target image to improve the accuracy of motion compensation. An object is to realize an image encoding / decoding device.

  In addition, the present invention provides a motion vector code amount that is as low as when motion compensation accuracy is initially set low when motion compensation is performed with low accuracy while setting motion compensation accuracy high. It is another object of the present invention to realize a moving image encoding / decoding device that can be suppressed to a low level.

Therefore, in order to solve the above problems, the present invention provides the following apparatus.
(1) An encoding target image is divided into encoding target image blocks each including a two-dimensional array of a plurality of pixels, and motion compensation prediction encoding, orthogonal transform encoding, quantization, A moving image that uses a reference interpolation image in which interpolation pixels are inserted between pixels of an image obtained by performing reencoding on an image that has already been encoded as an image that is subjected to entropy coding and is referred to when performing the motion compensation predictive coding In an image encoding device,
The prediction vector of the encoding target image block is calculated from the motion vectors of the plurality of image blocks that have already been encoded adjacent to the encoding target image block, respectively, and the calculated prediction vector and the actual encoding are encoded. Difference vector calculation means (110) for calculating a difference vector that is a difference from the motion vector of the encoding target image block obtained by performing;
If the difference vector calculated by the difference vector calculating means is an integer pixel precision difference vector indicating the actual pixel position of the reference interpolated image, either an even number or an odd number according to the integer value The first conversion difference vector of the type is generated, first code data is generated according to the generated first conversion difference vector, and the difference vector calculated by the difference vector calculation means is the reference interpolation In the case of a differential vector with fractional pixel precision indicating the position of an interpolated pixel in the image, the first converted differential vector according to an integer value obtained by dividing the numerator of the differential vector precision with fractional pixel by the denominator is used. Generate the second conversion difference vector of the other type different from the odd or even number, and generate the second code data according to the generated second conversion difference vector Motion vector code amount suppression means (111) for generating third code data corresponding to the vector accuracy information, using a remainder value obtained by dividing the numerator of the difference vector with fractional pixel accuracy by a denominator; ,
The entropy code using the first code data, the second code data, and the third code data generated by the motion vector code amount suppression unit and the encoded data obtained by encoding the encoding target image Entropy encoding means (112) for generating the code string data by outputting the code string data to the outside;
A moving picture encoding apparatus comprising:
(2) Using the code string data generated by the video encoding device as an input, entropy decoding, inverse quantization, inverse orthogonal transform decoding, and motion compensated prediction decoding are performed, and the encoding target image is decoded. In the moving picture decoding apparatus,
The entropy decoding is performed with the code string data as input, the encoded data of the encoding target image, the first code data, the second code data, and the third code data. Generating entropy decoding means (201);
Generating the first conversion difference vector according to the first code data generated by the entropy decoding means, calculating an integer pixel precision difference vector according to the generated first conversion difference vector; The second conversion difference vector corresponding to the second code data generated by the entropy decoding means is generated, and an integer value of the difference vector with fractional pixel accuracy corresponding to the generated second conversion difference vector is obtained. Calculating, generating the vector accuracy information according to the third code data generated by the entropy decoding means, and adding the calculated integer value of the difference vector of the fractional pixel accuracy and the vector accuracy information A difference vector calculating means (204) for calculating a difference vector with fractional pixel precision;
The prediction vector of the decoding target image block is calculated from the motion vectors of a plurality of image blocks that have already been decoded and are adjacent to the image block to be decoded this time, and the calculated prediction vector and the calculated vector The motion vector actually used for decoding the decoding target image block is generated by adding the difference vector of integer pixel accuracy or the difference vector of fractional pixel accuracy calculated by the difference vector calculation means, Motion vector calculation means (205);
A moving picture decoding apparatus comprising:

  According to the present invention, it is possible to suppress an increase in the amount of code of a generated motion vector while minimizing the accuracy of motion compensation by inserting interpolation pixels between actual pixels of an encoding target image. It becomes.

  In addition, in consideration of the bias of the vector accuracy information, it is possible to further suppress the increase in the code amount of the motion vector by assigning a short bit length code to the vector accuracy information having a high appearance probability.

  Furthermore, according to the present invention, when low-precision motion compensation is performed while the motion compensation accuracy is set high, the motion vector code amount is equivalent to that when the motion compensation accuracy is set low from the beginning. It becomes possible to hold down to a low amount.

  The contents of the motion vector code amount suppression means used in the video encoding / decoding device of one embodiment of the present invention, which can suppress the increase in the code amount of the motion vector even by using interpolation pixels, will be described below. . In this description, the motion vector is replaced with the difference vector described in the conventional example. This is because in actual encoding, in order to suppress the code amount, a motion vector is not used as it is, but a difference vector is often used.

  As described in the background art section above, the difference vector representing the motion compensation using the interpolated pixels is generally expressed as an integer as the difference vector indicating the position up to the actual pixel. In general, the difference vector indicating the position to the interpolation pixel is represented by an integer and a fraction, or an integer and a decimal. In this description, as described above, each is defined as a difference vector with integer pixel precision and a difference vector with fractional pixel precision.

  The motion vector code amount suppression means used in the moving picture coding / decoding apparatus of the present embodiment converts the difference vector with integer pixel precision and the difference vector with fractional pixel precision as follows.

The difference vector of integer pixel accuracy is obtained by multiplying the value of the difference vector by a predetermined even value (2 in this description) to convert it to an even value, and a new difference vector of integer pixel accuracy is obtained.
Also, the difference vector with fractional pixel precision is converted by dividing the difference vector into an integer part and a fractional part. The integer part of the difference vector is obtained by multiplying the value by a predetermined even value (2 in this description), adding a predetermined odd value (1 in this description) to the multiplication result, and converting it to an odd value with fractional pixel precision. The value of the difference vector of the integer part is used. The fractional part of the difference vector is converted in accordance with a conversion assignment specified separately, and the converted value is used as vector accuracy information. Then, a new fractional pixel precision difference vector is formed by combining the fraction vector precision integer vector difference vector value and the vector precision information.

Each difference vector subjected to the above-described conversion is referred to as a conversion difference vector in this description.
Thus, when the difference vector is used as code data, the difference vector of integer pixel accuracy is represented by an even value, and the difference vector of fractional pixel accuracy is represented by an odd value and vector accuracy information. Only in this case, it is sufficient to add vector accuracy information as code data. Therefore, when a difference vector with integer pixel accuracy is used as code data, since it is not necessary to add vector accuracy information, the code amount is about half of the difference vector with fractional accuracy, and the code amount of the overall difference vector is Can be suppressed.

  Further, when a differential vector with fractional pixel accuracy is used as code data, the vector accuracy information becomes independent code data. Therefore, when a bias is expected in the appearance probability of the vector accuracy information in advance, an increase in the code amount of the motion vector can be further suppressed by assigning a short bit length code to the vector accuracy information having a high appearance probability.

  In this description, the integer pixel precision difference vector is converted to an even value, and the integer part of the fraction pixel precision difference vector is converted to an odd value. The integer part of may be converted to an even value.

Hereinafter, an example will be described in which code amounts are compared in the case where code allocation is actually performed between the conventional case and the present embodiment.
FIG. 10 is a table showing the relationship between a conventional difference vector in the case of 1/4 pixel accuracy and allocation code data assigned to the difference vector. FIG. 11 is a table showing the relationship of allocation code data to be assigned to the position information of the neighboring blocks for which the difference vector has been obtained.

  FIG. 12 is a table showing the relationship between the conventional difference vector in the case of 1/4 pixel accuracy, the conversion difference vector of the present embodiment corresponding to the difference vector, and the allocation code data assigned to the conversion difference vector. is there. FIG. 13 is a table showing a conventional difference vector in the case of ¼ pixel accuracy, vector accuracy information of this embodiment corresponding to the difference vector, and allocation code data assigned to the vector accuracy information.

  FIG. 14 is a first example in which motion vector appearance probabilities at the time of actual motion compensation are assigned to the respective allocation code data allocated as described above. This is an example of the case where the appearance probabilities of motion vectors with 1/4 pixel accuracy, 2/4 pixel accuracy, and 3/4 pixel accuracy are equal. In this example, the code amount of the motion vector at the time of actual motion compensation is calculated and compared between the conventional example and this example.

Assume that 20 motion vectors appear according to the appearance probability. Therefore, the motion vector appears once per 5% appearance probability.
At this time, when a motion vector is used as code data by the conventional method, the code amount of the difference vector is
(1 bit x 6 times) [When difference vector is 0]
+ (3bit x 3 times) [Difference vector 1/4]
+ (3bit x 3 times) [Difference vector 2/4]
+ (5bit x 3 times) [Difference vector 3/4]
+ (5bit x 5 times) [Difference vector 1 o'clock]
= 64bit
In addition, the code amount of the position information of the neighboring blocks from which the difference vector is obtained
(1 bit x 6 times) [When difference vector is 0]
+ (3bit x 3 times) [Difference vector 1/4]
+ (3bit x 3 times) [Difference vector 2/4]
+ (1 bit x 3 times) [Difference vector 3/4]
+ (3bit x 5 times) [Difference vector 1 o'clock]
= 42bit
A total of 106 bits is the code amount of the difference vector.

On the other hand, when the motion vector is code data in the present invention, the code amount of the conversion difference vector is
(1 bit x 6 times) [When difference vector is 0]
+ (3bit x 3 times) [Difference vector 1/4]
+ (3bit x 3 times) [Difference vector 2/4]
+ (3bit x 3 times) [Difference vector 3/4]
+ (3bit x 5 times) [Difference vector 1 o'clock]
= 48bit
Also, the code amount of the vector accuracy information is
(1 bit x 3 times) [Difference vector 1/4]
+ (2bit x 3 times) [Difference vector 2/4]
+ (2bit x 3 times) [Difference vector 3/4]
= 12 bits A total of 60 bits is the code amount of the motion vector.

Therefore, the present embodiment can perform motion compensation with the same accuracy with a code amount of 46 bits less than the conventional example.
Next, FIG. 15 is a second example in which the appearance probability of the motion vector at the time of actual motion compensation is assigned to each assigned code data as in the description of FIG. This is an example when the appearance probability of a motion vector with ¼ pixel accuracy is higher than that with other fractional pixel accuracy motion vectors. At this time, attention is paid to the fact that the bit length of the assigned code data assigned to the motion vector of 1/4 pixel accuracy is shorter than the bit length of the assigned code data of other fractional accuracy motion vectors.

In this example, the code amount of the motion vector at the time of actual motion compensation is calculated and compared between the conventional example and this example.
Assume that 20 motion vectors appear according to the appearance probability. Therefore, the motion vector appears once per 5% appearance probability.

At this time, when a motion vector is used as code data by the conventional method, the code amount of the difference vector is
(1 bit x 8 times) [Difference vector 0]
+ (3bit x 2 times) [Difference vector 1/4]
+ (3bit x 1 time) [Difference vector 2/4]
+ (5bit x 1 time) [Difference vector 3/4]
+ (5bit x 8 times) [Difference vector 1 o'clock]
= 62bit
In addition, the code amount of the position information of the neighboring blocks from which the difference vector is obtained
(1 bit x 8 times) [Difference vector 0]
+ (3bit x 2 times) [Difference vector 1/4]
+ (3bit x 1 time) [Difference vector 2/4]
+ (1 bit x 1 time) [Difference vector 3/4]
+ (3bit x 5 times) [Difference vector 1 o'clock]
= 33bit
A total of 95 bits is the code amount of the motion vector.

On the other hand, when the motion vector is code data in the present invention, the code amount of the conversion difference vector is
(1 bit x 8 times) [Difference vector 0]
+ (3bit x 2 times) [Difference vector 1/4]
+ (3bit x 1 time) [Difference vector 2/4]
+ (3bit x 1 time) [Difference vector 3/4]
+ (3 bits x 8 times) [Difference vector 1 o'clock]
= 44bit
Also, the code amount of the vector accuracy information is
(1 bit x 2) [Difference vector 1/4]
+ (2bit x 1 time) [Difference vector 2/4]
+ (2bit x 1 time) [Difference vector 3/4]
= 6 bits A total of 50 bits is the code amount of the motion vector. Therefore, the present embodiment can perform motion compensation with the same accuracy with a code amount of 45 bits less than the conventional example.

Thus, according to the present embodiment, it can be seen that an increase in the code amount of the motion vector can be significantly suppressed as compared with the conventional case.
It can also be seen that an increase in the code amount of the motion vector can be further suppressed by assigning a short bit length code to the vector accuracy information having a high appearance probability in consideration of the bias of the vector accuracy information.

Next, an example will be described in which the motion compensation with 1/4 pixel accuracy is expanded to motion compensation with higher accuracy such as 1/8 pixel accuracy, and actually the operation with 1/4 pixel accuracy is performed.
FIG. 16 is a table showing the relationship between a conventional difference vector and assigned code data assigned to the difference vector when expanded to 1/8 pixel accuracy.

  FIG. 17 is a table showing the relationship between the conventional difference vector, the conversion difference vector of this embodiment corresponding to the difference vector, and the assigned code data assigned to the conversion difference vector when expanded to 1/8 pixel accuracy. It is. FIG. 18 is a table showing a conventional difference vector when expanded to 1/8 pixel accuracy, vector accuracy information of this embodiment corresponding to the difference vector, and allocation code data assigned to the vector accuracy information.

  FIG. 19 is an example in which the appearance probability of a motion vector at the time of actual motion compensation is assigned to the assigned code data allocated with 1/8 pixel accuracy expanded as described above. Since this example is an example in which the motion compensation accuracy is lowered to 1/4 pixel accuracy, the motion vectors of 1/8, 3/8, 5/8, and 7/8 are not actually used, so the appearance probability is 0. %It has become.

Under such conditions, the code amount of the motion vector at the time of actual motion compensation is calculated and compared between the conventional example and this embodiment.
Assume that 20 motion vectors appear according to the appearance probability. Therefore, the motion vector appears once per 5% appearance probability.

At this time, when a motion vector is used as code data by the conventional method, the code amount of the difference vector is
(1 bit x 8 times) [Difference vector 0]
+ (3bit x 0) [Difference vector 1/8]
+ (3bit x 2 times) [Difference vector 2/8]
+ (5bit x 0) [Difference vector 3/8]
+ (5bit x 1 time) [Difference vector 4/8]
+ (5bit x 0) [Difference vector 5/8]
+ (5bit x 1 time) [Difference vector 6/8]
+ (7bit x 0) [Difference vector 7/8]
+ (7bit x 8 times) [Difference vector 1 o'clock]
= 80bit
In addition, the code amount of the position information of the neighboring blocks from which the difference vector is obtained
(1 bit x 8 times) [Difference vector 0]
+ (3bit x 2 times) [Difference vector 2/8]
+ (3bit x 1 time) [Difference vector 4/8]
+ (1 bit x 1 time) [Difference vector 6/8 o'clock]
+ (3bit x 5 times) [Difference vector 1 o'clock]
= 33bit
A total of 113 bits is the code amount of the motion vector.

On the other hand, when the motion vector is code data in the present invention, the code amount of the conversion difference vector is
(1 bit x 8 times) [Difference vector 0]
+ (3bit x 0) [Difference vector 1/8]
+ (3bit x 2 times) [Difference vector 2/8]
+ (3bit x 0) [Difference vector 3/8]
+ (3bit x 1 time) [Difference vector 4/8]
+ (3bit x 0) [Difference vector 5/8 o'clock]
+ (3bit x 1 time) [Difference vector 6/8]
+ (3bit x 0) [Difference vector 7/8]
+ (3 bits x 8 times) [Difference vector 1 o'clock]
= 44bit
Also, the code amount of the vector accuracy information is
(5bit x 0) [Difference vector 1/8]
+ (1bit x 2 times) [Difference vector 2/8]
+ (5bit x 0) [Difference vector 3/8]
+ (3bit x 1 time) [Difference vector 4/8]
+ (5bit x 0) [Difference vector 5/8]
+ (3bit x 1 time) [Difference vector 6/4]
+ (5bit x 0) [Difference vector 7/8]
= 8 bits A total of 52 bits is the code amount of the motion vector. Therefore, the present embodiment can perform motion compensation with the same accuracy with a code amount of 61 bits less than the conventional example.

  Compared with the code amount obtained in the example of FIG. 15, in the conventional case, the code amount when motion compensation is performed with 1/4 pixel accuracy from the beginning is 95 bits. Becomes 113 bits, and the code amount of the motion vector is considerably increased.

  On the other hand, according to the present embodiment, the code amount when motion compensation is performed with 1/4 pixel accuracy from the beginning is 50 bits, whereas in this case, the code amount is 52 bits, and the code amount hardly increases. You can see that

The overall operation of the moving picture coding / decoding apparatus according to an embodiment of the present invention using the motion vector code amount suppression means described above will be described below with reference to the drawings.
The operation of the moving picture coding apparatus according to the present embodiment will be described with reference to FIG.
[Encoding device]
As shown in FIG. 1, the moving picture coding apparatus according to the present embodiment includes an image memory 101, an image interpolation unit 102, a motion vector detection unit 103, a motion compensation unit 104, an orthogonal transformation unit 105, and a quantization unit. 106, an inverse quantization unit 107, an inverse orthogonal transform unit 108, a motion vector memory 109, a difference vector calculation unit 110, a motion vector code amount suppression unit 111, and an entropy encoding unit 112. Yes.

  The image memory 101 stores a predetermined number (for example, one screen) of decoded image blocks obtained by encoding and re-decoding each block of the encoding target image for use in encoding each block of the next encoding target image. To do.

  The image interpolation unit 102 combines the plurality of decoded image blocks to generate a reference image, and further interpolates between the actual pixels of the reference image with a predetermined number (three in this description) of interpolation pixels. Is generated.

  The motion vector detection unit 103 detects a motion vector of the encoding target image block using the reference interpolation image generated by the image interpolation unit 102. The detected motion vector is supplied to the motion compensation unit 104 and the difference vector calculation unit 110 and stored in the motion vector memory 109.

  The motion compensation unit 104 performs motion compensation using the motion vector detected by the motion vector detection unit 103 and the reference interpolation image generated by the image interpolation unit 102 to generate a predicted image block.

The orthogonal transform unit 105 performs orthogonal transform on the difference signal between the encoding target image block and the predicted image block, and supplies the transformed orthogonal transform signal to the quantization unit 106.
The quantization means 106 quantizes the difference signal transformed by the orthogonal transformation means 105 and supplies the quantized quantized signal to the inverse quantization means 107 and the entropy coding means 112.

The inverse quantization means 107 inversely quantizes the quantized signal quantized by the quantization means 106 and supplies the inversely quantized inverse quantized signal to the inverse orthogonal transform means 108.
The inverse orthogonal transform means 108 performs inverse orthogonal transform on the inversely quantized signal inversely quantized by the inverse quantization means 107 to generate an inverse orthogonal transform signal.

The motion vector memory 109 stores a predetermined number (for example, four) of motion vectors detected by the motion vector detecting means 103.
The difference vector calculation unit 110 calculates a prediction vector from the motion vectors of the peripheral blocks of the encoding target image block stored in the motion vector memory 109. Next, a difference (hereinafter referred to as a difference vector) between the calculated prediction vector and the motion vector detected by the motion vector detection unit 103 is calculated. Then, the calculated difference vector is supplied to the motion vector code amount suppression unit 111.

  As a method of calculating the prediction vector, for example, there is a method of taking the median value (median) of each motion vector of MV1 to MV3 in FIG. For example, the median is (−2, 4) when MV1 is (−3,1), MV2 is (−2,5), and MV3 is (2,4). The calculation method of the prediction vector is not limited to the above method, and any method may be used as long as it is a method for calculating the representative value of the motion vector of each of the surrounding blocks.

The operation of the motion vector code amount suppression means 111 is as described in the best mode for carrying out the invention, but will be briefly described again.
If the value of the difference vector calculated by the difference vector calculation means 110 is integer pixel precision, the difference vector value is multiplied by a predetermined even number (2 in this description) to be converted into an even value. Further, the converted even value (integer precision conversion difference vector) is converted into predetermined allocation code data according to the allocation table of FIG. 12 and supplied to the entropy encoding means 112.

  If the value of the difference vector is fractional pixel precision, the integer part of the difference vector value is multiplied by a predetermined even number (2 in this description) and further converted to an odd value obtained by adding a predetermined odd value (1 in this description). At the same time, the fractional value after the decimal point is generated as vector accuracy information separately from the difference vector. Then, the converted odd value (conversion difference vector with fractional precision) is converted into predetermined allocation code data according to the allocation table of FIG. 12 and supplied to the entropy encoding means 112. Further, the vector accuracy information is converted into predetermined allocation code data according to the allocation table of FIG.

  In this description, the integer pixel precision difference vector is converted to an even value, and the integer part of the fraction pixel precision difference vector is converted to an odd value. The integer part of may be converted to an even value.

  The entropy encoding unit 112 multiplexes the quantized signal quantized by the quantizing unit 106 and each assigned code data generated by the motion vector code amount suppression unit 111 to perform entropy encoding, and generates the generated code string data externally. Output to.

  Here, the operation of the motion vector code amount suppression means 111 will be described again using the flowchart of FIG. FIG. 9 is a flowchart of the motion vector code amount suppression unit 111. First, it is determined whether or not the pixel accuracy of the difference vector calculated by the difference vector calculation means 110 is fractional accuracy (S11). If the accuracy is integer accuracy instead of fractional accuracy, the difference vector is converted to an even value and transmitted ( S12). In the case of fractional accuracy, the difference vector is converted into an odd value and transmitted (S13). In the case of fractional accuracy, vector accuracy information is transmitted separately from the difference vector (S14).

Encoding is performed by the above processing.
[Decryption device]
Next, the operation of the moving picture decoding apparatus according to the present embodiment will be described with reference to FIG.

  As shown in FIG. 2, the moving picture decoding apparatus according to the present embodiment includes an entropy decoding unit 201, an image memory 202, an image interpolation unit 203, a difference vector calculation unit 204, a motion vector calculation unit 205, a motion The vector memory 206, motion compensation means 207, inverse quantization means 208, and inverse orthogonal transform means 209 are configured.

  The entropy decoding unit 201 performs entropy decoding on the input code string data. Then, each allocation code data obtained by decoding is supplied to the difference vector calculation means 204 and the decoded quantized signal is supplied to the inverse quantization means 208.

The inverse quantization unit 208 dequantizes the quantized signal decoded by the entropy decoding unit 201 to generate an orthogonal transform signal, and supplies the orthogonal transform signal to the inverse orthogonal transform unit 209.
The inverse orthogonal transform unit 209 performs inverse orthogonal transform on the orthogonal transform signal inversely quantized by the inverse quantization unit 208 to generate an inverse orthogonal transform signal.

The difference vector calculation means 204 generates a conversion difference vector and vector accuracy information for each assigned code data decoded by the entropy decoding means 201 in accordance with the assignment tables shown in FIGS.

If the generated conversion difference vector is an even value (conversion difference vector with integer pixel accuracy), the difference vector is calculated by dividing by a predetermined integer value (2 in this description).
When the generated conversion difference vector is an odd value (conversion difference vector with fractional pixel precision), 1 is subtracted and then divided by a predetermined integer value (2 in this description) to calculate the integer part of the difference vector, and further the motion The vector accuracy information is added to calculate a difference vector.

For example, when the conversion difference vector is 3 and the vector accuracy information is 1/4, the difference vector is 1 and 1/4 (or 5/4). In addition, when the conversion difference vector is 6, the difference vector is 3. In this description, the integer vector precision difference vector is converted to an even value, and the fractional pixel precision difference vector is converted to an odd value. Of course, the same processing can be realized even if the difference vector of integer pixel accuracy is converted to an odd value and the integer part of the difference vector of fractional pixel accuracy is converted to an even value.

  The motion vector calculation means 205 calculates the median (median) of the motion vectors of the peripheral blocks of the decoding target image block stored in the motion vector memory 206. This median value becomes the prediction vector of the decoding target image block. Then, the motion vector is calculated by adding the difference vector calculated by the difference vector calculating means 204 to the prediction vector. The calculated motion vector is supplied to the motion vector memory 206 and motion compensation means 207.

The motion vector memory 206 stores the motion vectors calculated by the motion vector calculation means 205 for a predetermined number of blocks.
The motion compensation unit 207 performs motion compensation using the motion vector of the decoding target image block calculated by the motion vector calculation unit 205 and the reference interpolation image generated by the image interpolation unit 203 to generate a predicted image block.

The image memory 202 stores a predetermined number of decoded image blocks that have already been decoded for use in decoding each block of the next decoding target image.
The image interpolation means 203 combines the plurality of decoded image blocks to generate a reference image, and further interpolates between the actual pixels of the reference image with a predetermined number (three in this description) of interpolation pixels. Is generated.

  The inverse orthogonal transform signal generated by the inverse orthogonal transform unit 209 and the predicted image block generated by the motion compensation unit 207 are added to generate a decoded image block, which is supplied to the image memory 202 and the decoded image block is converted into an image signal. Output to the outside.

Decoding is performed by the above processing.
As described above, according to the moving picture coding / decoding apparatus of the present embodiment, it is possible to significantly suppress an increase in the amount of motion vector codes as compared with the conventional art.

  In addition, in consideration of the bias of the vector accuracy information, it is possible to further suppress the increase in the code amount of the motion vector by assigning a short bit length code to the vector accuracy information having a high appearance probability.

  In addition, when motion compensation with low accuracy is performed with high motion compensation accuracy, the amount of motion vector code is reduced to the same low amount as when motion compensation accuracy is initially set low. Is possible.

Furthermore, the present invention includes a program for causing a computer to realize the functions of the above-described moving picture coding apparatus. These programs may be read from a recording medium and loaded into a computer, or may be transmitted via a communication network and loaded into a computer.

It is a block diagram of one embodiment of a moving picture coding apparatus of the present invention. It is a block diagram of one embodiment of a moving picture decoding apparatus of the present invention. It is a block diagram of the conventional moving image encoder. It is a block diagram of the conventional moving image decoding apparatus. It is the figure which showed the mode of the 1/2 pixel precision interpolation pixel. It is the figure which showed the mode of the 1/4 pixel precision interpolation pixel. It is a flowchart of the conventional difference vector calculation means 310. It is the figure which showed the mode of the conventional prediction vector selection. It is a flowchart of the motion vector code amount suppression means 111. It is the table | surface which showed the assignment code to the difference vector in the case of the conventional 1/4 pixel precision. It is the table | surface which showed the assignment code to the conventional surrounding block position information. It is a table | surface which shows the example of the allocation code | symbol to the conversion difference vector in the case of 1/4 pixel precision of a present Example. It is a table | surface which shows the example of the allocation code | symbol to the vector accuracy information in the case of 1/4 pixel accuracy of a present Example. 6 is a table showing a first specific example of the appearance probability and the number of appearances of a motion vector in the conventional case and the present embodiment in the case of ¼ pixel accuracy. It is the table | surface which showed the 2nd specific example of the appearance probability of a motion vector, and the frequency | count of appearance with the past in the case of 1/4 pixel precision, and a present Example. It is the table | surface which showed the assignment code to the difference vector in the case of the conventional 1/8 pixel precision. It is a table | surface which shows the example of the allocation code | symbol to the conversion difference vector in the case of 1/8 pixel precision of a present Example. It is a table | surface which shows the example of the allocation code | symbol to the vector precision information in the case of 1/8 pixel precision of a present Example. 5 is a table showing a specific example of the appearance probability and the number of appearances of a motion vector in the conventional case and the present embodiment in the case of 1/8 pixel accuracy.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 Image memory 102 Image interpolation means 103 Motion vector detection means 104 Motion compensation means 105 Orthogonal transformation means 106 Quantization means 107 Inverse quantization means 108 Inverse orthogonal transformation means 109 Motion vector memory 110 Difference vector calculation means 111 Motion vector code amount suppression means DESCRIPTION OF SYMBOLS 112 Entropy encoding means 201 Entropy decoding means 202 Image memory 203 Image interpolation means 204 Motion vector memory 205 Motion vector calculation means 206 Motion compensation means 207 Inverse quantization means 208 Inverse orthogonal transform means

Claims (2)

An encoding target image is divided into encoding target image blocks each including a two-dimensional array of a plurality of pixels, and motion compensation prediction encoding, orthogonal transform encoding, quantization, and entropy encoding are sequentially performed in units of the encoding target image blocks. And using a reference interpolated image in which interpolated pixels are inserted between pixels of an image obtained by re-decoding an image that has already been encoded as an image to be referred to when performing the motion compensation predictive encoding In the device
The prediction vector of the encoding target image block is calculated from the motion vectors of the plurality of image blocks that have already been encoded adjacent to the encoding target image block, respectively, and the calculated prediction vector and the actual encoding are encoded. A difference vector calculation means for calculating a difference vector that is a difference from a motion vector of the encoding target image block obtained by performing;
If the difference vector calculated by the difference vector calculating means is an integer pixel precision difference vector indicating the actual pixel position of the reference interpolated image, either an even number or an odd number according to the integer value The first conversion difference vector of the type is generated, first code data is generated according to the generated first conversion difference vector, and the difference vector calculated by the difference vector calculation means is the reference interpolation In the case of a differential vector with fractional pixel precision indicating the position of an interpolated pixel in the image, the first converted differential vector according to an integer value obtained by dividing the numerator of the differential vector precision with fractional pixel by the denominator is used. Generate the second conversion difference vector of the other type different from the odd or even number, and generate the second code data according to the generated second conversion difference vector A motion vector code amount control means for said remainder value obtained molecules difference vector fractional pixel precision by dividing the denominator and vector accuracy information, to generate a third code data corresponding to the vector accuracy information,
The entropy code using the first code data, the second code data, and the third code data generated by the motion vector code amount suppression unit and the encoded data obtained by encoding the encoding target image Entropy encoding means for generating code string data and outputting the code string data to the outside;
A moving picture encoding apparatus comprising: A moving image for decoding the encoding target image by performing entropy decoding, inverse quantization, inverse orthogonal transform decoding, motion compensation predictive decoding, using code string data generated by the moving image encoding device as an input In the decryption device,
The entropy decoding is performed with the code string data as input, the encoded data of the encoding target image, the first code data, the second code data, and the third code data. Generating entropy decoding means;
Generating the first conversion difference vector according to the first code data generated by the entropy decoding means, calculating an integer pixel precision difference vector according to the generated first conversion difference vector; The second conversion difference vector corresponding to the second code data generated by the entropy decoding means is generated, and an integer value of the difference vector with fractional pixel accuracy corresponding to the generated second conversion difference vector is obtained. Calculating, generating the vector accuracy information according to the third code data generated by the entropy decoding means, and adding the calculated integer value of the difference vector of the fractional pixel accuracy and the vector accuracy information A difference vector calculating means for calculating a difference vector with fractional pixel accuracy;
The prediction vector of the decoding target image block is calculated from the motion vectors of a plurality of image blocks that have already been decoded and are adjacent to the image block to be decoded this time, and the calculated prediction vector and the calculated vector The motion vector actually used for decoding the decoding target image block is generated by adding the difference vector of integer pixel accuracy or the difference vector of fractional pixel accuracy calculated by the difference vector calculation means, A motion vector calculating means;
A moving picture decoding apparatus comprising:

Images