JP2003230150A - Method for coding dynamic image, program thereof, and recording medium recording this program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a motion vector of good coding efficiency and to obtain a reference image. <P>SOLUTION: An MR calculator 101 obtains pixel value totals M and R of overall input image. A reference image memory 104 stores the input image of a past frame. An M storage memory 102 stores M corresponding to the past input image. An MAD calculator 110 calculates the MAD between the reference image and the input image. An amount-of-motion vector coding evaluator 106 evaluates an amount of coding of the motion vector of a present macro block, by using motion vector information already obtained by the circumferential macro block. An amount-of-reference image designating evaluator 107 evaluates an amount of coding information designated by a searching frame. An SEA realizing unit 105 realizes an SEA method by using an inequality of motion searching from the R and the M and evaluated values B and C. A decoding image memory 109 stores decoded image of the past frame. An interframe predicting coding unit 103 forms the predicted image by sing the coded image and codes the input image. The unit 108 codes the coded data. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding method, and more particularly to a moving picture coding method for performing a motion search for each divided area (current area) of an input image.

[0002]

2. Description of the Related Art In moving picture coding, a motion-compensated interframe predictive coding method is generally used to improve coding efficiency. This method encodes a difference (referred to as a prediction error) in image information between corresponding pixels between an input image and a reference image created from a decoded image of another frame that has been encoded before. . The reference image can be created by using the previous frame, creating it from the preceding and succeeding frames, or affine transforming these frames.

In coding the prediction error, for example, the prediction error is subjected to orthogonal transform such as DCT (discrete cosine transform), and its coefficient is quantized and then variable length coded. Alternatively, there is a method of performing DPCM coding without performing orthogonal transform. In order to improve the coding efficiency, when the prediction error of the translated position is smaller than the prediction error between pixels at the same position in the screen, the prediction error of the translated position is coded. The translation amount is encoded as a motion vector.

The process of obtaining the amount of parallel movement is called motion search. Generally, the screen is divided into small areas and a motion vector is obtained for each area. For example, in the video coding method H.263, the blocks are separated by macroblocks of 16 pixels in the vertical and horizontal directions, and the motion vector is obtained for each macroblock. As the evaluation function of the prediction error for all the pixels in the area, the sum of the absolute value differences of the image information of all the pixels in the area,
Their dispersion or the like is used. If a motion vector with a small value is selected, the coding efficiency of the prediction error is increased, so that the motion vector with the smallest value is obtained. In addition, the larger the search range, the more motion vector candidates increase, and thus it is possible to obtain a motion vector with high coding efficiency. For example, in H.263, the search range is a position that is shifted by a maximum of 16 pixels in the vertical and horizontal directions due to the restriction on the expression of the motion vector. This state is shown in FIG. While shifting the search range pixel by pixel, the absolute value difference sum and variance are calculated, and the position where these values are minimum is obtained. But,
The larger the search range, the greater the amount of calculation for motion search.

When creating a reference image, there are a method of using an already encoded input image and a method of using a locally decoded image of a past frame created at the time of motion compensation.

Various techniques are used to reduce the amount of calculation for motion estimation. In particular, the SEA method is a method of reducing the calculation amount of motion search without impairing the search accuracy.

The SEA method will be outlined. The value of the pixel placed at the position (i + a, j + b) of the input image (time t) is f
If (i + a, j + b, t), the original pixel position (a,
b), the absolute value difference sum MAD for the reference frame (time t-1) at the search position (x + a, y + b)
(X + a, y + b) is calculated by the following equation (1).

In the SEA method, the search position (x + a, y +
In b), the calculated search position (m + a, n)
+ B) absolute value difference sum MAD (m + a, n + b), the search position (x +) obtained by the following equation (2):
The absolute value difference sum is obtained only when M (x + a, y + b) of (a, y + b) satisfies the following inequality (3).
R (a, b) is a value calculated by the following equation (4). Therefore, the number of times the absolute value difference sum is calculated can be reduced.

[0009]

[Equation 1]

For example, a rectangular search range (-w <= x <=
When performing the motion search within w, -w <= y <= w), the motion search is performed in the following procedure by using the determination by the above inequality (3). This motion vector determination flow is shown in FIG.

First, R (a, b) and M (x + a, y +
The value of b) is obtained at all positions (a, b) in the frame (Step 1). After that, the motion vector is obtained for each macroblock by using the inequality (3).

The initial position of the search is the upper left (m + a,
If n + b) = (− w + a, −w + b), first, MAD (m + a, n +) at the initial position (m + a, n + b).
b) is calculated (Step 2).

Next, the value of M (x + a, y + b) is calculated while shifting the search position to the right one pixel at a time, and MAD is calculated.
The position (x ₀ + a, y ₀ + b) that satisfies the inequality (3) is obtained for (m + a, n + b) (Step 3). At this time, next to the search position at the rightmost end, the leftmost end one line below is set as the search position.

A position (x that satisfies the inequality (3)₀+ A, y₀
If + b) is found, MAD (x ₀+ A, y₀+ B)
Ask. MAD (x₀+ A, y₀+ B) value and MAD (m
The values of + a, n + b) are compared (Step 4). MAD
(X₀+ A, y₀The value of + b) is MAD (m + a, n + b)
If the value is greater than or equal to, the process returns to Step 3 and the inequality (3) is returned.
Find a position that satisfies.

The value of MAD (x ₀ + a, y ₀ + b) is MAD
If smaller than the value of (m + a, n + b), MAD
The value of (m + a, n + b) is MAD (x ₀ + a, y ₀ + b)
Replace with the value of (Step 5), return to Step 3,
Find a position that satisfies the inequality (3).

In this way, St is reached to the last search position.
Repeat ep3 and Step4. And finally Ste
The position where MAD is replaced in p5 is set as a motion vector.

Next, in the coding of the motion vector, the difference between the motion vector (x ₀ , y ₀ ) obtained by the motion search and the motion vector of the surrounding area is obtained, and this difference information is entropy coded. . For example, when the current area and the surrounding area have the positional relationship shown in FIG. 4, the motion vectors of the surrounding area are (x ₁ , y ₁ ), (x ₂ , y ₂ ), (x ₃ , y ₃ ).
Then, the encoded motion vector information (x, y) of the current area can be obtained by the equation (5).

[0018]

[Equation 2]

Further, like H.263 Annex U,
There is also a method in which the encoding side and the decoding side are provided with reference image memories capable of accumulating images for a plurality of frames, and a reference image is selected from the reference image memory to create a predicted image. In this case, in addition to the motion vector information, it is necessary to code together information ref for designating the reference image used on the coding side.

For example, it is possible to specify the reference image by setting ref = 1 when the immediately preceding frame is used as the reference image, and by setting ref = 2 when setting the two frames before the reference image. These reference images can be designated for each small area such as a macroblock. On the encoding side, a motion search is performed in each frame stored in the reference image memory,
A frame from which a motion vector with the smallest prediction error is obtained is obtained. The ref is variable-length coded, and the smaller the value, the smaller the code length is assigned.

As the reference image used for the motion search, either a method using a past input image or a method using a past decoded image can be used.

[0022]

According to the conventional SEA method, the position (x ₀ + a, y ₀ + b) having the smallest MAD with respect to the position (a, b) can be obtained. However, since (x, y) obtained by using Expression (5) is encoded from this motion vector (x ₀ , y ₀ ), the code amount for each area is always the largest at the position where the absolute value difference sum is the smallest. It doesn't always get smaller. That is, in addition to the position obtained by the conventional SEA method, the sum of the absolute value difference sum and the code amount evaluation value of the motion vector (x, y), that is, the code amount of the prediction error and the code of the motion vector (x, y) The position (x
₀ + a, y ₀ + b) may exist and the conventional SEA
The method could not find the motion vector with the highest coding efficiency.

Further, in the method of selecting a reference image from a plurality of frames, the information designating the reference image is required separately from the motion vector. Therefore, in addition to the position obtained by the conventional SEA method, the absolute value difference sum is obtained. And the sum of the code amount evaluation value of the motion vector (x, y) and the reference image and the code amount evaluation value of the information to be specified, that is, the code amount of the prediction error, the code amount of the motion vector (x, y), and the reference image There may be a combination of the position (x ₀ + a, y ₀ + b) and the reference image in which the total code amount of the information to be used is the smallest. Therefore, the conventional SEA method cannot obtain the motion vector and the reference image with the highest coding efficiency.

An object of the present invention is to provide a moving picture coding method that solves the above problems, a program for this method, and a recording medium recording this program.

[0025]

In order to solve the above problems, the present invention is characterized by the following moving picture coding method, a program for this method, and a recording medium recording this program.

(Invention of Method) (1) A moving image coding method for performing a motion search for each current region obtained by dividing an input image, and between a current region pixel and a reference image pixel at a search position A in a reference image. Absolute value difference W of
And an absolute value difference sum calculating means for calculating the sum of the pixel values in the current area, and a search position A for calculating the sum R of the pixel values in the current area.
A reference image pixel calculation means for obtaining the sum M of the pixel values of the same number of pixels at a search position B different from the above, and an evaluation value of the motion vector code amount when the search position A or B is a motion vector. The motion vector code amount evaluating means for obtaining VA or VB, the absolute value of the difference between the search position R obtained by the current area pixel calculating means and the search position M obtained by the reference image pixel calculating means, and the motion A search position evaluation means for adding the evaluation value VB obtained by the vector code amount evaluation means to obtain a search position evaluation value D, an absolute value difference sum W obtained by the absolute value difference sum calculation means, and the motion vector code amount evaluation. And an absolute value difference sum evaluation means for adding the evaluation value VA obtained by the means to obtain an absolute value difference sum evaluation value E. The search position evaluation value D obtained by the position evaluation means is compared with the absolute value difference sum evaluation value E obtained by the absolute value difference sum evaluation means, and if the value of D is equal to or less than the value E, the absolute value is calculated. It is characterized in that the difference sum calculation means obtains an absolute value difference sum W between the current region pixel and the reference image pixel at the search position B.

(2) A moving picture coding method for selecting a reference picture from a reference picture memory capable of accumulating a plurality of frames (P1 to Pn) for each current area obtained by dividing an input picture and performing a motion search, At the search position A within the reference image Pi,
Absolute value difference sum calculation means for calculating the absolute value difference sum W between the current area pixel and the reference image pixel, and the sum R of the pixel values in the current area
A pixel in the current area for calculating a pixel in the current area, and a pixel in the reference image for calculating a total M of pixel values of the same number of pixels as the current area at a search position B in a reference image Pj different from the reference image Pi; Evaluation value VA or VB of the code amount of the motion vector when the search position A or B is used as the motion vector
, A reference image designation code amount evaluation unit for obtaining an evaluation value Ri or Rj of a code amount for encoding the information designating the reference image Pi or Pj, and the current area pixel calculation unit. In the absolute value of the difference between the total R obtained in step S1 and the total M obtained in the reference image pixel calculation means, the evaluation value VB obtained by the motion vector code amount evaluation means and the reference image designated code amount evaluation means are obtained. The evaluation value Rj obtained by the motion vector code amount evaluation means and the search position evaluation means for obtaining the search position evaluation value D by adding the obtained evaluation value Rj, and the absolute value difference sum W obtained by the absolute value difference sum calculation means. And an absolute value difference sum evaluation unit for adding the evaluation value Ri obtained by the reference image designation code amount evaluation unit to obtain an absolute value difference sum evaluation value E. In the image coding method, the search position evaluation value D obtained by the search position evaluation means is compared with the absolute value difference sum evaluation value E obtained by the absolute value difference sum evaluation means, and the value of D is E. When the value is less than or equal to the value of, the absolute value difference sum calculating means calculates the absolute value difference sum W between the current region pixel and the reference image pixel at the search position B in the reference image Pj.
It is characterized by seeking.

(Invention of Program) (3) This is a program of a moving picture coding method for performing a motion search for each current area obtained by dividing an input image, and presents a current area pixel and a reference image pixel at a search position A in the reference image. Absolute value difference sum calculation procedure for obtaining the absolute value difference sum W between them, a current area pixel calculation procedure for obtaining a total R of pixel values in the current area, and a current area at a search position B different from the search position A. And a reference image pixel calculation procedure for obtaining a sum M of pixel values of the same number of pixels, and a motion vector code amount evaluation for obtaining a motion vector code amount evaluation value VA or VB when the search position A or B is a motion vector. Procedure, and the absolute value of the difference between the search position R obtained by the pixel calculation procedure in the current region and the search position M obtained by the pixel calculation procedure in the reference image, obtained by the motion vector code amount evaluation procedure. Evaluation value VB
Is added to obtain a search position evaluation value D, and an absolute value difference sum W obtained in the absolute value difference sum calculation procedure and an evaluation value VA obtained in the motion vector code amount evaluation procedure are added. Absolute value difference sum evaluation procedure for obtaining absolute value difference sum evaluation value E, and search position evaluation value D obtained in the search position evaluation procedure and absolute value difference sum evaluation value E obtained in the absolute value difference sum evaluation procedure And when the value of D is less than or equal to the value of E, the absolute value difference sum calculation procedure determines the absolute value difference sum W between the current region pixel and the reference image pixel at the search position B. It is characterized by having.

(4) A program of a moving picture coding method for performing a motion search by selecting a reference picture from a reference picture memory capable of accumulating a plurality of frames (P1 to Pn) of the picture for each current area obtained by dividing the input picture. Yes, the absolute value difference sum calculation procedure for obtaining the sum W of absolute value differences between the current region pixel and the reference image pixel at the search position A in the reference image Pi, and the current region for obtaining the sum R of the pixel values in the current region Pixel calculation procedure, a reference image pixel calculation procedure for obtaining a total M of pixel values of the same number of pixels as the current region at a search position B in a reference image Pj different from the reference image Pi, and the search position A or B. Where V is the motion vector, the evaluation value V of the motion vector code amount
A motion vector code amount evaluation procedure for obtaining A or VB;
A reference image designated code amount evaluation procedure for obtaining an evaluation value Ri or Rj of a code amount for encoding information designating the reference image Pi or Pj, and a total R obtained in the pixel calculation procedure in the current region and the reference image The absolute value of the difference from the total M calculated in the pixel calculation procedure is added with the evaluation value VB calculated in the motion vector code quantity evaluation procedure and the evaluation value Rj calculated in the reference image designation code quantity evaluation procedure to search position. In the search position evaluation procedure for obtaining the evaluation value D and the absolute value difference sum W obtained in the absolute value difference sum calculation procedure, the evaluation value VA obtained in the motion vector code quantity evaluation procedure and the reference image designation code quantity evaluation procedure are used. Absolute value difference sum evaluation procedure for calculating the absolute value difference sum evaluation value E by adding the calculated evaluation values Ri, and the search position evaluation value D and the absolute value difference calculated in the search position evaluation procedure. The magnitudes of the absolute value difference sum evaluation values E obtained in the sum evaluation procedure are compared, and when the value of D is equal to or less than the value of E, the absolute value difference sum calculation procedure is performed at the search position B in the reference image Pj. And a step of obtaining an absolute value difference sum W between the region pixel and the reference image pixel.

(Invention of Recording Medium) (5) A recording medium in which a program of a moving image coding method for performing a motion search for each current region obtained by dividing an input image is recorded, and the current region at the search position A in the reference image is recorded. An absolute value difference sum calculation procedure for obtaining an absolute value difference sum W between a pixel and a reference image pixel, a current area pixel calculation procedure for obtaining a total R of pixel values in the current area, and a search different from the search position A Position B
In the reference image for calculating the sum M of the pixel values of the same number of pixels as in the current region, and the search position A or B.
Where V is the motion vector, the motion vector code amount evaluation procedure for obtaining the evaluation value VA or VB of the motion vector code amount, and the search position R obtained by the current region pixel calculation procedure.
And the search position M obtained by the pixel calculation procedure in the reference image
And the absolute value of the difference calculated by the absolute value difference sum calculation procedure, and the search position evaluation value D obtained by adding the evaluation value VB obtained by the motion vector code amount evaluation procedure to the absolute value of the difference. Absolute value difference sum evaluation procedure for obtaining the absolute value difference sum evaluation value E by adding the difference sum W and the evaluation value VA obtained in the motion vector code amount evaluation procedure, and the search position evaluation value obtained in the search position evaluation procedure D is compared with the absolute value difference sum evaluation value E obtained in the absolute value difference sum evaluation procedure, and when the value of D is equal to or less than the value of E, the absolute value difference sum calculation procedure is performed at the search position B. The procedure for obtaining the absolute value difference sum W between the current area pixel and the reference image pixel is recorded.

(6) A program of a moving picture coding method for performing a motion search by selecting a reference picture from a reference picture memory capable of accumulating a plurality of frames (P1 to Pn) of the picture for each current area obtained by dividing the input picture. The recording medium is a recording medium in which the absolute value difference sum calculation procedure for obtaining the absolute value difference sum W between the current area pixel and the reference image pixel at the search position A in the reference image Pi and the sum R of the pixel values in the current area R And a reference image Pj different from the reference image Pi.
In the reference image pixel calculation procedure for obtaining the sum M of the pixel values of the same number of pixels at the search position B in the reference area, and the evaluation value VA of the motion vector code amount when the search position A or B is the motion vector. Alternatively, a motion vector code amount evaluation procedure for obtaining VB, a reference image designation code amount evaluation procedure for obtaining an evaluation value Ri or Rj of a code amount for encoding information designating the reference image Pi or Pj, and the pixel in the current area The absolute value of the difference between the total R calculated in the calculation procedure and the total M calculated in the reference image pixel calculation procedure is set to the evaluation value VB calculated in the motion vector code quantity evaluation procedure and the reference image designated code quantity evaluation procedure. In the search position evaluation procedure of adding the evaluation value Rj obtained in step S3 to obtain the search position evaluation value D, and the absolute value difference sum W obtained in the absolute value difference sum calculation procedure The absolute value difference sum evaluation procedure for calculating the absolute value difference sum evaluation value E by adding the evaluation value VA calculated in the reference code quantity evaluation procedure and the evaluation value Ri calculated in the reference image designation code quantity evaluation procedure, and the search position. The search position evaluation value D obtained in the evaluation procedure is compared with the absolute value difference sum evaluation value E obtained in the absolute value difference sum evaluation procedure, and when the value of D is equal to or less than the value E, the absolute value difference sum is calculated. It is characterized in that the calculation procedure records a procedure for obtaining an absolute value difference sum W between the current area pixel and the reference image pixel at the search position B in the reference image Pj.

(Explanation of Action) According to the invention of claim 1 and the like,
It is possible to obtain the motion vector that minimizes the sum of the code amount evaluation value of the motion vector and MAD. Originally SEA
The method is based on the following equation (6) regarding the finally obtained position (x + a, y + b) and an arbitrary search position (m + a, n + b).
The inequality (3) is derived by using the relationship of (7).

[0033]

[Equation 3]

Assuming that the evaluation value of the code amount of the motion vector at the position (x + a, y + b) is B (x + a, y + b), the following expression is obtained from the expression (6).

[0035]

[Equation 4]

The position (x +) finally obtained from equation (7)
There is a relationship of the following expression (9) between a, y + b) and an arbitrary position (m + a, n + b).

[0037]

[Equation 5]

Therefore, the position (x + a, y + b) and the arbitrary position (m +) finally obtained by the relational expression of the following expression (10) are obtained.
It will be between a, n + b).

[0039]

[Equation 6]

By using the inequality (10) in the SEA method instead of the inequality (3), the sum of the motion vector code amount evaluation value and MAD, that is, the motion vector code amount and the prediction error code amount is the smallest. Can be obtained.

Further, according to the invention of claim 2 or the like, in the encoding method for selecting a reference image from a plurality of frames,
It is possible to obtain a combination of the reference image and the position where the sum of the MAD, the code amount of the motion vector, the reference image, and the designated information is the smallest. Using the following inequality (11) obtained in the same way as above instead of inequality (3), SEA
It can be realized by executing the method. Note that C (t, t
-K) is an evaluation value of the code amount of the information designating the reference image when the frame at time (t-k) is selected with respect to the input image (time t).

[0042]

[Equation 7]

Further, MAD (x + a, y + b, tk)
Is calculated by the following equation (12), and M (x + a, y
+ B, tk) is calculated by the following equation (13).

[0044]

[Equation 8]

FIG. 5 shows an operation flow when there are a plurality of reference images. In this operation flow, R
(A, b) and M (x + a, y + b, tk) (k> =
It is assumed that the value of 1) has already been obtained. The information ref that specifies the reference image is obtained for each area.

[0046]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described with reference to the drawings. In the present embodiment, a reference image memory for three frames is provided, motion search is performed in each frame, and the sum of the code amount of the motion vector, the code amount of the information designating the reference image, and the code amount of the prediction error is the most. A method of obtaining information that specifies a motion vector and a reference image that become smaller will be described. Information designating the motion vector and the reference image is obtained for each macroblock.

An outline of the image coding apparatus is shown in FIG. M (x + a, y + b, t) and R (a, b, of the entire input image
t), an MR calculation unit 101, a reference image memory 104 for accumulating input images of past frames, and M corresponding to past input images accumulated in the reference image memory.
Of the motion vector of the current macroblock by using the M storage memory 102 for storing the MAD, the MAD calculation unit 110 for calculating the MAD between the reference image and the input image, and the motion vector information already obtained in the surrounding macroblocks. A motion vector code amount evaluation unit 106 that evaluates the code amount, a reference image designation code amount evaluation unit 107 that evaluates the code amount of information that specifies the frame that is being searched, R obtained by the MR calculation unit, and M
From the M stored in the storage memory, the evaluation value B obtained by the motion vector code amount evaluation unit, and the evaluation value C obtained by the reference image designation code amount evaluation unit, the SEA method is calculated using the above inequality (11). A SEA execution unit 105 to be executed, a decoded image memory 109 for storing decoded images of past frames,
An interframe predictive coding unit 103 that creates a predicted image using a decoded image and codes an input image, and a decoding unit 108 that decodes coded data are provided.

In the above structure, the current image at time t is stored in the reference image memory 104 at times t−1, t.
Input images I1, I2, I3 at −2, t−3 are stored, and M information M1, M2, M3 at each time is stored in the M storage memory 102.
It is assumed that the decoded images D1, D2, D3 at each time are stored in 09. The search range in each frame is (-16, -16) to (+16, +16).

The motion vector code amount evaluation unit 106 uses B in the following equation (14) for the motion vector information (x, y).
It is assumed that (x + a, y + b) is obtained. However, the motion vector information (x, y) here is assumed to be obtained by the above equation (5) from the motion vector (x ₀ , y ₀ ) obtained by the motion search. It is assumed that the reference image designation code amount evaluation unit 107 obtains C (t, t-ref) by the following formula (15) for the information ref that designates the reference image.

[0050]

[Equation 9]

The operation when the current image at time t is input on the basis of such a premise is as follows. First, the MR calculation unit 101 uses R (a, b) and M (x + a, y) of the input image.
+ B, t). It should be noted that the position (a, b) indicates the upper left pixel position of the macroblock.

The SEA execution unit 105 obtains information designating a motion vector of each macroblock and a reference image according to the operation flow of FIG. The SEA execution unit 105 first obtains motion vectors of all macroblocks for the first reference image. Set the reference image to I1 (Step1
1).

The motion vector of the macroblock at the position (a, b) is obtained as follows. The initial value of the search position is set to (m, n) = (-16, -16), and MA
D (m + a, n + b, t-1) and B (m + a, n + b)
And C (t, t-1) are obtained (Step 12). Of these, MAD (m + a, n + b, t-1) is calculated from the I1 accumulated in the reference image memory and the input image by the MAD calculation unit 110.
The value of B (m + a, n + b) is obtained by the motion vector code amount evaluation unit 106, and the value of C (t, t−1) is obtained by the reference image designation code amount evaluation unit 107.

Next, the search position is set to the adjacent pixel (x, y) =
Set to (-15, -16). R (a, b) obtained by the MR calculation unit 101, M (x + a, y + b, t−1) stored in the M storage memory 102, and B (x + a, y + b) obtained by the motion vector code amount evaluation unit 106. ) And C (t, t−) obtained by the reference image designation code amount evaluation unit 107.
1) is used to check whether the inequality (11) is satisfied. By repeating this, the search position (x, y) that satisfies the inequality (11) is obtained (Step 13).

Search position (x ₀ , which satisfies inequality (11),
If y ₀ ) is obtained, the MAD calculator 110 calculates MA
D (x ₀ + a, y ₀ + b, t-1), MAD (x
_{0 + a, y 0 + b} , t-1) + B (x 0 + a, y 0 + b) +
C (t, t-1) and MAD (m + a, n + b, t-
1) + B (m + a, n + b) + C (t, t-1) are compared (Step 14).

[0056] When the former is smaller values, the MAD (m + a, n + b, t-1) and _{MAD (x 0 + a, y} 0 + b, t-1) in the inequality (11), B (m + a,
(n + b) is replaced with B (x ₀ + a, y ₀ + b) (St
ep15). Then, returning to Step 13, the above process is repeated to obtain the motion vector for the macroblock at the position (a, b). At this point, the information ref that specifies the reference image is 1.

Returning to Step 12, the above processing is repeated to obtain motion vectors of all macroblocks.

Next, returning to Step 11, the reference image is I
Set to 2. Information specifying the motion vector of the macroblock at the position (a, b) and the reference image is obtained as follows.

Step 12 is skipped. Step1
3, in the search position (x, y) = (− 16, −16), R (a, b) obtained by the MR calculation unit 101 and M (x + a, y + b, t stored in the M storage memory 102). −
2) and B obtained by the motion vector code amount evaluation unit 106
(X + a, y + b) and reference image designation code amount evaluation unit 10
Using C (t, t-2) obtained in step 7, the inequality (11)
Inspect to see if they meet. While repeating this,
A search position (x, y) that satisfies the inequality (11) is obtained (Step 13).

Search position (x ₀ , which satisfies inequality (11)
If y ₀ ) is obtained, the MAD calculator 110 calculates MA
D (x ₀ + a, y ₀ + b, t-2) is calculated. MAD (x
_{0 + a, y 0 + b} , t-2) + B (x 0 + a, y 0 + b) +
C (t, t-2) and MAD (m + a, n + b, t-
1) + B (m + a, n + b) + C (t, t-1) are compared (Step 4).

[0061] When the former is smaller values, the MAD (m + a, n + b, t-1) and _{MAD (x 0 + a, y} 0 + b, t-2) in the inequality (11), B (m + a,
n + b) to B (x ₀ + a, y ₀ + b) and C (t, t-
Replace 1) with C (t, t-2) (Step 1)
5). Returning to Step 13, the above processing is repeated to obtain information designating the motion vector and the reference image for the macroblock at the position (a, b).

Returning to Step 12, the above processing is repeated to obtain motion vectors of all macroblocks.

Returning to Step 11, the reference image is set to I3, and the same processing as when the reference image is t-2 is repeated.

When the motion search is completed for all the reference images, it is possible to obtain the information specifying the motion vector and the reference image in each macroblock, which satisfies the inequality (11) at all the reference images and the search positions.

Next, using the motion vector obtained here and the information for designating the reference image, the interframe predictive coding unit 1
03 creates a predicted image from the decoded image stored in the decoded image memory and encodes the input image. Decoding unit 108
Decodes the encoded data. When all the macroblocks have been encoded, the decoded image is stored in the decoded image memory 109.
The M calculated by the MR calculation unit 101 is stored in the M storage memory 102, and the input image is stored in the reference image memory.

As described above, according to this embodiment, when the reference image memory for a plurality of frames is provided and the motion compensation is performed while selecting the reference image for each macroblock, the code amount of the motion vector and the reference image are used. It is possible to obtain the information designating the motion vector and the reference image that minimizes the sum of the code amount of the information designating and the sum of the absolute differences of the prediction errors.

In this embodiment, the reference image is provided for a plurality of frames, but it may be configured to have one frame. In the SEA method in this case, the inequality (1
The inequality (10) is used instead of 1) to evaluate the code amount of the motion vector and the absolute value difference sum. In this case, the image coding apparatus does not include the reference image designated code amount evaluation unit.

Further, in the present embodiment, R and M are calculated at the same time from the input image by the MR calculation unit, but only R can be obtained from the input image and M can be obtained from the reference image stored in the reference image memory. Is. Further, in the present embodiment, the motion vector is obtained for each macroblock, but the motion vector may be obtained for each region having an arbitrary number of vertical and horizontal pixels.

In the present invention, a part or all of the processing functions of the apparatus shown in FIG. 1 are configured as a program and implemented by using a computer, or the processing procedure shown in FIGS. Can be configured to run on a computer. A computer-readable recording medium that stores a program for realizing the processing functions of the respective parts of the computer, or a program for causing the computer to execute the processing procedure,
For example, FD (floppy disk: registered trademark), M
O, ROM, memory cards, CDs, DVDs, removable disks, etc. can be recorded, saved, and provided, and can be distributed via a communication network such as the Internet. It is possible.

[0070]

As described above, according to the present invention, SEA
In the method, it is possible to obtain the motion vector having the minimum sum of absolute differences of the prediction errors and the motion vector code amount evaluation value, that is, the sum of the prediction error code amount and the motion vector code amount.

In addition, in the encoding method for selecting a reference image from a plurality of frames, the sum of the absolute value difference sum of prediction errors, the motion vector code amount evaluation value, and the code amount evaluation value of the information designated as the reference image, that is, A combination of the motion vector and the reference image that minimizes the sum of the code amount of the prediction error, the code amount of the motion vector, and the code amount of the information specifying the reference image can be obtained.

[Brief description of drawings]

FIG. 1 is a relationship diagram between a macroblock and a search range according to an embodiment of the present invention.

FIG. 2 is a motion vector determination flow according to the SEA method.

FIG. 3 is a relationship diagram between a current macroblock and surrounding macroblocks in motion vector coding.

FIG. 4 is a motion vector determination flow by a SEA method in a plurality of frame reference images.

FIG. 5 is a configuration diagram of an embodiment.

[Explanation of symbols]

101 ... MR calculation unit 102 ... M storage memory 103 ... Interframe predictive coding unit 104 ... Reference image memory 105 ... SEA execution unit 106 ... Motion vector code amount evaluation unit 107 ... Reference image designation code amount evaluation unit 108 ... Decoding unit 109 ... Decoded image memory 110 ... MAD calculator

Continued front page    F-term (reference) 5C059 MA00 MA05 MA12 MA23 ME01                       NN03 NN08 NN28 PP04 SS20                       TA63 TB07 TC02 TC03 TC06                       TC12 TC18 TC31 TC42 TD02                       TD05 TD06 TD11 TD16 UA02                       UA33 UA39                 5J064 AA03 AA04 BA09 BA16 BB03                       BC01 BC08 BC16 BD01

Claims (6)

[Claims] 1. A moving image coding method for performing a motion search for each current region obtained by dividing an input image, wherein a sum W of absolute difference between a current region pixel and a reference image pixel at a search position A in a reference image. An absolute value difference sum calculating means for calculating the sum of the pixel values in the current area, a pixel calculating means for calculating the sum R of pixel values in the current area, and a pixel value of the same number of pixels as the current area at the search position B different from the search position A. A pixel in the reference image for calculating a total M, a motion vector code amount evaluating unit for calculating an evaluation value VA or VB of the motion vector code amount when the search position A or B is a motion vector, and the pixel in the current area A search position by adding the evaluation value VB calculated by the motion vector code amount evaluation means to the absolute value of the difference between the search position R calculated by the calculation means and the search position M calculated by the reference image pixel calculation means. A search position evaluation means for determining the value D, the absolute value difference sum W obtained by the absolute value difference sum calculating means
And an absolute value difference sum evaluation means for adding the evaluation value VA obtained by the motion vector code amount evaluation means to obtain an absolute value difference sum evaluation value E, which is a moving image coding method for performing a motion search. , The search position evaluation value D obtained by the search position evaluation means and the absolute value difference sum evaluation value E obtained by the absolute value difference sum evaluation means are compared in size, and when the value of D is equal to or less than the value of E, A moving image coding method, wherein the absolute value difference sum calculating means obtains an absolute value difference sum W between the current region pixel and the reference image pixel at the search position B. 2. A moving image coding method for performing a motion search by selecting a reference image from a reference image memory capable of accumulating a plurality of frames (P1 to Pn) of an image for each current region obtained by dividing an input image. Absolute value difference sum calculating means for calculating the absolute value difference sum W between the current area pixel and the reference image pixel at the search position A in the image Pi, and current area pixel calculating means for calculating the sum R of the current area pixel values And a pixel-in-reference-image calculating means for obtaining a sum M of pixel values of the same number of pixels as the current region at a search position B in a reference image Pj different from the reference image Pi, and the search position A or B as a motion vector. And a motion vector code amount evaluation means for obtaining an evaluation value VA or VB of the motion vector code amount, and a code amount evaluation value Ri or Rj for encoding the information designating the reference image Pi or Pj. The absolute value of the difference between the reference image specifying code amount evaluating means to be obtained, the total R obtained by the pixel in the current region and the total M obtained by the pixel in the reference image calculating means is set to the motion vector code amount evaluating means. And an absolute value obtained by the absolute value difference sum calculating means, and a search position evaluating means for obtaining a search position evaluation value D by adding the evaluation value VB obtained by the above and the evaluation value Rj obtained by the reference image designation code amount evaluating means. Difference W
An absolute value difference sum evaluation means for adding an evaluation value VA calculated by the motion vector code quantity evaluation means and an evaluation value Ri calculated by the reference image designation code quantity evaluation means to obtain an absolute value difference sum evaluation value E; Is a moving image coding method for performing a motion search by using the search position evaluation value D obtained by the search position evaluation means and the absolute value difference sum evaluation value E obtained by the absolute value difference sum evaluation means. In comparison, if the value of D is less than or equal to the value of E,
A moving picture coding method, wherein the absolute value difference sum calculation means obtains an absolute value difference sum W between a current region pixel and a reference image pixel at a search position B in the reference image Pj. 3. A program of a moving picture coding method for performing a motion search for each current area obtained by dividing an input image, wherein an absolute value difference between a current area pixel and a reference image pixel at a search position A in the reference image. The absolute value difference sum calculation procedure for obtaining the sum W, the current area pixel calculation procedure for obtaining the sum R of the pixel values in the current area, and the pixels of the same number of pixels as the current area at the search position B different from the search position A Reference image pixel calculation procedure for obtaining the total M of values, and the evaluation value V of the motion vector code amount when the search position A or B is a motion vector
A motion vector code amount evaluation procedure for obtaining A or VB;
The evaluation value VB calculated in the motion vector code amount evaluation procedure is added to the absolute value of the difference between the search position R calculated in the current area pixel calculation procedure and the search position M calculated in the reference image pixel calculation procedure. And a search position evaluation procedure for obtaining a search position evaluation value D, and an absolute value obtained by adding the absolute value difference sum W obtained by the absolute value difference sum calculation procedure and the evaluation value VA obtained by the motion vector code amount evaluation procedure. The absolute value difference sum evaluation procedure for obtaining the difference sum evaluation value E is compared with the magnitude of the search position evaluation value D obtained in the search position evaluation procedure and the absolute value difference sum evaluation value E obtained in the absolute value difference sum evaluation procedure. Then, when the value of D is equal to or less than the value of E, the absolute value difference sum calculation procedure determines the absolute value difference sum W between the current region pixel and the reference image pixel at the search position B. Characterized by Moving picture coding method program that. 4. A program of a moving picture coding method for selecting a reference picture from a reference picture memory capable of accumulating a plurality of frames (P1 to Pn) for each current area obtained by dividing an input picture and performing a motion search. , An absolute value difference sum calculation procedure for obtaining an absolute value difference sum W between the current area pixel and the reference image pixel at the search position A in the reference image Pi, and a current area pixel for obtaining a total R of pixel values in the current area The calculation procedure and the reference image pixel calculation procedure for obtaining the total M of the pixel values of the same number of pixels at the search position B in the reference image Pj different from the reference image Pi and the search position A or B A motion vector code amount evaluation procedure for obtaining an evaluation value VA or VB of a motion vector code amount when a motion vector is used, and an evaluation value Ri of a code amount for encoding information designating the reference image Pi or Pj. Other R
In the reference image designation code amount evaluation procedure for obtaining j, the absolute value of the difference between the total R obtained in the current area pixel calculation procedure and the total M obtained in the reference image pixel calculation procedure is set to the motion vector code quantity. Evaluation value VB obtained in the evaluation procedure and evaluation value R obtained in the reference image designation code amount evaluation procedure
The search position evaluation procedure of adding j to obtain the search position evaluation value D, and the evaluation value VA obtained in the motion vector code amount evaluation procedure and the reference to the absolute value difference sum W obtained in the absolute value difference sum calculation procedure. An absolute value difference sum evaluation procedure for adding the evaluation values Ri calculated in the image designation code amount evaluation procedure to obtain an absolute value difference sum evaluation value E, and a search position evaluation value D and the absolute value calculated in the search position evaluation procedure. The magnitude of the absolute value difference sum evaluation value E obtained by the difference sum evaluation procedure is compared, and D
When the value of is less than or equal to the value of E, the absolute value difference sum calculation procedure obtains the absolute value difference sum W between the current region pixel and the reference image pixel at the search position B in the reference image Pj,
A program for a moving picture coding method, comprising: 5. A recording medium for recording a program of a moving picture coding method for performing a motion search for each current area obtained by dividing an input image, wherein a current area pixel and a reference image pixel at a reference image search position A are recorded. Absolute value difference sum calculation procedure for obtaining the absolute value difference sum W between them, a current area pixel calculation procedure for obtaining the total R of pixel values in the current area, and a current area at a search position B different from the search position A. A reference image pixel calculation procedure for obtaining the sum M of pixel values of the same number of pixels, and an evaluation value V of the motion vector code amount when the search position A or B is a motion vector.
A motion vector code amount evaluation procedure for obtaining A or VB;
The evaluation value VB calculated in the motion vector code amount evaluation procedure is added to the absolute value of the difference between the search position R calculated in the current area pixel calculation procedure and the search position M calculated in the reference image pixel calculation procedure. And a search position evaluation procedure for obtaining a search position evaluation value D, and an absolute value obtained by adding the absolute value difference sum W obtained by the absolute value difference sum calculation procedure and the evaluation value VA obtained by the motion vector code amount evaluation procedure. The absolute value difference sum evaluation procedure for obtaining the difference sum evaluation value E is compared with the magnitude of the search position evaluation value D obtained in the search position evaluation procedure and the absolute value difference sum evaluation value E obtained in the absolute value difference sum evaluation procedure. Then, when the value of D is equal to or less than the value of E, the absolute value difference sum calculation procedure obtains the absolute value difference sum W between the current region pixel and the reference image pixel at the search position B. Characterized by A recording medium recording a program of a moving image encoding method for. 6. A program of a moving picture coding method for performing a motion search by selecting a reference picture from a reference picture memory capable of accumulating a plurality of frames (P1 to Pn) of an picture for each current area obtained by dividing an input picture. And a total R of pixel values in the current area, and an absolute value difference sum calculation procedure for obtaining a sum W of absolute value differences between the current area pixels and the reference image pixels at the search position A in the reference image Pi. Pixel calculation procedure in the current area, reference pixel calculation procedure in the reference image Pj different from the reference image Pi, and a total M of pixel values of the same number of pixels as the current area at the search position B; A motion vector code amount evaluation procedure for obtaining an evaluation value VA or VB of the motion vector code amount when the position A or B is used as a motion vector, and information for specifying the reference image Pi or Pj are encoded. Issue amount of the evaluation value Ri or R
In the reference image designation code amount evaluation procedure for obtaining j, the absolute value of the difference between the total R obtained in the current area pixel calculation procedure and the total M obtained in the reference image pixel calculation procedure is set to the motion vector code quantity. Evaluation value VB obtained in the evaluation procedure and evaluation value R obtained in the reference image designation code amount evaluation procedure
The search position evaluation procedure of adding j to obtain the search position evaluation value D, and the evaluation value VA obtained in the motion vector code amount evaluation procedure and the reference to the absolute value difference sum W obtained in the absolute value difference sum calculation procedure. An absolute value difference sum evaluation procedure for adding the evaluation values Ri calculated in the image designation code amount evaluation procedure to obtain an absolute value difference sum evaluation value E, and a search position evaluation value D and the absolute value calculated in the search position evaluation procedure. The magnitude of the absolute value difference sum evaluation value E obtained by the difference sum evaluation procedure is compared, and D
When the value of is less than or equal to the value of E, the absolute value difference sum calculation procedure obtains the absolute value difference sum W between the current region pixel and the reference image pixel at the search position B in the reference image Pj,
A recording medium on which a program of a moving picture coding method is recorded.