JPH0678281A - Moving picture coder - Google Patents

Abstract

PURPOSE:To provide a moving picture coder with an excellent coding efficiency by controlling a code quantity of a picture and a quantization error of the picture simultaneously. CONSTITUTION:A code quantity measurement means 20 measures a code quantity of a picture and an upper limit and a lower limit of a buffer occupied quantity in a buffer 14. A quantization error measurement means 21 measures a quantity error in the picture based on the input and output of a quantization means 12. A quantization control means 22 controls a quantization step width (h) given to the quantization means 12 so that an actually measured quantization error measured by the quantization error measurement means 21 is a prescribed quantization error, and revises a prescribed quantization error so that the actually measured code quantity from the code quantity measurement means 20 is a prescribed code quantity.

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 apparatus for use in a multimedia computer or the like for coding a moving picture, and more particularly to a moving picture coding apparatus for improving the coding quality of a moving picture and controlling the code amount. It is related to the method.

[0002]

2. Description of the Related Art Conventionally, as a moving image coding method, particularly a moving image coding method for storage, for example, the method described in the following documents is often used. Reference: Hiroshi Yasuda "International Standard for Multimedia Coding"
(Head 3-6) Maruzen Co., Ltd. 127-156 FIG. 2 is a configuration diagram of an image group in the conventional moving image encoding system for storage described in the above document. This image group has N images as one group, and the group includes an intra-picture coded image I called an intra frame, a forward prediction coded image P called an inter frame, and bidirectional prediction called an interpolated frame. And the encoded image B. The inter-frame P is for predicting the image using an intra-frame I or an inter-frame P in front of the image and encoding a prediction error signal. Interpolated frame B
Is a method for predicting the image bidirectionally by using an intra-frame I or inter-frame P in the front of the image and an intra-frame I or an inter-frame P in the rear, and encoding a prediction error signal.

FIG. 3 is a block diagram showing an example of the configuration of a conventional moving image coding apparatus for storage described in the above-mentioned documents. This moving picture coding apparatus has input terminals 1, 2, 3 for inputting a current picture and a reference picture for predicting the current picture.
And an output terminal 4 for outputting encoded image data, and the prediction means 10 with motion compensation is connected to the input terminals 1, 2, and 3.

The motion compensation predicting means 10 has the following functions. That is, the input image of the intra frame I is output as it is. For the input image of the inter frame P, the intra frame I or the inter frame P temporally preceding is used to predict the inter frame image in M × M block units, and the difference between the current image and the input image (prediction error) ) Is output. Further, with respect to the input image of the interpolated frame B, the intra-frame I or the inter-frame P temporally forward and backward is used to predict the image bidirectionally in M × M block units, and a prediction error signal is obtained. Output. On the output side of the prediction unit with motion compensation 10, an orthogonal transformation unit 11, a quantization unit 12, an encoding unit 13, a buffer 14, and an output terminal 4 are provided.
Are cascaded. The orthogonal transformation unit 11 divides the difference pixels input from the prediction unit with motion compensation 10 into N × N blocks, and discrete cosine transform (hereinafter referred to as DCT).
It has a function of converting the image for each block using an orthogonal conversion method such as the above and outputting the conversion result to the quantizing means 12. The quantizing means 12 quantizes the output of the image block orthogonally transformed by the orthogonal transforming means 11 based on the given quantization step width h, and outputs the quantized image data to the encoding means 13. Is.

The coding means 13 performs variable length coding on the image data quantized by the quantizing means 12, for example, by the Huffman coding method in which the number of zero data and the level of non-zero data are combined, and codes the code. It has a function of outputting the converted image data to the buffer 14. The buffer 14 is, for example, a first-in / first-out (hereinafter, FIFO).
It has a function of temporarily storing encoded image data and outputting it to the output terminal 4 at a constant bit rate in accordance with the FIFO law. The buffer 14 is connected to the quantizing control means 16 via the code amount measuring means 15, and the output side of the quantizing control means 16 is connected to the quantizing means 12. Code amount measuring means 15
Measures the code amount b _r of the coded image data in units of a predetermined number of blocks from the occupied amount of the buffer 14,
This is given to the quantization control means 16. The quantization control unit 16 compares the measured code amount b _r input from the code amount measuring unit 15 with the assigned code amount b _a that is assigned in advance according to the image, and the measured code amount b _r is the assigned code amount. Quantity b _a
To have a function of controlling the quantization step width h given to the quantizing means 12.

Next, the operation of FIG. 3 will be described. Typically,
The images to be encoded are input terminals 1, 2, in the order shown in FIG.
3 and is encoded. When the current image is the intraframe I, the image is input from the input terminal 1. When the current image is the inter frame P, the image is input from the input terminal 1, and the intra frame I or the inter frame P immediately preceding in time for predicting the image is input from the input terminal 2. When the current image is the interpolated frame B, the current image is input from the input terminal 1, the intra frame I or the inter frame P immediately before the current image is input from the input terminal 2, and the intra frame immediately after the current image is input. The frame I or the interframe P is input from the input terminal 3.

The motion compensation predicting means 10 outputs the intra frame I input from the input terminal 1 as it is. Further, the inter-frame P input from the input terminal 1 is divided into M × M blocks of the current image, and the forward prediction image input from the input terminal 2 is used for each block, for example, the pattern matching method. The motion vector is detected by, and a block of the same size that has been moved by the motion vector from the position of the current image block of the previous image is used as the prediction image block of the current image block, and the current image block and the prediction image of the current image block Output the difference from the block. In the motion compensation predicting means 10, the interpolated frame B input from the input terminal 1 is divided into M × M blocks in the same manner as described above, and the forward prediction image and the forward prediction image input from the input terminal 2 are input. Predicted image blocks are selected from the backward predicted images input from the input terminal 3 in the same manner as described above, and the difference block between the current image block and the forward predicted image block, the current image block and the backward predicted image block are selected. Difference block of
The difference block between the current image block and the average of both the forward and backward predicted image blocks is obtained, and the smallest difference block from the sum of squares of these three types of difference blocks is output as the prediction error signal of the current image block. Then
Send to the orthogonal transformation means 11. The orthogonal transform means 11 calculates the difference image input from the motion-compensated prediction means 10 by N × N.
Block (for example, N = 8), and each block is subjected to orthogonal transformation using an orthogonal transformation method such as DCT of the following equation (1), and the transformed image block is quantized by the quantizing means 12 Send to.

[0008]

[Equation 1] C (u) and C (v) in the equation (1) are represented by the following equation (2).

[0009]

[Equation 2] The quantizing unit 12 quantizes the image block input from the orthogonal transforming unit 11 based on the quantizing step width h given by the quantizing controlling unit 16, for example, according to the following equation (3).

[0010]

[Equation 3] The quantized image data is sent to the encoding means 13 according to the zigzag order shown in Table 1 below.

[0011]

[Table 1] In the encoding means 13, the quantized data input from the quantization means 12 in block units is converted into a number of consecutive zeros (hereinafter referred to as zero run length) and a non-zero value (hereinafter referred to as level) that follows. The coded image data is combined with the code shown in the zero-run level composite code examples shown in FIGS. 4 and 5, and the encoded image data is buffered.
Send to. In FIG. 4, the last bit "s" indicates the level (LEVEL) positive or negative. That is, "0" is positive,
“1” indicates negative. For combinations of run and LEVEL other than those shown in FIG. 3, 6-bit ESC
APE code, 6-bit RUN and 8-bit LEVE
It is encoded by 20 bits of L. Buffer 1
Reference numeral 4 temporarily stores coded image data input from the coding means 13 at an irregular bit rate, and a FIFO
According to the above rule, data is output to the output terminal 4 at a constant bit rate. The code amount measuring means 15 measures the code amount b _r of the buffer 14 from the buffer 14 on a block-by-block basis or on a block-by-block basis and outputs it to the quantization controller 16.
In the quantization control means 16, the measured code amount b _r input from the code amount measuring means 15 in units of one block or several blocks and the allocated code amount b _a of the same number of blocks given according to the image prediction method. From the above, the quantization step width h given to the quantization means 12 is obtained from the following equations (4) and (5). B (n) = B (n -1) + (b r -b a) ··· (4) h (n) = f (n-1) + B (n) / k ··· (5) where, k; constant Here, the constant k of the equation (5) is determined by the capacity of the buffer 14 and the range of the quantization step width h. For example, if the capacity of the buffer 14 is 120 Kbits and the quantization step width h is 2 to 62, k = 2000.
Neighborhood is appropriate.

[0012]

However, since the storage moving picture coding apparatus having the above configuration controls only the code amount of the buffer 14, the coding quality of the next image is the same as that of the previous image. It depends on the amount. Therefore, when the code amount of the previous image becomes larger than the planned amount, the code amount assigned to the next image becomes small and the quality of the image deteriorates. On the contrary, when the code amount of the previous image is small, the code amount of the next image is increased and the quality is improved. When the image quality fluctuates in this way, the coding efficiency decreases. Moreover, with conventional devices,
It was difficult to optimize the encoding according to the image properties. The present invention solves the problem of low coding efficiency as a problem that the above-mentioned conventional art has, and a moving image with excellent coding efficiency by controlling the quantization error of the image at the same time as the code amount of the image. An encoding device is provided.

[0013]

[Means for Solving the Problems] In order to solve the above-mentioned problems, an input image of the intra frame is selected from among three types of coded images of an intra frame, an inter frame and an interpolated frame. On the other hand, the image is output as it is, and the image is M × M (however, M;
Make a prediction image by predicting in block units of (positive integer),
A motion-compensated prediction unit that outputs a difference image between the current image and the predicted image, and an output image of the motion-compensated prediction unit is divided into N × N (where N; positive integer) blocks, and the block Orthogonal transformation means for orthogonal transformation for each, quantization means for quantizing the output of the orthogonal transformation means using a given quantization step width, coding means for coding the output of the quantization means, A moving picture coding apparatus provided with a buffer for temporarily storing the output of the coding means and outputting it at a constant bit rate is provided with the following means.

That is, according to the present invention, the code amount measuring means for measuring the code amount of the image and the upper and lower limits of the buffer occupancy amount based on the contents of the buffer, and the input and output of the quantizing means, Quantization error measuring means for measuring the number of blocks in the image, the quantization error of the block or the whole image, and the quantization error of the measured image based on the outputs of the quantization error measuring means and the code amount measuring means. To a predetermined quantization error, and to give the predetermined quantization error and the quantization means such that the group average code amount of the image obtained from the measured code amount becomes a predetermined code amount. Quantization control means for controlling the quantization step width. In a second aspect of the invention, the quantization control means of the first aspect of the invention sets the quantization error for the input intraframe and interframe to be equal to or greater than the quantization error of the previous intraframe or interframe. And for the input interpolated frame, the setting quantization error thereof is larger than the setting quantization error of the temporally closer intra-frame or inter-frame before and after used for prediction. The configuration is such that the quantization step width is controlled. In a third aspect of the invention, the quantization control means of the first aspect of the invention is used, in which the quantization error of the input interframe is
Α times the quantization error of the intra frame or inter frame used for forward prediction (where 1 ≦ α ≦ 2)
Therefore, the quantization error of the input interpolated frame is such that the forward or backward nearer intra frame or inter frame used for prediction is β times the quantization error (where 1 ≦ β ≦ 2). So that
The quantization step width is controlled.

According to a fourth aspect of the invention, a motion compensation predicting means similar to the first aspect of the invention, an orthogonal transforming means, a quantizing means,
The following means is provided in the moving picture coding apparatus including the coding means and the buffer. That is, in the present invention, the code amount measuring means for measuring the code amount of the image and the upper and lower limits of the buffer occupancy amount based on the contents of the buffer, the current image, and the prediction means with motion compensation are output. From the predicted image of the current image to be calculated, statistical property calculation means for calculating statistical properties of the image, and based on the calculated statistical properties of the image, the average code amount in the image group is a predetermined code amount. And a code amount determining means for calculating an optimum code amount to be assigned to each image in the image group so that the average quantization error of the image is minimized, and the code amount measuring means and the code amount determining means. Quantization control means for controlling the quantization step width given to the quantization means based on the output.

In a fifth aspect of the invention, the code amount determining means of the fourth aspect of the invention is used so that the quantization error of the interframe is equal to or greater than the quantization error of the forward intraframe or interframe used for prediction. The quantization error of the interpolated frame is set to be larger than the quantization error of the forward or backward intra frame or inter frame used for prediction, and the optimum allocation code amount is obtained. In the sixth invention, the code amount determination means of the fifth invention presumes several kinds of statistical properties of the image in advance, calculates the optimum allocated code amount for them, and calculates the actual image from the statistical property calculation device. When the information of the statistical property is input, the one that most corresponds to the information of the statistical property is selected from the optimal allocation code amount calculated by the assumed image property.
In a seventh invention, the code amount determining means of the sixth invention sets the quantization error of the input interframe to be α times the quantization error of the intraframe or interframe used for forward prediction. Then, the quantization error of the input interpolated frame is set to be β times the quantization error of the forward or backward nearer intra frame or inter frame used for prediction, and the optimal code amount to be assigned to each image is set. It is configured to calculate.

[0017]

According to the first aspect of the present invention, since the moving picture coding apparatus is configured as described above, the code amount measuring means measures the code amount of the image and the upper and lower limits of the buffer occupation amount from the buffer. The measurement result is sent to the quantization control means. The quantization error measuring means measures the quantization error of the image from the input and the output of the quantizing means and sends the measurement result to the quantization control means. The quantization control means updates the predetermined quantization error so that the actually measured code quantity from the code quantity measuring means becomes a predetermined code quantity. Then, the quantization control means changes the quantization step width so that the measured quantization error measured by the code amount measuring means becomes the predetermined quantization error, and controls the quantization of the quantization means.

According to the second invention, the quantization control means is
The quantization step width given to the quantization means is controlled according to the input image group. According to the third invention, the quantization control means controls the quantization step width given to the quantization means according to the quantization error of the input image group. According to the fourth aspect, the code amount measuring means measures the code amount of the image and the upper and lower limits of the buffer occupation amount from the buffer, and sends the measurement information to the quantization control means. The statistical property calculation means calculates the statistical property of the image from the input of the image and the output of the prediction means with motion compensation, and sends the information to the code amount determination means. The code amount determination means determines the optimum code amount to be assigned to each image based on the calculated statistical properties of the image and sends it to the quantization control means. The quantization control means changes the quantization step width based on the outputs of the code amount measuring means and the code amount determining means, and controls the quantization of the quantizing means. According to the fifth invention, the code amount determining means obtains the optimum assigned code amount according to the quantization error of the input image group. According to the sixth aspect, the code amount determining means calculates the optimal allocation code amount based on the statistical property of the image, and when the information of the statistical property of the actual image is input from the statistical property calculating device, the calculation is performed. The optimum allocation code amount is calculated from the optimum allocation code amount, and the code amount is determined at high speed. According to the seventh invention,
The code amount determining means sets the quantization error according to the input image group and calculates the optimum code amount to be assigned to each image. Therefore, the above problem can be solved.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First Embodiment FIG. 1 is a block diagram showing the configuration of a moving picture coding apparatus for storage according to the first embodiment of the present invention, and FIG. 3 showing a conventional moving picture coding apparatus for storage. Elements that are the same as the elements inside have the same reference numerals. This moving picture coding apparatus includes a prediction unit with motion compensation 10 and an orthogonal transformation unit 1 which are the same as those of the conventional FIG.
1, quantizer 12, encoder 13, and buffer 1
It is equipped with 4. Code amount measuring means 20 is provided in the buffer 14.
Is connected to the input side and the output side of the quantizing means 12, and the quantizing error measuring means 21 is connected. Further, the quantizing control is performed on the output sides of the code amount measuring means 20 and the quantizing error measuring means 21. The means 22 is connected, and the quantization control means 22 is connected to the quantization means 12.

The code amount measuring means 20 measures the code amount b _{r of the} image based on the contents of the buffer 14 and also measures the upper and lower limits of the occupied amount of the buffer 14 and quantize control information thereof. 22 has a function to give.
The quantization error measuring means 21 is based on the input X (i) given to the quantization means 12 from the orthogonal transformation means 11 and the output Y (i) from the quantization means 12 to the encoding means 13,
Quantization error MSE is measured and measured quantization error MSE
To the quantization control means 22. The quantization control unit 22 controls the quantization step width h given to the quantization unit 12 so that the measured quantization error MSE becomes a predetermined quantization error, and the measured code amount b _r is predetermined. It has a function of updating the predetermined quantization error so that the assigned code amount is obtained.

Next, the operation of the moving picture coding apparatus shown in FIG. 1 will be described. When the image to be encoded is input to the input terminals 1, 2 and 3 in the order shown in FIG. 2, the motion compensating prediction means 10 obtains a differential image and the like, as in the conventional case. The orthogonal transform is performed by the orthogonal transform means 11, is quantized by the quantizing means 12 based on the quantizing step width h, is coded by the coding means 13, is sent to the buffer 14, and is coded from the buffer 14. The image data is output to the output terminal 4 at a constant bit rate. The code amount measuring means 20 measures the code amount b _r of one image unit based on the contents of the buffer 14 and sends it to the quantization control means 22, and at the same time, the occupied amount of the buffer 14 exceeds the upper limit or the lower limit. It is checked whether or not the upper limit or the lower limit is exceeded, and the information is sent to the quantization control means 22. In the quantizing error measuring means 21, the input X of the quantizing means 12 is inputted.
Based on (i) and the output Y (i), the quantization error MSE is calculated from the following equation (6), for example, and sent to the quantization control means 22.

[0022]

[Equation 4] In the equation (6), N can be one block, several blocks, or one image. In the quantization control means 22, the quantization error M for the intra frame I and the inter frame P in one image group
SE is the order of input, and the quantization error MS of the previous intra frame I or inter frame P is expressed by the following equation (7).
An interpolated frame (i-th intra frame in terms of time) such that α times (α ≧ 1) of E and between the (i−1) th and i-th intra frames I or P. I or the interpolated frame B immediately following the interframe P), the quantization error MSE of the interpolated frame B close to the (i-1) th intra frame I or interframe P is MSEP (i-1). Β times (β>
1), the quantization error is set so that the quantization error MSE of the interpolated frame B close to the i-th intra frame I or the inter frame P is β times MSEP (i), and given to the quantization means 12. The quantization step width h is controlled. MSEP (i) = MSEP (i−1) · α (7) where α; coefficient The quantization step width h is controlled by the following method, for example. The quantization error MSE input from the quantization error measuring means 21 and a predetermined quantization error MSE (i)
From, the quantization step width h given to the quantization means 12
(N) is calculated by the following equation (8).

[0023]

[Equation 5] However, when the information that the occupied amount of the buffer 14 exceeds the upper limit or the lower limit is input from the code amount measuring means 20, the quantization step width is unconditionally set to the maximum value h _max or the minimum value h _min . Here, the number of intra frames I in one group is N _i , and the number of inter frames P is N.
_P and the number of interpolated frames B are N _b = 2
(N _i + N _P ), the average quantization error in the group
MSE is obtained as in the following equation (9). Therefore, the quantization error of each image MSEP (1), MSEP (i),
MSEB (i) becomes the following equations (10) to (12).

[0024]

[Equation 6]

[Equation 7] The quantization control means 22 further includes the code amount measuring means 20.
The group average code amount b _ave is calculated from the code amount b _{r of} each image given by the above, and the average quantization error MSE is set so that the average code amount b _ave becomes a predetermined code amount BR (= transmission bit rate). To the following equation (13), and
Using the updated average quantization error MSE , the set quantization errors MSEP (i) and MSEB (i) of each image are calculated as in the equations (11) and (12).

[0025]

[Equation 8] In equations (8) and (13), constants k and r
Influences the property of control, and if the constants K and r are reduced, the control becomes slower but the sensitivity also lowers. On the other hand, when the constants K and r are increased, the control becomes quick and sensitive, but the vertical fluctuation may become severe. It is possible to change the constants K and r as needed.
Further, the coefficients α and β in the equations (9) to (12) are used.
Influences the efficiency of encoding. That is, the difference between the current image and the predicted image includes the fluctuation amount of the image and the quantization error MSE of the predicted image. Therefore, the accuracy of prediction and the magnitude of the prediction error depend on the quantization error MSE of the image used for prediction. Quantization error MSE of prediction image
If is smaller, the accuracy of prediction increases, the prediction error also decreases, and the coding efficiency improves.

On the other hand, since the prediction error signal of the current image also includes the quantization error MSE of the prediction image, there are many ways to reduce the quantization error MSE of the current image to the quantization error MSE of the prediction image or less. The code amount is required, and the coding efficiency is reduced.
Therefore, if the quantization error MSE of the predicted image is reduced,
The prediction error of the current image also becomes small, the quantization error MSE can be kept small even with a small code amount, and high coding efficiency can be realized. Therefore, for the images used for prediction such as the intra frame I and the inter frame P, the quantization error MSE is made as small as possible and gradually increased in the input order (α> 1), and the interpolated frame B is set. In particular, if a quantization error MSE larger than the image used for prediction is assigned (β> 1), high coding efficiency can be obtained. The values of the coefficients α and β depend on the nature of the image, but in natural images such as television images,
By simulation, α = 1.07, γ = 1.32.
It was confirmed that the best coding efficiency was obtained by using the neighboring values.

FIG. 6 shows, from a typical television image,
It is the result of simulating the coding effect in the moving picture coding apparatus shown in FIG. 1 by extracting statistical properties. In this figure, the horizontal axis DLTB represents -log ₂ β, the vertical axis BIT _- RATE represents MSE = 0.05 x the average code amount when the image is dispersed, and the parameter DLTP is -log ₂ β.
indicates α. As can be seen from FIG. 6, DLTP = −0.
1 (α = 1.07), and DLTB = −0.4 (β =
In the case of 1.32), the average code amount becomes the minimum. As described above, in the first embodiment, the quantization error measuring means 21 is used when the image group composed of the intra frame I, the inter frame P and the interpolated frame B as shown in FIG. 2 is encoded. Since the quantization error MSE of each image output from is controlled by the quantization control means 22 so that the coding efficiency is maximized, the most efficient coding can be realized.

Second Embodiment FIG. 7 is a block diagram showing the configuration of a moving picture coding apparatus for storage according to a second embodiment of the present invention, which is common to the elements in FIG. 1 showing the first embodiment. Common elements are denoted by common reference numerals. This moving picture coding apparatus includes the same motion compensation predicting means 10, orthogonal transforming means 11, quantizing means 12, coding means 13, buffer 14, and code amount measuring means 20 as in the first embodiment. There is. The statistical property calculation means 30 is connected to the output side of the prediction means 10 with motion compensation,
The code amount determining means 31 is connected to the output side thereof, and the quantization control means 32 is further connected to the output sides of the code amount determining means 31 and the code amount measuring means 20.

The statistical property calculation means 30 calculates the statistical properties of the image based on the input image from the input terminal 1 and the prediction image output from the motion-compensated prediction means 10, and obtains information on the statistical properties. It has a function of giving the code amount determining means 31. The code amount determination means 31 determines the optimum code amount to be assigned to each image based on the information on the statistical properties of the image, and gives the determination result to the quantization control means 32. The quantization control means 32 has a function of controlling the quantization step width h given to the quantization means 12 based on the outputs of the code amount measuring means 20 and the code amount determining means 31.

Next, the operation will be described. When the image to be encoded is input to the input terminals 1, 2 and 3 in the order shown in FIG. 2, the motion-compensated prediction means 1 is used as in the first embodiment.
0, orthogonal transformation means 11, quantization means 12, coding means 1
3, the buffer 14, and the code amount measuring means 20 perform a predetermined encoding process.

In the statistical property calculating means 30, when the current image from the input terminal 1 and the predicted image of the current image output from the motion compensating prediction means 10 are input, the variance δ _{x of the} current image.
² , the correlation coefficient ρ _x , ρ _y , the correlation coefficient ρ _p between the current image and the inter-frame predicted image, the correlation coefficient ρ _b between the current image and the interpolated frame predicted image, etc. (14) to (18), the code amount determining means 3 is obtained.
Send to 1.

[0032]

[Equation 9]

[Equation 10] In the code amount determining means 31, information (δ _x ² , ρ _x , ρ _y , information on the statistical properties of the image calculated by the statistical property calculating means 30 is used.
ρ _p , δ _b ) is used to assign the optimum code amount distribution in the statistical property obtained in advance by a logical operation or simulation of an actual image to each image, and the code amount is sent to the quantization control means 32. Examples of the logical operation method of the optimum code amount include the following methods. First, the quantization error MSEP (i) of the intra frame I and the inter frame P in one image group is input in the order of input as shown in the equation (7): MSEP (i) = MSEP (i-1). α ... (7) α; coefficient It is set to be α times (α ≧ 1) of the quantization error MSEP (i−1) of the previous intraframe I or interframe P, and (i−1) The quantization error MSEB of the interpolated frame B in the i-th and i-th intra frames I or P is
As shown in 9), it is set to be β times (β> 1) of the closer intraframe I or interframe P. MSEB (j) = MSEP (i) · β (19) Here, the number of intra frames in one image group is N _i , the number of inter frames is N _p , and the number of interpolated frames is N _b = 2 ( (N _i + N _p ), the average quantization error MSE within the group is obtained as in the following equation (20). Therefore, the average quantization error MSEP (1), MS
EP (i) and MSEB (j) are the same as (10) and (11) above.
The equation and the following equation (21) are obtained.

[0033]

[Equation 11]

[Equation 12] For the first intra-frame I in the group, the relationship between the code amount due to encoding and the quantization error is expressed by the following equation (22).
It is represented by.

[0034]

[Equation 13] Regarding the interframe P in the group, the relationship between the code amount and the quantization error due to encoding is affected by the quantization error of the intraframe I or the interframe P used for forward prediction, and ) ~ (25)
become that way.

[Equation 14] For the interpolated frame B in the group,
Under the influence of the quantization error of the preceding and following prediction intra-frame I or inter-frame P, the relationship between the code amount by the encoding and the quantization error is as shown in the following equations (26) to (28).

[0035]

[Equation 15] f (ρ _x , ρ _y ) has a correlation coefficient ρ in the horizontal and vertical directions.
A geometric mean of variances of respective frequencies after orthogonal transformation (for example, DCT) of a stochastic process having a normal distribution of _x and ρ _y is shown, and is obtained from correlation coefficients ρ _x and ρ _y .

Next, the average code amount b _r of the images in the group
Is a predetermined code amount (transmission bit rate), and the average quantization error MSE is the smallest, the equations (11) and (21) to (28) are changed to the following equation (2).
9) and obtain the coefficients α and β.

[0037]

[Equation 16] Then, the calculated coefficients α and β are substituted into the above equation, and the code amount b
Obtain _rp (i) and b _rb (j). This code amount b _rp (i)
And b _rb (j) are the optimal allocation code amount of the image having the statistical property. Therefore, the quantization control unit 32 is based on the outputs of the code amount determination unit 31 and the code amount measurement unit 20.
By controlling the quantizing step width h given to the quantizing means 12 by means of the above, efficient coding can be performed. In the second embodiment, the statistical property calculating means 30 obtains the statistical property from a typical television image, and the average quantization error M
If the average code amount when the coefficients α and β are changed when SE is 0.05 times the variance of the image is calculated by the logical operation, the simulation as shown in FIG. 6 is performed as in the first embodiment. The result is obtained.

As can be seen from FIG. 6, DLTP = −0.1 (α = 1.07), and D as in the first embodiment.
When LTB = −0.4 (β = 1.32), the average code amount becomes the minimum. That is, the coefficients α = 1.07 and β = 1.3
When 2, the code amount is optimally assigned. If the coefficients α and β are calculated so that the average code amount at this optimum time becomes a predetermined code amount, the code amount of each image at that time becomes optimum. As an example, FIG. 8 shows the code amount of each image in the optimum state. In FIG. 8, the horizontal axis indicates an image group, the 0th image is an intra frame I, the 1st to 4th images are interframes P, and the 5th to 10th images are interpolated frames B. The vertical axis is the assigned code amount. As described above, according to the second embodiment, when the image group including the intra frame I, the inter frame P and the interpolated frame B as shown in FIG. , The quantization error and the allocated code amount of each image are calculated so that the coding efficiency is the best, and the quantization step width h is controlled by the quantization control means 32 based on the calculation result. A good encoding can be realized.

[0039]

As described in detail above, the first to third aspects
According to the invention, the quantizing error measuring unit measures the quantizing error of the image, and the quantizing control unit also controls the quantizing error of the image at the same time as the coding amount of the image to change the quantizing step width, Since I tried to control the quantizer,
For an image group including an intra frame, an inter frame, and an interpolated frame, the quantization error of each image is controlled so that the encoding efficiency becomes the highest, and efficient encoding is possible. According to the fourth to seventh inventions, the statistical property of the image is calculated by the statistical property calculating means, and the optimum code amount to be assigned to each image is determined by the code amount determining means based on the calculated statistical property. However, the quantization step size is changed by controlling the code amount of the image and the quantization error of the image at the same time by the quantization control means on the basis of the determination result, so that the intra frame, the inter frame and the interpolated frame are At the time of coding an image group composed of, the quantization error and the allocated code amount of each image are calculated so that the coding efficiency becomes the best, and efficient coding is possible.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a moving picture coding apparatus showing a first embodiment of the present invention.

[Fig. 2] Fig. 2 is a diagram illustrating a configuration example of an image group in moving image encoding.

FIG. 3 is a configuration block diagram of a conventional moving image encoding device.

FIG. 4 is a diagram showing an example of the zero-run level composite code of FIG.

5 is a diagram showing an example of the zero-run level composite code of FIG.

FIG. 6 is a diagram showing a simulation result of FIG. 1.

[Fig. 7] Fig. 7 is a configuration block diagram of a moving picture coding apparatus according to a second embodiment of the present invention.

8 is a diagram showing an example of an optimal allocation code amount of the image of FIG.

[Explanation of symbols]

 10 Prediction Means with Motion Compensation 11 Orthogonal Transformation Means 12 Quantization Means 13 Coding Means 14 Buffer 20 Code Quantity Measuring Means 21 Quantization Error Measuring Means 22, 32 Quantization Control Means 30 Statistical Property Calculating Means 31 Code Quantity Determining Means

Claims (7)

[Claims] 1. An intra frame among three types of coded images, that is, an intra frame that is an intra-coded image, an inter frame that is a forward predictive coded image, and an interpolated frame that is a bidirectional predictive coded image. Output as it is to the input image of the inter frame and the interpolated frame, the image is predicted in block units of M × M (where M is a positive integer) to form a predicted image, A motion compensation prediction unit that outputs a difference image between the current image and the prediction image, and an output image of the motion compensation prediction unit is N × N (however,
N; a positive integer), orthogonal transform means for orthogonally transforming each block, and quantizer means for quantizing the output of the orthogonal transform means using a given quantization step width, In the moving picture coding apparatus, the coding means for coding the output of the quantizing means, and the buffer for temporarily storing the output of the coding means and outputting at a constant bit rate, Based on the code amount measuring means for measuring the code amount of the image and the upper limit and the lower limit of the buffer occupancy, based on the input and output of the quantizing means, one block of the image, a few blocks or the whole image is quantized. Quantization error measuring means for measuring an error, and based on the outputs of the quantization error measuring means and the code amount measuring means, a quantization error of the measured image becomes a predetermined quantization error. Quantization for controlling the predetermined quantization error and the quantization step width given to the quantization means so that the group average code amount of the image obtained from the measured code amount becomes a predetermined code amount. A moving picture coding apparatus, comprising: a control means. 2. The quantization control means sets the input quantization error of the input intra-frame and inter-frame to be equal to or larger than the setting quantization error of the previous intra-frame or inter-frame, and For the interpolated frame that is set, the quantization step width is set so that the set quantization error becomes larger than the set quantization error of the temporally closer intra-frame or inter-frame before and after used for prediction. The moving picture coding apparatus according to claim 1, wherein the moving picture coding apparatus is configured to control. 3. The quantization control means, wherein the quantization error of the input inter-frame is α times the quantization error of the intra-frame or inter-frame used for forward prediction (where 1 ≦ α ≦ 2). ), The quantization error of the input interpolated frame is β times the quantization error of the forward or backward nearer intra-frame or inter-frame used for prediction (where 1 ≦ β ≦ 2). The moving picture coding apparatus according to claim 1, wherein the quantization step width is controlled so that 4. The intra frame among three types of coded images, that is, an intra frame that is an intra-coded image, an inter frame that is a forward predictive coded image, and an interpolated frame that is a bidirectional predictive coded image. Output as it is to the input image of the inter frame and the interpolated frame, the image is predicted in block units of M × M (where M is a positive integer) to form a predicted image, A motion compensation prediction unit that outputs a difference image between the current image and the prediction image, and an output image of the motion compensation prediction unit is N × N (however,
N; a positive integer), orthogonal transform means for orthogonally transforming each block, and quantizer means for quantizing the output of the orthogonal transform means using a given quantization step width, In the moving picture coding apparatus, the coding means for coding the output of the quantizing means, and the buffer for temporarily storing the output of the coding means and outputting at a constant bit rate, Based on the code amount measuring means for measuring the code amount of the image and measuring the upper and lower limits of the buffer occupancy amount, the current image, and the predicted image of the current image output from the motion-compensated prediction means, Statistical property calculating means for calculating the statistical properties of the image, based on the calculated statistical properties of the image, so that the average code amount in the image group becomes a predetermined code amount, and the average amount of the image A code amount determining means for calculating an optimal code amount to be assigned to each image in the image group so that the sub-sizing error is minimized, and the quantization based on the outputs of the code amount measuring means and the code amount determining means. And a quantization control means for controlling the quantization step width given to the means. 5. The code amount determining means sets the quantization error of the interframe to be equal to or greater than the quantization error of a forward intra frame or interframe used for prediction, and the quantization error of the interpolated frame. 5. The moving image according to claim 4, wherein is set so as to be larger than a quantization error of a forward or backward intra frame or inter frame used for prediction, and the optimum allocation code amount is obtained. Encoding device. 6. The code amount determining means presumes several kinds of statistical properties of the image in advance, calculates the optimum allocation code amount for them, and calculates the statistical properties of the real image from the statistical property calculating means. 6. The moving image according to claim 5, wherein when the information is input, the one that most corresponds to the information of the statistical property is selected from the optimal allocation code amount calculated by the assumed image property. Image coding device. 7. The code amount determining means sets the input quantization error of the inter-frame to be α times the quantization error of the intra-frame or inter-frame used for forward prediction, and inputs the quantization error. The quantization error of the interpolated frame is set to β times the quantization error of the forward or backward nearer intra frame or inter frame used for prediction, and the optimal code amount to be assigned to each image is calculated. The moving picture coding device according to claim 6.