JP2004357079A - In-frame rate control method in dynamic image coding, in-frame rate control system, dynamic image coding device, and in-frame rate control program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide in-frame rate control with accuracy and stability in dynamic image coding. <P>SOLUTION: A target code amount computing unit 102 calculates a target generating code amount of a succeeding control unit for each control unit in a picture on the basis of a differential between a target value of an accumulated generated code amount and an actual value of the accumulated generated code amount in the picture till then. A quantization parameter computing unit 103 calculates an estimation value of a quantization parameter for generating the target generating code amount from a past encoding result. A quantization parameter smoothing unit 104 calculates a weighted mean value between the estimation value of the quantization parameter and the quantization parameter for each control unit which has been coded until then and uses it as a succeeding quantization parameter for each control unit. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an intra-frame rate control technique in video coding, and more particularly to a method, apparatus, and program for performing more accurate and stable rate control than before.
[0002]
[Prior art]
In video coding, rate control is generally performed to maintain the coding bit rate at a desired bit rate. In general, rate control can be divided into inter-frame rate control in which a target code amount is assigned to each picture, and intra-frame rate control in which a generated code amount of a picture is controlled in accordance with a given target code amount.
[0003]
Generally, a moving image is composed of a plurality of frames, and particularly in the case of interlaced video, each frame is composed of two fields. In moving image coding, there are a method of coding in units of frames and a method of coding in units of fields. These coding units are collectively referred to as a picture.
[0004]
Here, from the name of the control unit, the inter-frame rate control and the intra-frame rate control should be called inter-picture rate control and intra-picture rate control. Is generally used, so this name is used interchangeably. That is, the inter-frame rate control and the intra-frame rate control in this specification do not exclude the control of the coding in the field unit.
[0005]
In the intra-frame (in-picture) rate control, encoding is generally performed while controlling the quantization parameter of each encoding unit in the picture.
[0006]
In the intra-picture rate control method shown in MPEG-2 test model 5 (hereinafter referred to as TM5), in order to make the generated code amount of a picture equal to a given picture target code amount, quantization is performed for each macroblock. The parameters are determined (see Non-Patent Document 1).
[0007]
For each macroblock coding, a difference value is calculated by subtracting the target value from the actual value of the accumulated code amount in the picture up to that time. The target value of the cumulative code amount in a picture is the cumulative value of the target code amount of each macroblock when the target code amount of the picture is distributed to all macroblocks on average. The quantization parameter of each macroblock is a value proportional to a value obtained by adding an initial value to the difference value. As a result, when the difference value is large, the quantization parameter becomes large, and the generated code amount decreases. When the difference value is small, the quantization parameter becomes small, and the generated code amount increases. Is achieved.
[0008]
This is represented by the following equation.
[0009]
Q _j = D _j ・ 31 / r …… (1)
d _j = D ₀ + (B _j-1 − (J−1) · T / N) (2)
j is the macroblock number to be processed, N is the total number of macroblocks, B _j-1 Is a cumulative generated code amount up to the macroblock number j-1, T is a picture target code amount, and r is a parameter for determining the strength of control response. Also, d ₀ Is the variable d _j Of the last macroblock of the immediately preceding picture of the same type. _j Is used.
[0010]
This d ₀ Determines the quantization parameter of the first macroblock of the picture, the value of the quantization parameter is defined below as the initial value Q of the quantization parameter. ₀ I will call it. That is, Q ₀ = D ₀ ・ Set to 31 / r.
[0011]
In actual control, the quantization parameter is Q _j To the coefficient N according to the characteristics of each macroblock _act Multiplied by 2 and then the value mquant which is clipped to a value (2-62 or 1-112) that can be taken in the standard _j Is used, but here Q _j Is referred to as a quantization parameter.
[0012]
mquant _j = Clip (2.0 · Q _j ・ N _act ) …… (3)
Since the above-described method has a relatively small amount of calculation and well-known control characteristics, a control method based on this control method is widely used.
[0013]
Although an example of control in units of macro blocks has been described, similar coding control may be performed in other units. For example, in TM5, the control unit is a macroblock unit, but from the viewpoint of mounting, a larger control unit may be advantageous because the calculation amount is smaller. Therefore, there is a method in which the control unit is a slice unit which is a set of horizontally continuous macroblocks, and the above expression is calculated only once for each slice to determine the quantization parameter of the slice. In this case, although the response of the control is slightly delayed, there is an advantage that the amount of calculation for determining the quantization parameter can be reduced.
[0014]
FIG. 5 is a diagram showing a configuration example of a conventional encoding device using the intra-picture rate control method shown in the TM5. In FIG. 5, reference numeral 2 denotes an encoding device; 11, a subtraction unit for subtracting a signal from the motion search / compensation unit 18 from an input image signal to obtain a prediction error; 12, a DCT unit for performing DCT conversion of the prediction error; A quantizing unit that quantizes the transform coefficient obtained in 12 is an inverse quantizing unit that inversely quantizes the quantized transform coefficient obtained in the quantizing unit 13.
[0015]
Reference numeral 15 denotes an inverse DCT unit for performing an inverse DCT transform of the inversely quantized coefficient, 16 an addition unit for adding a signal from the motion search / compensation unit 18 to the difference image obtained by the inverse DCT unit 15, and 17 a decoded image. A frame memory 18, a motion search / compensation unit for performing motion search and motion compensation using the decoded image stored in the frame memory 17 as a reference frame, and 19 a variable-length quantization transform coefficient obtained by the quantization unit 13. This is a variable length coding unit to be coded.
[0016]
Reference numeral 20 denotes an encoding control unit for controlling the generated code amount of a picture; 60, an intra-frame quantization control unit for controlling a quantization parameter in a frame (in a picture); The accumulated code amount accumulating unit 101 is a target accumulated code amount calculating unit that calculates a target value of the accumulated code amount in the picture up to the present time.
[0017]
A subtraction unit 105 subtracts the value of the accumulated code amount recorded in the accumulated code amount accumulation unit 100 from the target value of the accumulated code amount calculated by the target accumulated code amount calculation unit 101, and a subtraction unit 106 subtracts 106. A quantization parameter calculation unit that calculates a quantization parameter based on a result.
[0018]
First, the DCT unit 12 performs DCT on the prediction error obtained by the subtraction unit 11, and the quantization unit 13 quantizes the transform coefficient according to the quantization parameter Q determined by the coding control unit 20. The inverse quantization unit 14 and the inverse DCT unit 15 respectively perform inverse quantization and inverse DCT transform on the quantized transform coefficient, and the addition unit 16 adds the signal from the motion search / compensation unit 18 to the result. to add. The decoded image obtained as a result is stored in the frame memory 17.
[0019]
On the other hand, the variable length coding unit 19 performs variable length coding on the transform coefficient after quantization by the quantization unit 13 and accumulates information on the generated code amount R in the intra-frame quantization control unit 60. Transmit to the unit 100. The accumulated code amount accumulating unit 100 accumulates and records the transmitted generated code amount R up to that time.
[0020]
Then, the subtraction unit 105 subtracts the value of the actual accumulated code amount recorded in the accumulated code amount accumulation unit 100 from the target value of the accumulated code amount calculated by the target accumulated code amount calculation unit 101. The quantization parameter calculation unit 106 calculates the quantization parameter Q of the next control unit based on the result of the subtraction by the subtraction unit 105, and transmits information on the calculated quantization parameter Q to the quantization unit 13.
[0021]
[Non-patent document 1]
MPEG-2 Test Model 5, Document ISO / IEC JTC1 SC29 WG11 / 93-400, Test Model Editing Committee, April 1993.
[0022]
[Problems to be solved by the invention]
However, the above-mentioned prior art method has the following problems. One is the problem of control stability. Generally, since the quantization parameter Q and the generated code amount R have a relationship close to inverse proportion, as shown in FIG. 6, the change in the code amount when the quantization parameter is small is smaller than the change in the code amount when the quantization parameter is large. Is big.
[0023]
In TM5, since the value of the quantization parameter is changed in proportion to the difference between the actual value of the accumulated code amount in the picture and the target value, the difference takes a negative value and the quantization parameter is small. Sometimes the response is particularly sensitive. Therefore, in systems where the control unit is not a macroblock unit but a slice unit, and the feedback is slow, the code amount explosively increases after the quantization parameter becomes unnecessarily small in the control process and the control becomes oscillatory. Problem.
[0024]
The other is the problem of control accuracy. In TM5, if the actual value of the accumulated code amount in the picture and the target value match during the control process, d _j = D ₀ And the quantization parameter is the initial value Q ₀ It becomes. But Q ₀ Is not an appropriate value for the picture of the picture and the target code amount, the generated code amount acts again in a direction in which the actual value of the intra-picture accumulated generated code amount deviates from the target value.
[0025]
For example, the initial value Q ₀ Is smaller than an appropriate value, the code amount is too large, and the difference value between the actual value of the accumulated code amount in the picture and the target value increases again. For this reason, Q ₀ Is not an appropriate value, a deviation between the generated code amount and the target code amount always remains.
[0026]
From another viewpoint, if the generated code amount continues to be T / N with a certain quantization parameter in the control process, the difference between the actual value and the target value of the accumulated code amount in the picture does not change. There is a problem that the quantization parameter does not change, and the system falls into a steady state where it does not approach the target code amount any more. For this reason, there is a problem that the target code amount given to the picture cannot be accurately realized.
[0027]
SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems of the prior art and realize an intra-frame rate control that enables accurate and stable rate control.
[0028]
[Means for Solving the Problems]
In order to solve the above problem, the present invention determines a target code amount of the next control unit for each control unit in a picture, and multiplies the generated code amount of the control unit encoded up to that by the quantization parameter. Is divided by the target value of the generated code amount to estimate the quantization parameter for generating the target code amount of the next control unit. Alternatively, a value proportional to the reciprocal of the target code amount is set as the quantization parameter.
[0029]
Then, rate control is performed by using the weighted average value of the quantization parameter thus obtained and the quantization parameter of the control unit coded so far as the quantization parameter of the next control unit. By performing such control, it is possible to solve the above-mentioned problems relating to the stability and accuracy of the control.
[0030]
The operation of the present invention is as follows. In the present invention, the target code amount of the next control unit is determined, and a quantization parameter for generating the code amount is estimated, or a value proportional to the reciprocal of the target generated code amount is used as the quantization parameter. This makes it possible to prevent the control amount from becoming unstable due to an abnormally large or non-existent code amount. Here, the estimation of the quantization parameter for generating the target code amount is performed, for example, by multiplying the product of the generated code amount of the control unit encoded so far and the quantization parameter by the target value of the generated code amount of the next control unit. This is performed by, for example, using a value obtained by dividing by an estimated value.
[0031]
The difference between the above-described conventional method (TM5) and the rate control method according to the present invention will be described with reference to the conventional rate control conceptual diagram shown in FIG. 7 and the rate control conceptual diagram according to the present invention shown in FIG. In the conventional method, as shown in FIG. 7, the target value (jT / N) and the accumulated code amount (B _j ), A quantization parameter Q is calculated based on the difference value, coding is performed based on the calculated quantization parameter Q, and a generated code amount R is calculated by an integrator (accumulated code amount in FIG. 5). The accumulating unit 100) employs a method of accumulating. However, in the conventional method, since the control target does not have a linear characteristic and the characteristic is not corrected, the whole control is difficult to stabilize.
[0032]
On the other hand, in the present invention, the target value (jT / N) and the accumulated code amount (B _j In calculating the quantization parameter Q on the basis of the difference value from the above, a target code amount R ′ is calculated from the difference value, and the quantization parameter Q ′ for generating the calculated target code amount R ′ is estimated. The quantized parameter Q ′ is smoothed to determine the quantized parameter Q.
[0033]
In the present invention, in the calculation of the quantization parameter, a conversion having a characteristic opposite to the characteristic of the control target, that is, a value (temporary quantization parameter) proportional to the reciprocal of the target generated code amount is obtained or the target code amount is calculated. In order to perform the operation of estimating the quantization parameter for generating the error, the relationship between the difference value between the accumulated code amount and the target value and the code amount generated by the coding becomes linear, and the stability of the entire control system increases.
[0034]
By appropriately determining the target code amount for each control unit, the target code amount given to the picture can be accurately realized. For example, from the value obtained by dividing the target code amount T of the picture by the number N of control units in the picture, the amount by which the code amount that has passed so far is returned by a constant rate K is subtracted to obtain the target code amount of the next control unit. Set. That is, the target code amount of the next control unit is
Target code amount = T / NK- (B _j-1 − (J−1) · T / N)
By doing so, the target code amount given to the picture can be realized more accurately than in the past.
[0035]
On the other hand, since the relationship between the generated code amount and the quantization parameter for each control unit in a picture is usually greatly different, the quantization parameter is set such that the generated code amount for each control unit is relatively close to a constant value as described above. Is determined, the fluctuation of the quantization parameter for each control unit becomes large. Therefore, the variation is suppressed by taking a weighted average value of the obtained quantization parameter and the quantization parameter of the control unit coded so far.
[0036]
For example, a 7: 1 weighted average is calculated between the quantization parameter of the control unit encoded immediately before and the quantization parameter obtained this time, and this value is used as the quantization parameter. As a result, although the accuracy of the control is slightly lowered, the fluctuation of the quantization parameter for each control unit in the picture can be smoothed, and visually excellent control can be realized. In addition, it is possible to obtain superior performance with respect to control stability and accuracy.
[0037]
That is, the present invention determines a target generated code amount of the next control unit for each control unit in a picture, estimates a quantization parameter for generating the target generated code amount from past coding results, A weighted average value of the quantization parameter of the control unit and the quantization parameter of the control unit encoded up to that time is used as the quantization parameter of the next control unit.
[0038]
Further, the present invention determines a target generated code amount of the next control unit for each control unit in a picture based on a difference between a target value and an actual value of the accumulated generated code amount in the picture up to that time. The product of the generated code amount of the control unit coded up to and the quantization parameter is divided by the target value of the generated code amount of the next control unit to determine the quantization parameter, and the value and the control coded so far are determined. It is characterized in that a weighted average value with a unit quantization parameter is used as a quantization parameter of the next control unit.
[0039]
Also, the present invention determines a target generated code amount of the next control unit for each control unit in a picture, determines a quantization parameter in proportion to the reciprocal thereof, and determines the determined value and the coding value up to that. The weighted average value of the obtained control unit and the quantization parameter is used as the quantization parameter of the next control unit.
[0040]
BEST MODE FOR CARRYING OUT THE INVENTION
An embodiment of the present invention will be described below with reference to the drawings.
[0041]
[Example of device configuration]
FIG. 2 is a diagram showing a configuration example of an encoding device using the intra-picture rate control method of the present invention. 1 is an encoding device, 10 is an encoding control unit for controlling the amount of generated code of a picture, and 50 is an intra-frame quantization control unit for controlling a quantization parameter in a frame (in a picture). The configuration of the encoding apparatus 1 excluding the encoding control unit 10 is the same as the configuration of the conventional encoding apparatus 2 except for the encoding control unit 20 described with reference to FIG.
[0042]
In FIG. 2, reference numeral 100 denotes a cumulative code amount accumulating unit that accumulates the generated code amount up to that point in the picture, and 101 denotes a target cumulative code amount calculating unit that calculates a target value of the current cumulative generated code amount in the picture. , 102 are target code amount calculation units for calculating the target code amount R ′ of the next control unit based on the result of the subtraction by the subtraction unit 105.
[0043]
Reference numeral 103 denotes a quantization parameter calculation unit that calculates a quantization parameter Q ′ corresponding to the target code amount R ′ calculated by the target code amount calculation unit 102, and 104 denotes a quantization parameter Q ′ calculated by the quantization parameter calculation unit 103. This is a quantization parameter smoothing unit that performs smoothing and calculates a quantization parameter Q of the next control unit.
[0044]
The operation of the encoding device 1 is as follows. First, the DCT unit 12 performs DCT on the prediction error obtained by the subtraction unit 11, and the quantization unit 13 quantizes the transform coefficient according to the quantization parameter Q determined by the coding control unit 10. The inverse quantization unit 14 and the inverse DCT unit 15 respectively perform inverse quantization and inverse DCT transform on the quantized transform coefficient, and the addition unit 16 adds the signal from the motion search / compensation unit 18 to the result. to add. The decoded image obtained as a result is stored in the frame memory 17.
[0045]
On the other hand, the variable-length coding unit 19 performs variable-length coding on the transform coefficient after quantization by the quantization unit 13, and stores information on the generated code amount R in the cumulative code amount accumulating unit 100 in the intra-frame quantization control unit 50. Send to
[0046]
The accumulated code amount accumulating unit 100 accumulates and records the transmitted generated code amount R up to that time. Then, the subtraction unit 105 subtracts the value of the actual accumulated code amount recorded in the accumulated code amount accumulation unit 100 from the target value of the accumulated code amount calculated by the target accumulated code amount calculation unit 101. The target code amount calculation unit 102 calculates a target code amount R ′, which is a target value of the generated code amount of the next control unit, based on the result of the subtraction by the subtraction unit 105, and information on the calculated target code amount R ′. Is transmitted to the quantization parameter calculation unit 103. The quantization parameter calculator 103 calculates a quantization parameter Q ′ for generating the target code amount R ′ based on the target code amount R ′ transmitted from the target code amount calculator 102.
[0047]
Here, as will be described later, the quantization parameter Q ′ for generating the target code amount R ′ is, for example, a product of the generated code amount R of the control unit encoded so far and the quantization parameter, It is calculated by dividing by R ′ or as a value proportional to the reciprocal of the target code amount R ′.
[0048]
The quantization parameter smoothing unit 104 smoothes the quantization parameter Q ′ calculated by the quantization parameter calculation unit 103 so that the quantization parameter does not change rapidly, and information of the quantization parameter Q obtained as a result of the smoothing. Is transmitted to the quantization unit 13. Here, in the smoothing process performed by the quantization parameter smoothing unit 104, for example, the weighted average value of the quantization parameter Q ′ and the quantization parameter of the control unit encoded up to that point is calculated by using the weighted average value of the next control unit. This is performed by setting the quantization parameter Q.
[0049]
[First Embodiment]
FIG. 3 is a diagram showing a flow of the rate control process according to the first embodiment of the present invention. In the first embodiment of the present invention, it is assumed that the intra-picture rate control is performed on a slice-by-slice basis. It is assumed that the number of slices in a picture is N.
[0050]
First, a target code amount T of a picture to be encoded is determined by an inter-picture rate control algorithm (step S1). Although this algorithm is not within the scope of the present invention, for example, TM5 introduces a method of calculating a picture target code amount from past coding results (generated code amounts and quantization parameters) and remaining GOP code amounts. .
[0051]
At the beginning of picture coding, the quantization parameter value Q of the first slice (slice 0) ₀ To determine. If there is no past picture having the same picture type as the picture to be encoded (step S2), the I and P pictures have Q ₀ = 10, Q for B picture ₀ = 14 (step S3). If there is, the quantization parameter of the last slice of the closest past picture of the same picture (the immediately preceding same picture) is set to the quantization parameter Q of the first slice. ₀ (Step S4).
[0052]
This quantization parameter Q ₀ Is used to encode the first slice (slice 0) (step S5). In practice, the quantization parameter Q ₀ To the coefficient N according to the characteristics of each macroblock _act Multiplied by 2 and then the value mquant which is clipped to a value (2-62 or 1-112) that can be taken in the standard _j , Each macro block in slice 0 is encoded. When the encoding of slice 0 is completed, the generated code amount R of slice 0 ₀ And the quantization parameter Q of the next slice 1 ₁ Is determined as follows.
[0053]
Here, more generally, the quantization parameter Q of slice j _j Describes how to determine Here, j is 1 or more and N-1 or less. When the encoding of the slice j-1 is completed, the generated code amount R of the slice _j-1 (Step S6), and the generated code amount R of slice j-1 _j-1 And the quantization parameter Q _j-1 And the complexity index X of slice j-1 _j-1 Is obtained (step S7).
[0054]
X _j-1 = Q _j-1 ・ R _j-1 …… (4)
Next, the generated code amount R of slice j-1 _j-1 , The accumulated code amount B in the picture up to slice j-1 _j-1 Is obtained (step S8). If j = 1, B ₀ = R ₀ And when j ≧ 2, it is as follows.
[0055]
B _j-1 = R _j-1 + B _j-2 …… (5)
Next, the target code amount R ′ of slice j _j Is determined (step S9). Target code amount R ' _j Is determined by “(average target code amount per slice) − (repayment code amount)”. The average target code amount per slice is a value obtained by dividing the target code amount T for one picture by the number of slices N in a picture, and the repayment code amount is the difference between the actual value of the accumulated code amount at that time and the target value. The value is multiplied by a constant rate K (for example, 0.1).
[0056]
R ' _j = T / N-0.1 · (B _j-1 − (J−1) · T / N) (6)
The target code amount R ′ of this slice j _j The quantization parameter Q ′ for generating _j To the complexity index X of slice j-1 _j-1 Is determined using That is, as shown below, the complexity index X of slice j-1 _j-1 With the target code amount R ′ of slice j _j Divided by the quantization parameter Q _j 'Is obtained (step S10).
[0057]
Q ' _j = X _j-1 / R ' _j ...... (7)
This is based on the approximation that the quantization parameter and the generated code amount are inversely proportional, and the approximation that the complexity index of the immediately preceding slice and the complexity index of the slice to be encoded do not change much.
[0058]
The value Q 'found this time _j And the value Q determined in the previous slice _j-1 And a: (1-a) (a is, for example, 1/8), a weighted average is taken, and the quantization parameter Q of the slice j is calculated. _j Is determined (step S11).
[0059]
Q _j = 1/8 Q ' _j + 7/8 · Q _j-1 …… (8)
This has the effect of smoothing the quantization parameter, because if the quantization parameter changes abruptly for each slice, it is undesirable in terms of image quality and stability.
[0060]
In practice, the quantization parameter Q _j To the coefficient N according to the characteristics of each macroblock _act Multiplied by 2 and then the value mquant which is clipped to a value (2-62 or 1-112) that can be taken in the standard _j Is used to encode each macroblock in slice j (step S12).
[0061]
Steps S6 to S12 are repeated for all slices in the picture (step S13), and when the picture encoding is completed, the quantization parameter Q _j Is stored for the subsequent initial value of the same kind of picture (step S14).
[0062]
This is repeated for all picture encodings (step S15). By doing so, more accurate and stable rate control than before can be performed.
[0063]
[Second embodiment]
FIG. 4 is a diagram showing a rate control processing flow according to the second embodiment of the present invention. In the second embodiment of the present invention, it is assumed that in-picture rate control is performed by using M macroblocks as one control unit (hereinafter referred to as MB group). It is assumed that the number of MB groups in a picture is N.
[0064]
First, a target code amount T of a picture to be encoded is determined by an inter-picture rate control algorithm (step S21). This algorithm is not within the scope of the present invention. At the beginning of picture coding, the quantization parameter value Q of the first control unit (MB group 0) ₀ To determine. If there is no past picture having the same picture type as the picture to be encoded (step S22), Q ₀ = 10, Q for B picture ₀ = 14 (step S23). If there is, the quantization parameter of the last MB group of the closest past picture of the same picture (the immediately preceding same picture) is Q ₀ (Step S24).
[0065]
This quantization parameter Q ₀ Is used to encode MB group 0 (step S25). In practice, the quantization parameter Q ₀ To the coefficient N according to the characteristics of each macroblock _act Multiplied by 2 and then the value mquant which is clipped to a value (2-62 or 1-112) that can be taken in the standard _j Is used to encode each macroblock in MB group 0.
[0066]
When the encoding of the MB group 0 is completed, the generated code amount R of the MB group ₀ And quantize parameter Q of MB group j _j Is determined as follows. Here, more generally, the quantization parameter Q of MB group j _j Describes how to determine Here, j is 1 or more and N-1 or less.
[0067]
When the encoding of the MB group j-1 is completed, the generated code amount R of the MB group _j-1 Are counted (step S26). Next, the complexity index X of the MB group _j-1 (X _j-1 = R _j-1 ・ Q _j-1 ) Is calculated and stored (step S27). Then, the generated code amount R of the MB group j-1 _j-1 , The accumulated code amount B in the picture up to the MB group j−1 _j-1 Is obtained (step S28). If j = 1, B ₀ = R ₀ And when j ≧ 2, it is as follows.
[0068]
B _j-1 = R _j-1 + B _j-2 …… (9)
Next, the target code amount R ′ of the MB group j _j Is determined (step S29). Target code amount R ' _j Is determined by “(average target code amount per 1 MB group) − (repayment code amount)”. The average target code amount per MB group is a value obtained by dividing the target code amount T for one picture by the number N of MB groups in a picture, and the repayment code amount is the actual value and the target value of the accumulated code amount at that time. Is divided by the number of remaining control units N-j.
[0069]
R ' _j = T / N- (B _j-1 − (J−1) · T / N) / (N−j) (10)
The target code amount R ′ of this MB group j _j The value proportional to the reciprocal of _j (Step S30).
[0070]
Q ' _j = X / R ' _j …… (11)
The proportionality constant X used at this time is the average value of the complexity index of all MB groups of the immediately preceding same picture, that is, the average value of the product of the generated code amount of the MB group and the average quantization parameter. Use the same value. By doing so, the code amount actually generated when encoding the MB group j is equal to the target code amount R ′ of the MB group j. _j Is not necessarily the same as the temporary quantization parameter Q ′ _j Has the advantage that the fluctuation is reduced and it is visually desirable.
[0071]
Then, the temporary quantization parameter Q ′ obtained this time _j And the tentative quantization parameters Q ′ of the previous two MB groups _j-1 And a weighted average (moving average) at a ratio of 1: 1: 1 to obtain the quantization parameter Q of the MB group j as follows: _j Is determined (step S31).
[0072]
Q _j = (Q ' _j + Q ' _j-1 + Q ' _j-2 ) / 3 ... (12)
In practice, the quantization parameter Q _j To the coefficient N according to the characteristics of each macroblock _act Multiplied by 2 and then the value mquant which is clipped to a value (2-62 or 1-112) that can be taken in the standard _j Is used to encode each macroblock in MB group j (step S32). Steps S26 to S32 are repeated for all the MB groups in the picture (step S33), and when the picture encoding is completed, the quantization parameter Q of the last MB group is _j Is stored for the subsequent initial value of the same type of picture (step S34). Further, the average value X of the complexity index of the MB group is stored (step S35).
[0073]
This is repeated for all picture encodings (step S36). By doing so, more accurate and stable rate control than before can be performed.
[0074]
The processing of each of the embodiments described above can be realized by hardware or a processor dedicated to encoding processing, and can also be realized by a general-purpose computer and a software program. The software program can be provided by being recorded on a computer-readable recording medium or can be provided through a network.
[0075]
【The invention's effect】
According to the present invention, the target code amount of the next control unit is determined, and a quantization parameter for generating the code amount is estimated, or a value proportional to the reciprocal of the target generated code amount is set as the quantization parameter. By doing so, it is possible to prevent the control amount from becoming unstable due to abnormal or excessive output of the code amount.
[0076]
Further, according to the present invention, for example, a value obtained by dividing a target code amount T of a picture by the number N of control units in a picture and subtracting an amount of the code amount that has been passed so far by a constant rate K is subtracted from the next control value. By appropriately determining the target code amount for each control unit, for example, by setting the target code amount in units, the target code amount given to the picture can be accurately realized.
[Brief description of the drawings]
FIG. 1 is a conceptual diagram of rate control according to the present invention.
FIG. 2 is a diagram illustrating a configuration example of an encoding device using the intra-picture rate control method of the present invention.
FIG. 3 is a diagram showing a flow of a rate control process according to the first embodiment of the present invention.
FIG. 4 is a diagram showing a rate control processing flow according to a second embodiment of the present invention.
FIG. 5 is a diagram illustrating a configuration example of a conventional encoding device.
FIG. 6 is a diagram illustrating a relationship between a quantization parameter and a generated code amount.
FIG. 7 is a conceptual diagram of a conventional rate control.
[Explanation of symbols]
1,2 coding device
10,20 encoding control unit
11 Subtraction unit
12 DCT section
13 Quantization unit
14 Inverse quantization unit
15 Inverse DCT section
16 Addition unit
17 Frame memory
18 Motion search / compensation unit
19 Variable length coding unit
50,60 In-frame quantization controller
100 Cumulative code amount accumulator
101 Target cumulative code amount calculation unit
102 Target code amount calculation unit
103, 106 Quantization parameter calculator
104 Quantization parameter smoothing unit
105 Subtraction unit

Claims (8)

In an intra-frame rate control method in video coding,
Determining, for each control unit in the picture, a target generated code amount for the next control unit;
Estimating a quantization parameter for generating the target generated code amount from a past coding result;
Moving the weighted average value between the estimated value of the quantization parameter and the quantization parameter of the control unit coded so far as the quantization parameter of the next control unit. In-frame rate control method in multiplexing. In an intra-frame rate control method in video coding,
Determining, for each control unit in the picture, a target generated code amount of the next control unit based on a difference between the target value of the accumulated generated code amount in the picture and the actual accumulated generated code amount;
Deciding a quantization parameter by dividing a product of the generated code amount of the control unit encoded up to then and the quantization parameter by a target value of the generated code amount of the next control unit;
A step of using a weighted average value of the determined value and the quantization parameter of the control unit encoded so far as the quantization parameter of the next control unit. Internal rate control method. In an intra-frame rate control method in video coding,
Determining, for each control unit in the picture, a target generated code amount for the next control unit;
Determining a quantization parameter proportional to the reciprocal of the determined target generated code amount;
A step of using a weighted average value of the determined value and the quantization parameter of the control unit encoded so far as the quantization parameter of the next control unit. Internal rate control method. In the intra-frame rate control device in video coding,
A target code amount calculation unit for determining a target generated code amount of the next control unit for each control unit in the picture;
A quantization parameter calculator for estimating a quantization parameter for generating the target generated code amount from a past coding result;
A quantization parameter smoothing unit that uses a weighted average value of the estimated value of the quantization parameter and the quantization parameter of the control unit encoded so far as a quantization parameter of the next control unit. Intra-frame rate control device in video coding. In the intra-frame rate control device in video coding,
Target code amount calculation for determining the target generated code amount of the next control unit for each control unit in the picture based on the difference between the target value of the cumulative generated code amount in the picture up to that point and the actual accumulated generated code amount Department and
A quantization parameter calculator for dividing a product of the generated code amount of the control unit encoded up to that and the quantization parameter by a target value of the generated code amount of the next control unit to determine a quantization parameter;
A moving image, comprising: a quantization parameter smoothing unit that uses a weighted average value of the determined value and the quantization parameter of the control unit encoded so far as a quantization parameter of the next control unit. An intra-frame rate control device in image coding. In the intra-frame rate control device in video coding,
A target code amount calculation unit for determining a target generated code amount of the next control unit for each control unit in the picture;
A quantization parameter calculator for determining a quantization parameter proportional to a reciprocal of the determined target generated code amount;
A moving image, comprising: a quantization parameter smoothing unit that uses a weighted average value of the determined value and the quantization parameter of the control unit encoded so far as a quantization parameter of the next control unit. An intra-frame rate control device in image coding. An intra-frame rate control device in video coding according to any one of claims 4 to 6,
Means for performing variable-length coding of a moving image using a quantization parameter determined for each control unit in a picture by the intra-frame rate control device in the moving image coding. . 4. An intra-frame rate control program in video encoding for causing a computer to execute the intra-frame rate control method in video encoding according to any one of claims 1 to 3.

Images