JPH07322245A - Video signal coding system - Google Patents

Abstract

PURPOSE:To make a generated code quantity stable corresponding to malfunction of motion compensation and corresponding to a local change in an image by using an in-image signal power and an inter-image signal power thereby calculating an adaptive quantization coefficient. CONSTITUTION:An output of a target complexity calculation circuit 3a is given to a 1st input of a target bit quantity calculation circuit 4a and a buffer residual amount 207 received from an input terminal 1f is given to a 2nd input of the circuit 4a. An output of the circuit 4a is given to a 1st input of a virtual buffer residual amount calculation circuit 5a and a generated code quantity 209 received from a terminal 1g is given to a 2nd input of the circuit 5a. An output of the circuit 5a is given to an input of a quantization step parameter calculation 6a. An output of the circuit 6a is given to a 1st input of an adaptive quantization circuit 7a, an activity 212 is given from a terminal 1h to the 2nd input, and a newest average inter-image signal power 213 from a terminal 1i is given to the 3rd input. A quantization step parameter 214 being an output of the circuit 7a is provided as an output.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a video signal coding system, and more particularly to a video signal quantization control system.

[0002]

2. Description of the Related Art FIG. 5 shows, for example, ISO-IEC / JTC1 / SC29 / WG.
11 is a schematic block diagram showing a conventional video signal encoding method shown in 11 MPEG 92 / N0245 Test Model 2.

In FIG. 5, a digitized video signal 1101 input from an input terminal 101a has a first input of a subtractor 9, a first input of a motion compensation prediction circuit 16 and a second input of a quantization circuit 11. Given to the input of. The output 1102 of the subtractor 9 is given to the first input of the quantization circuit 11 via the DCT circuit 10. The output 1104 of the quantization circuit 11 is given to the input of the transmission buffer 19 via the variable length coding circuit 18, and the dequantization circuit 1
2 and the IDCT circuit 13 to the first input of the adder 14. The output 1107 of the adder 14 is supplied to the first input of the memory circuit 15, and the output 1108 of the memory circuit 15 is supplied to the second input of the motion compensation prediction circuit 16 and the first input of the switching circuit 17. .

The second input of the memory circuit 15 is supplied with the first output 1111 of the motion compensation prediction circuit 16. On the other hand, the zero signal is given to the second input of the switching circuit 17, and the second output 1110 of the motion compensation prediction circuit 16 is given to the third input of the switching circuit 17. The output 1109 of the switching circuit 17 is given to the second input of the subtractor 9 and the second input of the adder 14. On the other hand, the second output 1113 of the transmission buffer 19 is the quantization circuit 11
And the first output 1114 of the transmission buffer 19 is output from the output terminal 2d.

FIG. 6 shows, for example, ISO-IEC / JTC1 / SC29 / WG11.
FIG. 11 is a schematic block diagram showing an example of a quantization control method in the conventional video signal coding method shown in MPEG 92 / N0245 Test Model 2.

In FIG. 6, the generated code amount 1201 in the previous I picture input from the input terminal 101d is given to the first input of the complexity calculation circuit 20a, and is averaged in the previous I picture input from the input terminal 101e. The quantization step parameter 1202 is the complexity calculation circuit 20a.
Given to the second input of. The I-picture complexity 1203, which is the output of the complexity calculation circuit 20a, is given to the first input of the target bit amount calculation circuit 4e. The generated code amount 1204 in the previous P picture input from the input terminal 101f is given to the first input of the complexity calculation circuit 20b, and the average quantization step parameter 1205 in the previous P picture input from the input terminal 101g is complicated. It is given to the second input of the degree calculation circuit 20b.

The complexity 1206 of the P picture, which is the output of the complexity calculating circuit 20b, is given to the second input of the target bit amount calculating circuit 4e. Generated code amount 1207 in the previous B picture input from the input terminal 101h
Is given to the first input of the complexity calculating circuit 20c, and the average quantization step parameter 1208 in the previous B picture inputted from the input terminal 101i is given to the second input of the complexity calculating circuit 20c. Complexity calculation circuit 20
The complexity 1209 of the B picture, which is the output of c, is given to the third input of the target bit amount calculation circuit 4e.

The fourth input of the target bit amount calculation circuit 4e is the remaining number 1210 of P pictures in the current GOP after the previous picture is encoded from the input terminal 101j.
However, for the fifth input, the remaining number of B-pictures 121 in the current GOP after the previous picture encoding from the input terminal 101k is 121.
1 is input. The sixth input has a constant Kp1212 which is the output of the constant generation circuit 21a, and the seventh input has a constant Kb1213 which is the output of the constant generation circuit 21b.
Is entered. The eighth input has an input terminal 101.
From l, the residual code amount 1214 assigned to the current GOP after the previous picture encoding is input. A picture type 1215 is input to the ninth input from the input terminal 101m.

The target bit amount 1217 which is the output of the target bit amount calculating circuit 4e is given to the first input of the virtual buffer selecting circuit 22a. The generated code amount up to the current macroblock in the current picture input from the input terminal 101c is input to the second input of the virtual buffer selection circuit 22a. A picture type 1215 is input to the third input from the input terminal 101m. The target bit amount 1218, which is the first output of the virtual buffer selection circuit 22a, is input to the first input of the virtual buffer remaining amount calculation circuit 5e for I-pictures up to the current macroblock in the current picture that is the second output. The generated code amount 1219 is input to the second input of the virtual buffer remaining amount calculation circuit 5e for I pictures.

The target bit amount 1220, which is the third output of the virtual buffer selection circuit 22a, is input to the first input of the virtual buffer remaining amount calculation circuit 5f for P pictures, and the current macro of the current picture, which is the fourth output, is input to the first input. The generated code amount 1221 up to the block is input to the second input of the virtual buffer remaining amount calculation circuit 5f for P picture. The target bit amount 1222, which is the fifth output of the virtual buffer selection circuit 22a, is input to the first input of the virtual buffer remaining amount calculation circuit 5g for B pictures, and is output to the current macroblock in the current picture that is the sixth output. The generated code amount 1223 is
It is input to the second input of the virtual buffer remaining amount calculation circuit 5g for B pictures.

The output 1224 of the virtual buffer remaining amount calculation circuit 5e for I pictures is input to the first input of the virtual buffer selection circuit 22b, and the output 1225 of the virtual buffer remaining amount calculation circuit 5f for P pictures is the virtual buffer selection. Circuit 22b
To the second input of the virtual buffer remaining amount calculation circuit 5g for B picture
provided to the third input of b. Virtual buffer selection circuit 2
The picture type 1215 is input to the fourth input of 2b from the input terminal 101m.

Output 122 of virtual buffer selection circuit 22b
7 is given to the quantization step parameter calculation circuit 6e, and the reference quantization step parameter 1228 which is the output of the quantization step parameter calculation circuit 6e is given to the first input of the adaptive quantization circuit 7e. The adaptive quantization coefficient 1229 is applied to the second input of the adaptive quantization circuit 7e from the input terminal 101n. The quantization step parameter 1230, which is the output of the adaptive quantization circuit 7e, is output from the output terminal 2e.

FIG. 7 is a conceptual diagram of the refresh system, which is one of the prominent properties of the hybrid coding system.

Next, the operation will be described. As one of high-efficiency coding schemes for coding a video signal, there is a hybrid coding scheme in which inter-picture predictive coding using motion compensation prediction and intra-picture transform coding are combined. This conventional example also employs the above hybrid coding method. Figure 5
Shows a schematic block diagram of the hybrid coding scheme. The digitized input signal is subjected to DCT in the spatial axis direction by taking a difference between images using motion compensation prediction to reduce redundancy in the time axis direction. The transformed coefficients are quantized, variable-length coded, and then transmitted through a transmission buffer.

The basic concept of quantization control will be described below. Generally, in a video signal encoding system, the transmission rate is fixed, so that a method for converging the generated code amount to the transmission rate is required.

The concept of the method of converging the generated code amount to the fixed transmission rate will be shown below. In general, the transformed coefficients are
It is not always possible to represent with a finite number of digits. Representing this coefficient with a finite number of digits is called quantization, a discrete value by quantization is called a quantization level, and an interval between quantization levels is called a quantization step. Also, the difference between the coefficient before quantization and the quantization level after quantization is called quantization noise. If the quantization step is made smaller, the quantization noise will be reduced, but the generated code amount will be increased.If the quantization step is made larger, the quantization noise will be increased, but the generated code amount will be reduced. Become. Controlling the quantization step in this way is called quantization control. That is, by quantization control,
It is possible to converge the generated code amount to the transmission rate. At this time, it is an important subject to converge to a fixed rate while maintaining good image quality.

In this conventional example, a quantization step parameter is used as an index similar to the quantization step. The function of the quantization step parameter may be considered to be equivalent to the quantization step.

The following two are defined as units for performing quantization control. (1) Standard fixed rate period (2) Standard control period

The reference fixed rate period means a unit or period in which it is determined that the generated code amount is a fixed rate. For example, the sum of the generated code amounts in a certain number of screens may be a fixed rate, or the generated code amount in one screen may be a fixed rate.
If this fixed rate period is large, the tolerance of fluctuations in the generated code amount within that period increases. For example, in a moving image, the generated code amount changes for each picture as the image moves. Therefore, in order to continuously obtain stable image quality, it is effective that the fixed rate period is large.
However, a buffer with a large capacity is required in order to increase the tolerance of fluctuations in the generated code amount. A large capacity buffer means an increase in delay time in the encoding / decoding system. On the other hand, if the fixed rate period is small, the admissibility of the generated code amount becomes low. For example, if the fixed rate period is set to a fraction of the screen, it is not possible to sufficiently allow the fluctuation of the generated code amount in the screen. For example, in a moving image, there is often a case where only one part in the screen is moving. In such a case, the degree of image quality deterioration in the screen varies greatly depending on the position in the screen. there is a possibility. However, in this case, the buffer capacity may be small, which means that the delay time in the coding / decoding system is small.

The reference control period is defined as follows. Basically, the quantization control is to check the state of the generated code amount every certain period, verify whether or not it matches the desired generated code amount, and change the quantization step parameter if it does not match. That is the work. The above period is defined as the standard control period. If this reference control period is large, the code amount becomes excessively large or small, which easily causes a buffer overflow or underflow. If this reference period is short, overflow or underflow is unlikely to occur, but a high degree of accuracy is required to predict the desired generated code amount. Uncertain prediction of the desired amount of generated code causes unnecessary fluctuation of the quantization step parameter,
This causes local deterioration of image quality.

In describing the quantization control method in this conventional example, the hierarchical structure of the video signal in this conventional example will be defined, and one property of the hybrid coding related to the hierarchical structure will be described below.

The hierarchical structure of the video signal in this conventional example is outlined as follows. Sequence Consists of one or more consecutive GOPs. GOP It is composed of a plurality of consecutive pictures (screens). Picture This is one screen and is composed of multiple slices. Slice It consists of one or more macroblocks. Macro block It consists of four luminance blocks and a color difference block at the same position on the screen. Block One block is composed of 8 × 8 pixels.

In this conventional example, three types of pictures are defined. There are three types: I picture that does not perform inter-picture predictive coding but only intra-picture conversion coding, P picture that predicts from only one direction, and B picture that predicts from both directions.

Next, one of the prominent properties of the hybrid coding system will be listed below. In this conventional example, inter-picture predictive coding using motion compensation prediction is performed. this is,
Since it is basically time-domain predictive coding, it is necessary to set an initial value. Further, in order to prevent the propagation of an error that occurs accidentally after encoding, it is necessary to set the initial value at an appropriate cycle.

This periodic initial value setting operation is generally called periodic refresh. The periodic refresh in the case of the hybrid method using inter-picture predictive coding using motion compensation prediction and intra-picture transform coding is specifically intra-picture transform coding without motion-compensated prediction.

Generally, as the refresh method, there are a method of collectively performing screens and a method of performing divided screens. The method referred to in the conventional example is the former,
The refresh screen is called an I picture.

In the case of the screen division refresh method, for example, the screen is divided into several areas, and one screen is divided into one area.
The area will be refreshed one by one. FIG. 7 shows a conceptual diagram of the refresh method. In general, intra-picture transform coding has a larger amount of generated code than inter-picture predictive coding, and therefore, when performing periodic refresh in screen units, a large reference fixed rate period of about several pictures is adopted. This means that the delay time in the encoding / decoding system is large.

Next, the quantization control method in the conventional example will be described. FIG. 6 shows an example of a quantization control method in the conventional example. In this conventional example, basically, one GOP period is used as the reference fixed rate period and one macroblock period is used as the reference control period, and the quantization control is performed in the following procedure. (1) Estimate the bit amount (target bit amount) that can be used in the encoding of the next picture. (2) The reference value of the quantization step parameter for each macroblock is set based on the estimated target bit amount and the actually generated code amount. (3) The reference value of the quantization step parameter is changed according to the image feature of each macroblock, and the final quantization step parameter of each macroblock is determined.

The details will be described below. As the first procedure, the amount of bits that can be used in encoding the next picture is estimated. Introducing the concept of complexity in this work.
In this conventional example, complexity is defined for each picture type. Also, the complexity is updated every time one picture is encoded. In reality, since the coded picture has a certain picture type, one complexity will be updated each time one picture is coded.

Complexity Xi (g, f) of the I picture immediately before encoding the fth picture of the gGOP, and the complexity of the P picture
Xp (g, f) and B picture complexity Xb (g, f) are defined as follows. Xi (g, f) = Si (s, x) × Qi (s, x) Xp (g, f) = Sp (t, y) × Qp (t, y) Xb (g, f) = Sb (u , z) × Qb (u, z)

At this time, Si (s, x) is the actual generated code amount in the xth picture of the sGOP, which is the I picture that was present in the closest past, and Sp (t, y) is the closest in the past. Sb (u, z) is the actual amount of generated code in the y-th picture of the t-th GOP, which is the P-picture, and is the actual amount of generated code in the z-th picture of the u-th GOP, which is the closest B picture that existed in the past. . (At this time, one of the above three pictures should correspond to the f-1th picture of the gGOP.) Also, Qi (s, x), Qp (t, y), Qb
(u, z) is the average value of the actual quantization step parameters in the above three pictures.

Based on the complexity, the target bit amount T (g, f) of the f-th picture of the g-th GOP is estimated as follows.

[0033]

[Equation 1]

At this time, Kp: constant Kb: constant BIT_RATE: transmission rate PICTURE_RATE: number of pictures per second.

Further, R (g, f) is the total number of remaining bits assigned to the gth GOP immediately before the fth picture of the gth GOP is encoded, and is updated as follows after the encoding of each picture. To be done. R (g, f) = R (g, 1)-{S (g, 1) + S (g, 2) +… + S (g, f-
1)} Here, S (g, f-1) represents the actual amount of generated codes in the (f-1) th picture of the gth GOP.

Further, the following calculation is performed before encoding the head picture of the GOP. R (g + 1,1) = G + R (g, f + 1) G = BIT_RATE × (F / PICTURE_RATE) At this time, F is the number of pictures per GOP, and R (1, 1) = G Further, Fp (g, f) and Fb (g, f) are the number of pictures of the P picture and the B picture which remain unencoded in the current GOP immediately before encoding the fth picture of the gth GOP. Show.

Next, as the second procedure, the reference value of the quantization step parameter for each macroblock is set based on the estimated target bit amount and the actual generated code amount.

A virtual buffer is assumed in setting the reference value. The virtual buffer is a buffer that roughly stores the difference between the target bit amount and the actual generated code amount. A virtual buffer is assumed for each picture.
That is, if the picture is an I picture, the corresponding I
A virtual buffer for a picture is used if the picture is a P picture, a virtual buffer for the corresponding P picture is used, and if the picture is a B picture, a virtual buffer for the corresponding B picture is used. Occupancy rate is changed.

For example, if the picture type of the f-th picture of the g-th GOP is an I-picture, the occupation ratio di (g, g, immediately before encoding the m-th macro block of the f-th picture of the g-th GOP of the virtual buffer for I-pictures. f, m) is defined as follows. di (g, f, m) = di (g, f, 1) + B (g, f, 1) + B (g, f, 2) +… +
B (g, f, m-1)-{T (g, f) × ((m-1) / MB_cnt)} where MB_cnt is the number of macroblocks per picture, and B (g, f) , m-1) is the actual generated code amount in the m-1th macroblock of the fth picture of the gGOP. As the di (g, f, 1), the buffer occupancy ratio di (a, b, MB_cnt + 1) in the b-th picture of the a-th GOP, which is the picture of the same type that existed in the closest past, is adopted.

Also, the occupancy rate of the virtual buffer for P pictures dp (g, f, m) and the occupancy rate of the virtual buffer for B pictures db
Similarly, (g, f, m) is updated according to the picture type.

Virtual buffer occupancy ratio di (g, f, m), dp
Based on (g, f, m) and db (g, f, m), the reference value Qr (g, f, m) of the quantization step parameter of the m-th macroblock of the f-th picture of the g-th GOP is as follows. Is calculated. If (the f-th picture of the g-th GOP is an I-picture) Then Qr (g, f, m) = (di (g, f, m) / VB) × 31 If (the f-th picture of the g-th GOP is a P-picture) Then Qr (g, f, m) = (dp (g, f, m) / VB) × 31 If (the f-th picture of the g-th GOP is a B-picture) Then Qr (g, f, m) = (db (g, f , m) / VB) × 31 At this time, VB: virtual buffer capacity.

Next, as the third procedure, the above Qr (g, f,
Based on m), the final quantization step parameter for each macroblock is determined. In this conventional example, the adaptive quantization coefficient K (g, f, m) is calculated based on the activity of each macroblock.
Is calculated as follows.

[0043]

[Equation 2]

At this time, actj: activity of macroblock Pk: pixel value of original image 8 × 8 block avg_act: average value of actj of the latest encoded image.

Next, based on the above K (g, f, m), the final quantization step parameter Q for each macroblock is obtained as follows.
Calculate (g, f, m). Q (g, f, m) = Qr (g, f, m) × K (g, f, m)

[0046]

In the conventional quantization control method in the video signal coding method, the adaptive quantization coefficient is obtained based only on the activity of each macroblock, and the result of motion compensation prediction is not taken into consideration. . Also, scene change is not considered.

The present invention has been made to solve the above problems, and an object thereof is to obtain a quantization control system for performing adaptive quantization using activity, inter-picture signal power and scene change information. To do.

[0048]

A video signal coding system according to the present invention is a video signal coding system having intra-picture coding and inter-picture coding by motion compensation as constituent elements. It is configured to use the activity and the inter-image signal power as an index of conversion.

Further, the video signal coding system according to the present invention is a video signal coding system having intra-picture coding and inter-picture coding by motion compensation as its constituent elements, and the activity is used as an index of adaptive quantization. And the inter-picture signal power and scene change information are used.

Further, the video signal coding system according to the present invention is a video signal coding system having intra-picture coding and inter-picture coding by motion compensation as its constituent elements, and the rate of change is used as an index of a scene change. And the activity is used for the determination.

[0051]

The video signal coding method according to the present invention has intra-picture coding and inter-picture coding by motion compensation as constituent elements, and estimates the complexity or generated code amount of a coded picture for each divided area. Therefore, even when performing periodic refresh with divided screens, good image quality is maintained and rate stability is achieved.By using activity and inter-picture signal power as indicators of adaptive quantization, motion compensation can be achieved. A quantization control method that can cope with a malfunction and a local change of an image and that also stabilizes the generated code amount is realized.

Further, the video signal coding method according to the present invention has intra-picture coding and inter-picture coding by motion compensation as constituent elements, and estimates the complexity or generated code amount of the coded picture for each divided area. By doing so, good image quality is maintained and rate stability is achieved even when periodic refreshing is performed with the screen divided, and activity, inter-picture signal power, and scene change information are used as indicators of adaptive quantization. As a result, it is possible to realize a quantization control method capable of coping with a malfunction of motion compensation, coping with a local change of an image, and stabilizing a generated code amount.

Further, the video signal coding method according to the present invention has intra-picture coding and inter-picture coding by motion compensation as constituent elements, and estimates the complexity or generated code amount of the coded picture for each divided area. By doing so, good image quality can be maintained and rate stability can be achieved even when periodical refreshing with divided screens is performed.By using the change rate and activity as indicators for scene change determination, the generated code amount can be Realize a stable quantization control method.

[0054]

【Example】

Example 1. The first embodiment of the present invention will be described below.

FIG. 1 is a diagram showing a main hierarchical structure of a video signal in the first embodiment of the present invention.

FIG. 2 is a schematic block diagram showing an example of the quantization control system in the first embodiment of the present invention.

In FIG. 2, the average inter-image signal power 201 input from the input terminal 1a is the first input of the target complexity calculation circuit 3a, and the average intra-image signal power 202 input from the input terminal 1b is The reference quantization step parameter 203 input from the input terminal 1c to the second input of the target complexity calculation circuit 3a is the type identification input from the input terminal 1d to the third input of the target complexity calculation circuit 3a. The signal 204 is supplied to the fourth input of the target complexity calculation circuit 3a, and the generated code amount 205 independent of quantization input from the input terminal 1e is supplied to the fifth input of the target complexity calculation circuit 3a.

Output 20 of target complexity calculation circuit 3a
Reference numeral 6 denotes the first bit of the target bit amount calculation circuit 4a, and the remaining buffer amount 207 input from the input terminal 1f.
Is given to the second input of the target bit amount calculation circuit 4a. The target bit amount 208 output from the target bit amount calculation circuit 4a is the first input of the virtual buffer remaining amount calculation circuit 5a, and the generated code amount 209 input from the input terminal 1g is the virtual buffer remaining amount calculation circuit 5a. Given to the second input of. Virtual buffer remaining amount calculation circuit 5
The output 210 of a is provided to the input of the quantization step parameter calculation circuit 6a.

The reference quantization step parameter 211, which is the output of the quantization step parameter calculation circuit 6a, is given to the first input of the adaptive quantization circuit 7a. The activity 212 is applied to the second input of the adaptive quantization circuit 7a from the input terminal 1h. The latest average inter-picture signal power 213 input from the input terminal 1i is supplied to the third input of the adaptive quantization circuit 7a. The quantization step parameter 214 for each macroblock, which is the output of the adaptive quantization circuit 7a, is output from the output terminal 2a.

FIG. 3 is a diagram showing a schematic relationship between the complexity referred to in the quantization control method and the generated code amount in the first embodiment of the present invention.

The operation will be described below. In recent years, the development of high-efficiency coding apparatus for video signals has been shifting to the development of high-resolution systems including HDTV signals. At this time, the high-resolution system inevitably requires high-speed processing. Generally, by adopting parallel processing, high speed performance is supplemented.

FIG. 1 is a diagram showing a main hierarchical structure of a video signal in the first embodiment. FIG. 1B is the same as FIG.
It is a figure in which the picture in (a) is vertically divided into S pieces. At this time, each of the S divided areas is called a sub-picture, and each sub-picture becomes a unit for performing parallel processing. It should be noted that the divided area of the sub-picture is not limited to this example. FIG. 1C is a diagram in which each sub-picture is further divided into L pieces. At this time, each of the L divided areas in each sub-picture is called a GOB. It should be noted that the GOB division area does not follow only this example. Further, as shown in FIG. 1D, GOBs existing at the same position in each sub-picture are collectively called a slice.
As in the conventional example, there are macroblocks and blocks as lower layers of GOB.

Even in the case of parallel processing, in order to prevent the image quality from deteriorating depending on the position on the screen, the quantization control needs to handle all sub-pictures collectively, that is, as a picture. At this time, basically, when GOBs existing at the same position in each sub-picture are processed at the same time, the quantization control is processed in slice units. It should be noted that the slice is for the purpose of collectively processing the regions processed at the same time, and the divided regions do not follow only this example.

FIG. 2 shows an example of the quantization control method in the first embodiment of the present invention. This embodiment basically adopts one picture period as the reference fixed rate period and one slice period as the reference control period, and the quantization control is performed by the following procedure based on the main hierarchical structure. . (1) Estimate the bit amount (target bit amount) that can be used for encoding each slice of the next picture. (2) The reference value of the quantization step parameter for each macroblock is set based on the estimated target bit amount and the actually generated code amount. (3) The reference value of the quantization step parameter is changed according to the image feature of each macroblock, and the final quantization step parameter of each macroblock is determined.

In the present embodiment, the case where parallel processing is performed, or the case where a screen is divided and periodic refresh is performed is also considered. Therefore, in this embodiment, GOB will be described as a basic unit.

The details will be described below. As the first procedure, the amount of bits that can be used for encoding each slice of the next picture is estimated. For this work, in this embodiment,
For each GOB, a pattern target complexity and a motion target complexity are defined. The pattern target complexity is
The target complexity is defined based on the intra-image signal power, and the motion target complexity is the complexity defined based on the inter-image signal power.

The image signal power Eint is defined as follows, for example, in macroblock units. The macroblock size is P × Q, and (p,
When the signal value of the q) component is V (p, q)

[0068]

[Equation 3]

It is assumed that Alternatively,

[0070]

[Equation 4]

It is assumed that The in-image signal power Eint does not have to comply with the above definition as long as it is an index indicating the complexity of the pattern of the input image. For example, the absolute value sum or the absolute value sum of squares of the conversion coefficients excluding the DC component after the input signal is subjected to the conversion coding and the specific quantization is performed for each conversion coefficient may be used.

The inter-image signal power Emc is defined as follows, for example, in macroblock units. The macroblock size is P × Q, and (p,
When the signal value of the q) component is V (p, q) and the signal value of the (p, q) component of the motion compensation prediction block of the macroblock is R (p, q),

[0073]

[Equation 5]

It is assumed that Alternatively,

[0075]

[Equation 6]

It is assumed that The inter-image signal power Emc does not have to comply with the above definition as long as it is an index representing the complexity of the movement of the input image. For example, the sum of absolute values or the sum of absolute values squared after transform-encoding the signal after motion compensation prediction and performing specific quantization for each transform coefficient may be used.

The picture target complexity Yint (f, h, s) and motion target complexity Ymc (f, h, s) of the sth subpicture of the hth slice of the fth picture are the function fint and the function fm.
It is defined as follows using c. Yint (f, h, s) = fint (GOBaverage_Eint (f-1, h, s), Q
(f, h, s)) Ymc (f, h, s) = fmc (GOBaverage_Emc (f-1, h, s), Q
(f, h, s), Sv (f-1, h, s)) At this time, GOBaverage_Eint (f-1, h, s) and GOBavera
ge_Emc (f-1, h, s) is the average Eint and the average Emc of the s-th sub-picture of the h-th slice of the (f-1) -th picture, and Q
(f, h, s) is the average value of the quantization step parameters actually used in the s-th sub-picture of the h-th slice of the f-th picture, and Sv (f-1, h, s) is the This is a code amount that does not depend on quantization among the code amounts actually generated in the s-th sub-picture of the h-th slice of the f-1 picture. Sv (f, h, s)
Indicates the code amount of information that is not quantized, such as the address of a macroblock, the quantization step parameter, and the information of a motion vector in motion compensation. In addition, Sv
As the approximate value of (f, h, s), only the code amount of the motion vector in motion compensation may be used.

In the above definition, the functions fint and fmc
Is defined as follows. FIG. 3A illustrates GOBaverage_of the sth subpicture of the hth slice of the fth picture.
It is a figure which shows the schematic relationship of Eint (f, h, s) and the generated code amount Sint (f, h, s) when the said GOB is intra-coded. As is clear from the figure, GOBaverage_Eint (f, h, s) and Q (f, h, s
It is possible to estimate Sint (f, h, s) from (s). This estimated generated code amount is the pattern target complexity Yint.

3B and 3C show the GOB average of the s-th sub-picture of the h-th slice of the f-th picture.
FIG. 9 is a diagram showing a schematic relationship between _Emc (f, h, s) and the generated code amount Smc (f, h, s) when the GOB is inter-coded. Figure 3
(B) shows a code amount Sv (f, h, s) of the generated code amount Smc (f, h, s) that does not depend on the quantization step parameter Q (f, h, s) such as the code amount of the motion vector in motion compensation. , s) is small, and FIG. 3C shows the generated code amount Smc (f,
It shows a case where the code amount Sv (f, h, s) that does not depend on the quantization occupies a large proportion of the h, s). Therefore, Smc (f, h, s)
Can be estimated from GOBaverage_Emc (f, h, s), Q (f, h, s) and Sv (f, h, s), and the estimated generated code amount is the motion target complexity Ymc. The functions fint and fmc are defined based on the above quantitative property.

By the way, in the present embodiment, both the screen batch refresh and the screen division refresh are assumed as the periodic refresh. Further, it is assumed that the screen division refresh area is a slice unit. That is, it may be considered that there are cases where intra-picture coding is performed in slice units. Based on the above, the target complexity Y (f, h, s) of the s-th phase of the h-th slice of the f-th picture is the GOB
Is defined as follows depending on whether or not the slice to which is belongs to is an intra-coded slice. If (The slice type to which the GOB belongs is an intra-frame coding slice) Then Y (f, h, s) = Yint (f, h, s) Else Y (f, h, s) = Ymc (f, h , s)

The target complexity of each slice is defined based on the target complexity of each GOB. The f
The target complexity Y (f, h) of the h-th slice of the frame is
It is defined as follows. Y (f, h) = Y (f, h, 1) + Y (f, h, 2) + ... + Y (f, h, S) (S; number of subpictures)

Next, the bit amount (target bit amount) usable in each slice of the next picture is estimated based on the target complexity of each slice and the remaining buffer capacity. Target bit amount of each slice of f-th picture
T (f, h) (1 ≦ h ≦ L) is assumed as follows after the encoding of the f−1th picture is completed. T (f, h) = [Y (f, h) / {Y (f, 1) + Y (f, 2) +… + Y (f, L)}
] × [BIT_RATE / PICTURE_RATE-{BE (f, 1)-BIT_
RATE × TE}] However, in the process of obtaining Y (f, h, s), Q (f-1, L, s) is used instead of Q (f, h, s). This is because the value of Q (f, h, s) is not yet determined at the time when the target bit amount of each slice of the f-th picture is obtained, that is, immediately before the f-th picture is encoded.

In the above equation, BIT_RATE: transmission rate PICTURE_RATE: number of pictures per second TE: reference accumulation time constant, BE (f, h) is immediately before encoding the h-th slice of the f-th picture Indicates the remaining amount of the transmission buffer.

If all Y (f, h) cannot be defined in the initial state immediately after the start of encoding, T (f, h) is defined as follows. T (f, h) = [1 / L] × [BIT_RATE / PICTURE_RATE-{B
E (f, 1)-BIT_RATE x TE}]

BE (f, 1) in the initial state is defined as follows, for example. BE (f, h) = BIT_RATE / PICTURE_RATE

When BE (f, 1) cannot be used due to the delay due to the hardware configuration, for example, the latest available buffer remaining amount is used.

Next, as the second procedure, the reference value of the quantization step parameter for each macroblock is set based on the estimated target bit amount and the actually generated code amount. A virtual buffer is assumed when setting the reference value. The virtual buffer is a buffer that roughly stores the difference between the target bit amount and the actual generated code amount.

In this embodiment, one virtual buffer is assumed. The occupancy rate d (f, h) of the virtual buffer immediately before encoding the h-th slice of the f-th frame is defined as follows. At this time, S (f-1, h) is the actual generated code amount in the h-th slice of the f-th picture.

The quantization step parameter is changed by associating the occupation rate of the virtual buffer with the quantization step parameter. For example, if there is no difference between the target bit amount set for each slice and the actual generation amount, the virtual buffer occupancy does not change and the quantization step parameter does not change. If there is a difference between the above two,
The virtual buffer occupancy changes, and the quantization step parameter is modified according to the change. For example, when the generated code amount exceeds the target bit amount,
The difference is the increase amount of the virtual buffer occupation rate. The virtual buffer occupancy corresponds to the quantization step parameter, and an increase in the virtual buffer occupancy leads to an increase in the quantization step parameter. Increasing the quantization step parameter will reduce the subsequent generated code amount.

Based on the virtual buffer occupancy ratio d (f, h),
The reference value Qr (f, h) of the quantization step parameter of the h-th slice of the f-th picture is calculated as follows using the function fq. Qr (f, h) = fq (d (f, h)) Function fq is, for example, Qr (f, h) = 31 × {d (f, h) / r} where r = BIT_R
ATE / PICTURE_RATE or Qr (f, h) = 31 ^{d (f, h) / r where} r = BIT_R
ATE / PICTURE_RATE is possible. Alternatively, it is determined based on the generated code amount and the quantitative property of the quantization step parameter.

Next, as the third procedure, the final quantization step parameter for each macroblock is determined based on Qr (f, h). Therefore, the adaptive quantization coefficient K (f, h, m) of the m-th macroblock of the h-th slice of the f-th picture is calculated. K (f, h, m) = fk (GOBaverage_Emc (f-1, h, s), MBaverage_
act (f, h, m)) At this time, GOBaverage_Emc (f-1, h, s) is the average Emc of the s-th sub-picture of the h-th slice of the f-1st picture, and MBaverage_act (f, h, m) ) Is the m-th macroblock average activity of the h-th slice of the f-th picture.
Function fk Determined based on the above quantitative properties.

GOBaverage_ as a parameter of K (f, h, m)
By using Emc (f-1, h, s), it is possible to take the quantizer finer when the motion compensation does not work properly and takes a large value. .

Using K (f, h, m), the final quantization step parameter Q (f, h, m) for each macroblock is as follows.
m) is calculated. Q (f, h, m) = Qr (f, h, m) × K (f, h, m)

Example 2. The second embodiment of the present invention will be described below. FIG. 4 is a schematic block diagram showing an example of a quantization control system in the third embodiment of the present invention.

The second embodiment differs from the first embodiment in the process of calculating the adaptive quantization coefficient. In FIG. 4, the input terminal 1
The average inter-image signal power 301 input from j is input to the first input of the target complexity calculation circuit 3b at the input terminal 1
The average intra-image signal power 302 input from k is input to the second input of the target complexity calculation circuit 3b at the input terminal 1
Reference quantization step parameter 303 input from l
Is the type identification signal 304 input from the input terminal 1m to the third input of the target complexity calculation circuit 3b, and the quantization input from the input terminal 1n to the fourth input of the target complexity calculation circuit 3b. Generated code amount 30 independent of
5 is given to the fifth input of the target complexity calculation circuit 3b.

Output 30 of target complexity calculation circuit 3b
Reference numeral 6 indicates the remaining buffer capacity 307 input from the input terminal 1o to the first input of the target bit amount calculation circuit 4b.
Is supplied to the second input of the target bit amount calculation circuit 4b. The target bit amount 308 which is the output of the target bit amount calculation circuit 4b is the first input of the virtual buffer remaining amount calculation circuit 5b, and the generated code amount 309 input from the input terminal 1p is the virtual buffer remaining amount calculation circuit 5b. Given to the second input of. Virtual buffer remaining amount calculation circuit 5
The output 310 of b is given to the input of the quantization step parameter calculation circuit 6b.

The reference quantization step parameter 311 which is the output of the quantization step parameter calculation circuit 6b is given to the first input of the adaptive quantization circuit 7b. The activity 312 is applied to the second input of the adaptive quantization circuit 7b from the input terminal 1q. The latest average inter-image signal power 313 input from the input terminal 1r is supplied to the third input of the adaptive quantization circuit 7b. The average activity 314 of the image input from the input terminal 1s is input to the first input of the scene change determination circuit 8a at the input terminal 1t.
The change rate 315 of the image input from is supplied to the second input of the scene change determination circuit 8a. Scene change determination circuit 316 outputs the scene change determination circuit 316
Is given to the fourth input of the adaptive quantization circuit 7b. The quantization step parameter 317 for each macroblock, which is the output of the adaptive quantization circuit 7b, is output from the output terminal 2b.

The operation will be described below. The second embodiment differs from the first embodiment in the process of calculating the adaptive quantized coefficient. In the first embodiment, the adaptive quantization coefficient is calculated using the activity and the inter-image signal power. In this case, an appropriate coefficient may not always be set for an image in which the scene changes. Based on this, scene change determination is added and used to set the adaptive quantization coefficient.

Scene change determination Sce of f-th picture
neChange (f) is defined as follows. if (fsc (change_rate (f), PICTaverage_act (f))> Ksc) SceneChange (f) = 1 (with scene change) else SceneChange (f) = 0 (without scene change) At this time, change_rate (f) is the fth Picture change rate (sum of absolute differences between frames), PICTaverage_a
ct (f) is the average activity of the f-th picture, and Ksc
Is a threshold for making a scene change determination.
The function fsc is determined based on the above quantitative property.

Next, the formula for calculating the adaptive quantized coefficient K (f, h, m) of the mth macroblock of the hth slice of the fth picture is defined as follows. K (f, h, m) = fk (GOBaverage_Emc (f-1, h, s), MBaverage_a
ct (f, h, m), SceneChange (f)) At this time, GOBaverage_Emc (f-1, h, s) is the average Emc of the sth sub-picture of the h-th slice of the f-1 picture, and MBaverage_act ( f, h, m) is the average activity of the mth macroblock in the hth slice of the fth picture, and SceneChange (f) is the scene change information of the fth picture. The function fk is determined based on the above quantitative property.

SceneChange as a parameter of K (f, h, m)
By using (f), GOBaverage_Emc (f-1, h, s)
Even if takes a large value, it is possible to prevent a motion compensation malfunction from occurring in a picture in which a scene change has occurred.

[0102]

As described above, according to the present invention, by calculating the adaptive quantization coefficient using the intra-picture signal power and the inter-picture signal power, it is possible to cope with the malfunction of the motion compensation.
There is an effect of obtaining a quantization control method capable of coping with local changes in the image and stabilizing the generated code amount.

Further, according to the present invention, by calculating the adaptive quantization coefficient using the intra-image signal power, the inter-image signal power, and the scene change information, it is possible to deal with the malfunction of the motion compensation, and the local of the image There is an effect of obtaining a quantization control method that can cope with changes and stabilize the generated code amount.

Further, according to the present invention, by judging the scene change by using the change rate and the activity, it is possible to deal with various images and obtain the quantization control method in which the generated code amount is stable.

[Brief description of drawings]

FIG. 1 is a diagram showing a main hierarchical structure of a video signal in a first embodiment of the present invention.

FIG. 2 is a schematic block diagram showing an example of a quantization control method in the first embodiment.

FIG. 3 is a diagram showing a schematic relationship between complexity and a generated code amount, which is referred to in the quantization control method in the first embodiment.

FIG. 4 is a schematic block diagram showing an example of a quantization control system according to a second embodiment of the present invention.

FIG. 5 is a schematic block diagram showing a conventional video signal encoding system.

FIG. 6 is a schematic block diagram showing an example of a quantization control method in a conventional video signal encoding method.

FIG. 7 is a conceptual diagram of a refresh method.

[Explanation of symbols]

3 target complexity calculation circuit, 4 target bit amount calculation circuit, 5 virtual buffer remaining amount calculation circuit, 6 quantization step parameter calculation circuit, 7 adaptive quantization circuit,
8 Scene change judgment circuit.

Front page continuation (72) Inventor Takashi Shinohara No. 1 Baba Institute, Nagaokakyo City Video System Development Laboratory (72) Inventor Takahiro Nakai No. 1 Baba Institute, Nagaokakyo City Video System Development, Mitsubishi Electric Corporation In the laboratory

Claims (3)

[Claims] 1. A video signal coding system for coding a video signal, the video signal coding system having intra-picture coding and inter-picture coding as constituent elements, wherein an activity is provided for each divided area. A video signal coding method characterized in that adaptive quantization is performed using inter-picture signal power as a parameter. 2. A video signal coding system for coding a video signal, the video signal coding system comprising intra-picture coding and inter-picture coding as constituent elements, comprising means for judging a scene change. , Activity for each divided area,
A video signal coding method characterized in that adaptive quantization is performed using inter-image signal power and scene change determination as parameters. 3. A video signal coding system for coding a video signal, the video signal coding system having intra-picture coding and inter-picture coding as constituent elements, wherein a scene is defined with a change rate and activity as parameters. A video signal encoding method characterized by determining a change.