JPH11196422A - Encode method of image signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress screen deterioration in a scene change by judging the scene change, allocating a bit amount equal to that of an intra-picture with respect to the screen of the scene change, dividing the increased amount of generated bits through geometrical progression and correcting a lot of screens. SOLUTION: The scene change is detected by using an image error value from a motion estimating part 2. A rate control part 31 discriminates the scene change from a scene change flag outputted from a scene change detecting part 30 and determines bit distribution to each screen. Namely, when the scene change flag from the scene change detecting part 30 is turned on, the rate control part 31 corrects the bit distribution to that screen. Consequently, the distribution of bits equal to these of the intra-picture is set and further, the correction of increased amount by that distribution is performed on a lot of images after the scene change. Then, the calculation of a correction value is performed by the geometrical progression.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an encoder for an image signal, particularly to an encoding technique applicable to MPEG.

[0002]

2. Description of the Related Art An encoding method of a conventional MPEG encoder will be described with reference to FIG. 1. A picture rearranging unit 1 changes the encoding order of an input original image,
The motion estimating unit 2 estimates the motion of the image in the time direction, and outputs a motion vector detected in the determined encoding mode. The difference calculation unit 3 outputs the “image before compression” which is the output of the motion estimation unit 2.
And a difference between the compressed image and the motion-predicted compressed image. The DCT unit 4 performs DCT on the difference output, and the quantization unit 5 performs DCT on the DCT.
The output is quantized using the bit rate control value calculated from the utilization rate of the buffer 10. The variable length coding unit 6 converts the quantized value into a variable length code. The inverse quantization unit 7 performs inverse quantization on the quantized value using the above-mentioned bit rate control value, and the inverse DCT is performed by the IDCT unit 8. The multiplexing unit 9 multiplexes the output from the variable length coding 6 and the motion estimating unit 2 and sends the multiplexed output to the buffer 10, which stores the input,
The data is output as a bit stream. The addition unit 11 adds the output of the IDCT unit 8 and the output of the motion prediction unit 13 and outputs the result. The memory 12 stores the compressed image. The motion prediction unit 13 predicts and outputs an optimal motion using a compressed image, a motion vector, and a coding mode.

[0003] However, the conventional MPEG encoding method has a problem that image quality deterioration is conspicuous with respect to an image scene change. That is, the bit rate control is performed by GOP (Gr.
oupof picture), for example 1
This is based on a method in which five images are grouped into one group, and feedback is performed so that the amount of a bit stream generated by compressing 15 images is kept as constant as possible. For example, if the bit rate is 4 Mbit / sec and the input image rate is 30 images / sec, the bit amount allocated to 15 images is

## EQU1 ## (15 sheets / 30 sheets / sec) × 4 Mbit / sec = 2 Mbit (1) A GOP is composed of one Intra picture and 14 non-Intra pictures.
The difference information of the 14 pictures is compressed on the assumption that one intra picture is similar to the "picture". Here, an intra picture is a picture that compresses the image itself, and a non-intra picture is a picture that calculates a difference if there is a correlation between pictures, and the difference is also a picture to be compressed.

[0004] In the prior art, as described above, 2 Mbits are set to 1 on the premise that the picture in the GOP is similar to the "picture".
Distribute to 5 cards. At this time, a relatively large amount, for example, 0.5 Mbit is allocated to the intra picture, and the remaining 1.5 Mbit is allocated to other pictures, and the compression ratio is feedback-controlled (rate controlled). However, in this method, if the "picture" is changed in the GOP due to a scene change or the like, the assumption is broken, and the image quality is noticeably deteriorated. The reason will be described with reference to FIG. 1 GOP
When the rate control is performed with the total bit amount of occurrence of 2M bits as a target, the distribution of each screen is shown in FIG.
Decide as in (a). Here, the first screen is an intra picture, and the allocation amount is large (0.5 Mbit).
When feedback is applied so as to achieve the distribution shown in (a), the total number of bits generated in the GOP becomes (b)
It becomes as shown in. However, if a scene change occurs on the fifth screen as shown in (c), there is no correlation between the first to fourth images on the fifth screen and the same as the first screen. In order to obtain the image quality of, the number of bits to be distributed needs to be the same as that of the first image. However, since the distribution is set to the value determined in (a), the fifth screen becomes a “highly compressed picture” and the deterioration is conspicuous. If the fifth image is degraded, the next sixth to fifteenth images are also degraded.
Since the difference compression processing is performed on the basis of the sheet, the image is deteriorated. This causes the image quality deterioration to be dragged to the next GOP, resulting in long-term deterioration. The control is performed by feedback so as to achieve the distribution shown in FIG. 2A, and the feedback parameter at this time is calculated based on the value determined in FIG. 2A and the usage rate of the buffer 10 in FIG.

[0005]

SUMMARY OF THE INVENTION The MPE of the present invention
The G encoder detects a scene change screen and sets the same bit allocation as that of an intra picture for the screen. Further, the correction for increasing the distribution is performed on a large number of images after the scene change, and the correction values are calculated by geometric progression. In addition, the detection of a scene change uses the output result of the motion correction portion. Further, the determination of the scene change is performed based on the power of the error image.

[0006]

An embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of an MPEG encoder according to one embodiment.
And the rate control unit 31 is provided. That is, a scene change is detected using the image error value from the motion estimating unit 2. The rate control unit 31 determines a scene change based on a scene change flag output from the scene change detection unit 30, and determines a bit distribution for each screen.
Here, the image error value is a “pre-compression image” (output of the motion estimation unit 2) to be compressed at the current time,
This is the difference from the image obtained as a result of the motion vector correction of the “image before compression” of the output of the motion estimator 2 used when encoding the output compressed image.

[0007] The scene change detection unit 30 calculates an image error for one image, determines that the image error is a scene change image when the value is larger than a predetermined value, and sets a scene change flag. In this embodiment, a scene change is determined when the average error of an 8-bit image per pixel is 6 or more.

Next, the rate control section 31 will be described.
When there is no scene change (when the scene change flag from the scene change detection unit 30 in FIG. 1 is off), the rate control unit 31 allocates bits to each screen as shown in FIG. The method of distribution is shown below.

The buffer 10 shown in FIG. 1 is a 16 Mbit memory, and the bit rate control value is controlled so that 8 Mbits are used when the usage rate of this buffer is checked in a cycle of GOP. First, the total bit usage of the GOP is determined. The total bit usage is determined by the value obtained by the above equation 1 and the usage of the buffer 10. For example:-Bit rate = 4 Mbit / sec-Picture rate-> 30 images / sec-1 GOP-> 15 images / GOP-Bits used in buffer 10 If 0c = 7.8 Mbits, the total bit usage (W) of the GOP is

## EQU2 ## W = 4M / 30 sheets × 15 sheets− (0c−8M) = 4M / 30 × 15− (7.8M−8M) = 2.2 Mbit (2)

Next, the distribution of an intra picture which is the first screen in the GOP is obtained. Now, if the distribution of intra pictures is １ ／ of the entire GOP, 2.2M × １ ／ =
It becomes 2.2 / 5 Mbit.

The encoding process of the intra picture is as follows.
Bit rate control is performed so as to be 2.2 / 5 Mbit. After the encoding processing of the intra picture, the bit amount actually generated with respect to the allocation bits of 2.2 / 5 Mbit can be determined.

[0012] This bit amount is expressed as REAL _I = 2.2 / 5M
+ 0.1M, the remaining bit amount L in the current GOP
AST ₍₂₎

(Equation 3)

Next, allocation is determined for the picture following the intra picture. Here, if the distribution is made evenly over the remaining 14 screens, the next distribution T ₍₂₎ is

(Equation 4) Becomes

Similarly, the encoding process is performed on this distribution, and if the actually generated bit amount is REAL ₍₂₎ , the remaining bit amount LAST ₍₃₎ of the GOP becomes

(Equation 5) Becomes In the same way, the remainder is equally distributed to the next screen, so the next distribution T ₍₃₎ is

(Equation 6) Becomes

As described above, the allocation is determined for each of the fifteen screens and the encoding is performed. Then, the above is repeated for the next GOP.

Next, the bit distribution when a scene change occurs will be described.

First, the total bit usage (W) of the GOP
Is determined. W is similarly obtained by using the above-described equation (2). If the scene change flag from the scene change detection unit 30 in FIG. 1 is off, the same processing as described above is performed for distribution of each screen. If the scene change flag is turned on, the bit distribution for the screen is corrected.

The case where the scene change flag is turned on two screens after the intra picture in the GOP will be described.

First, the remaining bit amount of the GOP is calculated by the following equation (5).
Is given as LAST ₍₃₎ . Here, the distribution T ₍₃₎ is corrected as follows

(Equation 7) Equation (7) expresses that “the screen with the scene change flag on is allocated the same amount as the intra picture”, and indicates that this screen is a screen with little deterioration of the intra picture level. .

Assuming that the bit amount actually generated by the encoding processing of this screen is REAL ₍₃₎ , the remaining bit amount LAST ₍₄₎ of the GOP is

(Equation 8) Becomes

Further, the distribution T ₍₄₎ for the next screen is corrected as follows. Here, COMP is an initial value of the correction amount.

(Equation 9)

(Equation 10) Here, γ is a correction parameter and in this embodiment, γ = 0.97
And

Assuming that the bit amount actually generated by the encoding process of this scene is REAL ₍₄₎ , the remaining bit amount LAST ₍₅₎ of the GOP is

[Equation 11] Becomes

Further, the distribution T ₍₅₎ for the next screen is

(Equation 12) And in the same way

(Equation 13)

[Equation 14]

(Equation 15)

(Equation 16) Are sequentially set.

As described above, in this embodiment, the correction amount of the distribution is set by the equation of γ ^x · COMP (x is an integer), and γ
^The correction process is continuously performed until ^x · COMP = 0. That is, as long as γx · COMP is not “zero”, correction is performed regardless of the GOP unit.

According to this method, sufficient bits can be allocated to a scene-changed screen, and the increased amount is apparently corrected by reducing the allocation amount of another screen.
However, since a large number of screens are used to reduce the distribution amount, and a small bit amount is subtracted from each screen, deterioration is not visually noticeable.

When the next scene change occurs before γ ^x · COMP is not zero, γ ^x · CO
The optimum correction value is determined by adding the value of the following equation to the value of MP.

[Equation 17] (N is the order of encoding in the GOP) If the contents described above are approximated and described again, the amount ADD obtained by increasing the bit allocation due to the occurrence of a scene change is expressed by Expression (18).

(Equation 18) This equation (18) is converted into a geometric progression, and the right side thereof is made to correspond to the distribution bit amount of each screen for each item.

That is, the amount obtained by subtracting each term on the right side from the “distribution bit amount of each screen when it is assumed that no scene change has occurred” is set as the actual distribution bit amount.

Here, by transforming the equation (18) into a geometric progression, it is possible to use a single memory for storing the value of the COMP for correction in the actual system construction (design), thereby reducing cost. Is also excellent.

Next, the correction of the increased ADD will be described. As for LAST (n) (n is arbitrary), LAST ′ (n) when there is no scene change and LAST ″ (n) when a scene change occurs and the present invention is applied. LAST '''corrected for (n)
For (n), LAST '(n) and LAST ""
The difference from (n) is defined as one parameter. Assuming that the difference is ΔL (n), ΔL (n) is expressed by equation (19).

[Equation 19] Here, LAST (n) ″ is the same as LAST (n) ′ if there is no scene change.

If a scene change occurs at m = 0 and bit addition processing for ADD shown in equation (18) is performed, ΔL (m) (m ≧ 0) becomes

(Equation 20) Indicated by here

[Outside 1] Is equivalent to the denominator of the equations shown on the right side of Equations (12) to (16).

Also,

[Outside 2] Is equivalent to the second term on the right side of Expressions (12) to (16).

Formula (20) will be described in detail. m is a screen number. When m = 0, a scene change occurs, and when m = 1, it is a number indicating the next screen.

(Equation 21)

(Equation 22)

(Equation 23)

ΔL (1) in the equation (21) indicates that the difference of ADD = LAST (1) ′ − LAST (1) ″ is corrected by γ · ADD and becomes ADD−γ · ADD due to a scene change. ing. Here, a value obtained by dividing ΔL (1) by N (1) is an image distribution value of m = 1.

Equation (22) is different from equation (21) in that γ · AD
Since D becomes γ ² · ADD and the distribution value of m = 1 is reflected, ADD− △ L (1) / N (1) −γ ² · ADD
Becomes

Similarly, since equation (23) reflects the distribution values of m = 1 and m = 2, ADD− (△ L (1) / N (1) +
ΔL (2) / N (2)) − γ ³ · ADD. Therefore, if m is arbitrary, Expression (20) is obtained.

By transforming equation (20), equation (24) is obtained.
Further, the equation (25) is obtained by setting m to m-1.

(Equation 24)

(Equation 25) From Equation (25) -Equation (24)

(Equation 26)

Now, if m → ∞, equations (24) and (2)
6) are expressed by Expressions (27) and (28), respectively.

[Equation 27]

[Equation 28] Here, 0 ≦ γ <1. From equation (28), for m → ∞

(Equation 29) And the values in parentheses are (1-1 / N (m)) <1

[Equation 30] And the equation (31) is obtained.

(Equation 31)

Equations (30) and (31) mean the following. * According to equation (30), the correction in the present invention is in the same state after a long time as when no scene change has occurred. * From equation (31), the amount to be corrected is equal to ADD.

From equation (29), ΔL (m) is ΔL
(1) = ADD-γ · ADD is always smaller, and there is no image quality deterioration due to a scene change.

As described above, since the value to be corrected is calculated in the form of a geometric progression, it is possible to use a single memory for storing the value of the COMP for correction in the actual system construction (design). Excellent in terms of cost.

Although the present invention has been described with reference to the embodiments, the present invention is not limited thereto.

[0042]

The MPEG encoder of the present invention determines a scene change by using the result of the motion estimation circuit, and allocates a bit amount equivalent to that of an intra picture to a scene change screen, and allocates the allocated bit amount. The increase in the number of generated bits is divided by a geometric progression and corrected on a large number of screens, whereby image quality deterioration due to a scene change can be suppressed.

[Brief description of the drawings]

FIG. 1 is a block diagram of an MPEG encoder according to one embodiment of the present invention.

FIG. 2 is a view for explaining conventional bit allocation for each screen.

[Explanation of symbols]

 2 motion estimation unit 30 scene change detection unit 31 rate control unit

Claims (2)

[Claims] 1. An image of a scene change is detected based on an image error value, a bit allocation equivalent to an intra picture is set for the image of the scene change, and an increase in bits caused by the setting of the bit allocation is performed. A method for encoding an image signal, comprising a step of dividing a minute and assigning the divided image to a plurality of images after a scene change. 2. The image signal encoding method according to claim 1, wherein division of the bit increment is performed by a geometric progression.