JPH0541860A - Picture compression coder - Google Patents

Abstract

PURPOSE:To realize the picture compression coder whose picture quality and bit rate are stable. CONSTITUTION:Upon the receipt of an input picture Si, a statistic quantity arithmetic operation means 21 obtains a statistic quantity S23 to give it to a code quantity prediction means 23. A buffer occupancy quantity detection means 22 detects an occupied quantity S14 of a buffer 14 and gives buffer information S22 to the code quantity prediction means 23. The code quantity prediction means 23 predicts a quantization step width suitable for the property of each input picture and gives the result to a quantization means 12. The quantization means 12 quantizes a picture data S11 whose redundancy is eliminated by the information quantity compression means 11 based on the quantization step width from the code quantity prediction means 23. The quantized output data S12 is coded by a coding means 13 and outputted externally via the buffer 14.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compression / encoding apparatus for compressing and encoding a moving image or the like, and more particularly to an image quality and bit rate control system.

[0002]

2. Description of the Related Art Conventionally, as a technique in such a field,
For example, some documents were described in the following documents.

IEEE TRANSACTIONS ON Communications
COMMUNICATIONS), COM-32 [3] (1984-)
3) (US) Wen-Hsiung CHEN and William K.PRATT “Scene Adaptive Code (Scene Adaptive Code)
r) "P.225-232 Image compression, especially in real-time compression of moving images, it is necessary to control the code amount of the image due to restrictions such as transmission channels. The buffer control method described in 1 is most often used, and the configuration example will be described below with reference to the drawings.

FIG. 2 is a block diagram showing the structure of a buffer control type image compression encoding apparatus described in the above document.

This image compression coding apparatus uses an input image S _i.
The information amount compressing means 11 for removing the redundancy of and outputting the image data S11 is provided. The information amount compression means 11 is, for example, a discrete cosine transform (DCT) means or a prediction (DPC) means.
M) means, etc., and the quantizing means 1 is provided on the output side thereof.
2 and the encoding means 13 are connected. Quantization means 1
2 quantizes the image data S11 from which redundancy has been removed based on the quantization step width S15, and supplies the quantized output data S12 to the encoding means 13,
It is composed of a dread type linear quantizer or the like.

The encoding means 13 has quantized output data S1.
2 is encoded, and a variable length code (Huffman code or the like) is often used. A buffer 14 is connected to the output side of the encoding means 13, and the buffer 14
Is feedback-connected to the quantization means 12 via the buffer control means 15.

The buffer 14 temporarily stores the encoded image data S13 and outputs the output image S _o at a predetermined timing.
And a function of outputting the occupancy S14 (the code amount of the image accumulated in the buffer 14).
It is composed of IFO (First In First Out). Generally, the input of the buffer 14 is irregular and the speed (bit rate) is indefinite. Buffer control means 15
Measures the occupancy S14 of the buffer 14 at regular intervals, and from the occupancy S14 to the next quantization step width S15.
And has a function of giving it to the quantizing means 12.

Next, the operation of FIG. 2 will be described with reference to FIGS. 3 and 4.

FIG. 3 is a diagram for explaining the Midread type linear quantizer, where x on the horizontal axis is input, y on the vertical axis is output,
h indicates the quantization step width. FIG. 4 is a diagram showing a method of determining the quantization step width S15 from the occupied amount S14 of the buffer 14. B of FIG. 4 shows a buffer occupancy amount, B _max is an upper limit buffer occupancy amount, and B _min is a lower limit buffer occupancy amount. h represents a quantization step width, h _max represents a maximum step width, and h _min represents a minimum step width.

When the input image S _i is input to the information amount compression means 11, the information amount compression means 11 removes the redundancy of the input image S _i and sends the image data S 11 to the quantization means 12. The quantizing means 12 quantizes the image data S11 based on the quantizing step width S15 given from the buffer controlling means 15 and sends the quantized output data S12 to the coding means 13. If the quantizing means 12 is constituted by, for example, the Midread type linear quantizer shown in FIG. 3, the quantization can be expressed by the following equation (1).

[0011]

[Equation 1]

The encoding means 13 encodes the quantized quantized output data S12 and sends the encoded image data S13 to the buffer 14. The buffer 14 outputs the temporarily stored image data S13 to the transmission channel at a constant speed (bit rate) according to the transmission channel.

In the conventional image compression coding apparatus of the buffer control system, the occupancy S14 of the buffer 14 provided at the final stage of the compression coding apparatus is measured by the buffer control means 15,
By the buffer control means 15, the characteristics of the quantization means 12, for example, the quantization step width S15, according to the occupation amount S14.
Is changed to control the quantizing means 12 so that the occupancy S14 of the buffer 14 does not exceed a predetermined upper limit or lower limit.

That is, the buffer control means 15 measures the occupancy amount S14 of the buffer 14 at regular intervals, determines the next quantization step width S15 from the occupancy amount S14, and supplies it to the quantization means 12. .. A method of determining the quantization step width S15 from the occupancy S14 of the buffer 14 is shown in FIG.

As shown in FIG. 4, for the buffer occupancy S14 = B measured at a certain time, the next quantization step width S15 = h is determined by the following equation (2).

[0016]

[Equation 2]

[0017]

However, the apparatus having the above structure has the following problems.

(A) In the conventional image compression encoding device,
The occupancy S14 of the buffer 14 regardless of the image quality.
The quantization step width S15 of the quantization means 12 is controlled by. Therefore, in the same image, a part with good image quality and a part with poor image quality are mixed, and the image quality is deteriorated more than a numerical evaluation value (for example, a signal-to-noise ratio (SN ratio)) as a whole.

(B) In a moving image, there is a large variation in image quality between frames, and similarly, it is felt that the image quality deteriorates more than the numerical evaluation value.

(C) FIGS. 5 (a) and 5 (b) are explanatory views of conventional problems showing the above phenomenon.

FIG. 5A is a diagram showing the relationship between the passage of time and the buffer occupancy B. As shown in this figure, the code amount by the encoding means 13 changes due to the variation in local characteristics of the image, and as a result, the buffer occupancy B changes.

FIG. 5B is a diagram showing the relationship between the passage of time and the image quality evaluation value (eg, SN ratio). As shown in this figure, if the buffer occupancy B changes, the quantization step width S15 output from the buffer control means 15 changes accordingly, so the SN ratio also changes. As a result, S
The N ratio varies, and the image quality deteriorates.

The present invention provides an image compression encoding apparatus which solves the problem that the above-mentioned prior art has, such as local variations in the quality of compressed images.

[0024]

In order to solve the above-mentioned problems, the first invention is to quantize the information amount compression means for removing the redundancy of the input image, and the image data redundantly removed by the information amount compression means. An image compression coding including a quantizing means for encoding, an encoding means for encoding the quantized output data of the quantizing means, and a buffer for temporarily storing the image data encoded by the encoding means. The apparatus is provided with a statistic calculator for calculating the statistical quality of the input image and outputting the statistic, a buffer occupancy detector for detecting the occupancy of the buffer, and a code amount predictor. ing.

Here, the code amount predicting means makes the compression code amount of the image by the quantizing means and the encoding means equal to the transmission bit rate or the set code amount based on the statistic, and occupies the buffer. It has a function of predicting and controlling the quantizing means so that the quantity does not exceed a predetermined upper and lower limit.

In the second invention, the code amount predicting means of the first invention calculates beforehand the relationship between the predictive compression code amount of the image data normalized by the information amount compressing means and the quantizing means. The data is obtained and tabulated, and control data corresponding to the input signal is read out to control the quantizing means.

In the third invention, the code amount predicting means of the first or second invention determines the predictive distribution characteristic of the image based on the occupied amount of the buffer obtained from the difference between the predictive code amount and the actual code amount. It has the function of approximating the distribution characteristics of the actual image.

[0028]

According to the first aspect of the present invention, since the image compression coding apparatus is configured as described above, when the input image is input to the statistical amount calculating means and the information amount compressing means, the statistical amount calculating means: For example, the statistic amount of one input image is calculated and sent to the code amount prediction means. The information amount compression means sends to the quantizing means image data from which redundancy between input pixels of one image or redundancy between the current image and the previous image is removed.

The quantizing means quantizes the image data from which the redundancy has been removed on the basis of the control data given from the code amount predicting means, for example, the quantizing step width, and sends the quantized output data to the coding means. The encoding means encodes the quantized output data and sends the encoded image data to the buffer. The buffer temporarily stores the encoded data and outputs it to the outside at a predetermined timing.

The buffer occupancy detection means measures the occupancy of the buffer and gives an instruction to the code amount prediction means so that the occupancy does not exceed the upper and lower limits of the buffer. The code amount predicting unit predicts control data, for example, a quantization step width, so that the code amount of the image matches the transmission bit rate or the determined bit rate, based on the image statistical amount from the statistical amount calculating unit. Then, the quantizing means is controlled by the predicted control data.

Thus, by predicting the quantization step width suitable for the nature of the image and controlling the quantization means,
The local variations in the quality of the compressed image are eliminated, and the stability of image quality and bit rate can be improved.

According to the second invention, since the code amount predicting means is tabulated, the control data corresponding to the input signal is read out, and the control processing of the quantizing means for the real image is in the form of a lockup table. , Can be done at high speed.

According to the third invention, the code amount predicting means is
Based on the occupancy of the buffer, it has the function of making the predicted distribution characteristics of the image closer to the distribution characteristics of the actual image and improving the prediction accuracy. Therefore, the above problem can be solved.

[0034]

1 is a block diagram showing the configuration of an image compression coding apparatus according to an embodiment of the present invention, in which elements common to those in FIG. 2 of the related art are designated by common reference numerals.

This image compression coding apparatus is similar to the conventional one, in which orthogonal transformation (for example, DCT) and prediction (for example, DP) are performed.
A redundant amount (correlation, etc.) of the input image S _i is removed using a means such as CM or DPCM with motion compensation, and an information amount compression means 11 for outputting the redundantly removed image data S11 is provided, and on the output side thereof, Quantizing means 12, encoding means 13,
And the buffer 14 are connected. Quantization means 12
Is for quantizing the image data S11 using the given control data S23, for example, the quantization step width, and outputting the quantized output data S12 to the encoding means 13.
The encoding means 13 has a function of encoding the quantized output data S12 and outputting the image data S13 to the buffer 14. The buffer 14 has a function of temporarily storing the coded image data S13 and outputting the output image _S0 coded at a constant bit rate to the outside according to, for example, the FIFO rule.

This image compression coding apparatus is different from the conventional one in that the statistical amount calculating means 21 for inputting the input image S _i , the buffer occupation amount detecting means 22 connected to the buffer 14, and the buffer The point is that the occupancy amount detecting means 22 and the code amount predicting means 23 connected to the output side of the statistic calculating means 21 are provided.

The statistic calculating means 21 obtains statistical properties (for example, average, variance and correlation coefficient of images) of the input one input image S _i , and calculates those statistics by S.
It has a function of outputting 23 to the code amount predicting means 23. The buffer occupancy detection means 22 regularly measures the code occupancy S14 in the buffer 14 and
Has a function of giving the buffer information S22 to the code amount predicting means 23 so that the occupied amount of does not exceed the predetermined upper and lower limits.

The code amount predicting means 23 determines the code amount of the image to match the determined bit rate on the basis of the statistical amount of the given image based on S23 and the buffer information S22.
Alternatively, it has a function of predictively calculating the control data S23 (for example, an optimum quantization step width) so that the predetermined occupation amount does not exceed the upper and lower limits of the buffer 14 and giving the quantization step width to the quantization means 12. is doing.

Next, the operation of FIG. 1 will be described with reference to FIGS. When the input image S _i is input, the statistic calculator 21 calculates the following equations (3) to (6) for the input M × N input image S _i = g (x, y), for example. ), The average value μ, the variance σ ² , the horizontal correlation coefficient ρ _H , and the vertical correlation coefficient ρ _v are obtained.

[0040]

[Equation 3]

Then, those statistics S23 = (μ, σ
² , ρ _H , ρ _v ) is output to the code amount prediction means 23.

In the information amount compression means 11, for example, the image is divided into small blocks n × n, and the discrete cosine transform (D
CT) and remove the correlation. In the DCT, output image data S11 = y _b (u, v) for an input image S _i = g _b (x, y) of a certain block is _expressed by the following equations (7) to (9).

[0043]

[Equation 4]

The quantizing means 12 uses the quantizing step width h given from the code quantity predicting means 23, with respect to the inputted image data S11 = y _b (u, v), as shown in FIGS. The expression is quantized, and the output level L is output to the encoding means 13 as the quantization result.

[0045]

[Equation 5]

The encoding means 13 sets a preset code H (L) for the quantized quantized output data S12 = L.
Is assigned and the code H (L) is output to the buffer 14 as image data S13.

Further, in the encoding means 13, the inputted non-zero value L and the number S of zero data (L = 0) inputted before that are regarded as one composite data (S, L). 1 H
There is also a method of assigning a composite code (zero-run composite code) having a (S, L) word length and outputting the code. The code length H generally has a variable length.

The following method is often used to determine the code. That is, a method of determining the appearance probability of certain quantized data (or zero-run composite data) from theoretical calculation or experimental image data in advance and determining the optimum code (for example, Huffman code) of each quantized data from the probability. Is.

The buffer 14 is composed of, for example, a FIFO memory, and on the other hand, the coded image data S13 is input at any time and temporarily stored. On the other hand, the accumulated data is output at a constant rate (bit rate) R on a first-come-first-served basis.

The buffer occupancy detection means 22 measures the occupancy of the buffer 14 at regular time intervals (eg, every other frame), and outputs the occupancy S14 = B of the buffer 14 to the code amount prediction means 23. ..

The code amount predicting means 23 first calculates the energy of the compressed output data by the information amount compressing means 11 of the normalized image input data using the given correlation coefficients ρ _H and ρ _v . For example, when a block DCT is used as the information amount compression means 11, the code amount prediction means 23 has energy (dispersion) σ _N ² (n, n of each component of the normalized image input data (σ ² = 1) after DCT. v) is calculated by the following equation (11). Then, assuming that each component after the DCT has a distribution having a probability density function of the following equation (12), the quantization means 12 shown in FIG. The probability ρ (L) of the level L or the appearance probability q (S, L) of the zero-run composite data (S, L) is obtained as follows.

Note that γ of the probability density function f (x, σ, γ)
Is a parameter indicating distribution characteristics. A Gaussian distribution often used as a distribution function of an image is γ = 2, and a Laplace distribution is γ = 1. Probability density function f (x, σ,
γ) is generally expressed by the following equations (12), (13), (14)
Can be represented by.

[0053]

[Equation 6]

Under the above conditions, the probability that the quantized output of each component after DCT becomes the quantized level L is given by the following equation (15).
become that way. The same code length H for the quantization level L of each DCT component
If the code of (L) is assigned, the probability ρ (L) of the quantization level L and the average code amount C are given by the following equations (16), (1
It is required as in 7).

[0055]

[Equation 7]

(2N + 1) (N> 0) is the number of quantization steps. When the above-described zero run and non-zero composite encoding method is used, first, two-dimensional image data (y _b
(U, v)) by a scanning method using a zigzag scan or the like 1
Convert to the dimension (y _b ^(k) , k = 0, ..., N ² −1), and determine the appearance probability q (S, L) of the zero-run composite data (S, L) as follows: A code having a code length H (n, L) is assigned to the data having the appearance probability q (S, L), and the following equation (1
The average code amount is obtained from 8) and (19). In the expression (18), ρ (S, L) is the one-dimensionalization of the probability ρ (u, v, L) in the expression (15) by the same method as described above.

As described above, in the above-mentioned calculation, assuming the correlation coefficients ρ _H and ρ _v with respect to the determined quantization method and coding method, the pre-normalized image data (σ ² = 1)
In contrast, the average code amount can be calculated.

FIGS. 6 to 8 are diagrams showing the relationship between the average predicted code amount and the normalized quantization step width when the Huffman code is assigned to the above-mentioned normalized image data (σ ² = 1). .. The parameter ρ in the figure is the correlation coefficient ρ _H = ρ _v =
ρ. Figure 6 shows the gamma (1/2) distribution (γ = 1 /
In the case of 2), FIG. 7 shows the case of Laplace distribution (γ = 1)
Is the case of Gaussian distribution (γ = 2).

In the actual processing, first, the above-mentioned calculation is performed in advance on the normalized image data and tabulated. Then, with respect to the input image, a data string of the code amount and the quantization step width at which the correlation coefficients match is extracted from the table of the preset distribution (for example, γ = 1). From that, the normalized quantization step width h at which the average code amount matches or is closest to the transmission bit rate is selected and multiplied by the standard deviation σ of the actual image (h
Since * σ) is the actual quantization step width, the value (h * σ) is output to the quantization means 12.

Further, in the code amount prediction means 23, the buffer information S22 input from the buffer occupancy amount detection means 22.
Therefore, the buffer occupancy S14 is constantly checked.
When the buffer occupancy S14 exceeds a predetermined upper limit, the quantization step width h is forcibly increased and the compression code amount is decreased. When the buffer occupancy S14 exceeds the predetermined lower limit, the quantization step width h is forcibly reduced and the compression code amount is increased. Alternatively, a predetermined specific code is input to the buffer 14 to prevent underflow of the buffer 14.

Furthermore, as another embodiment, the buffer occupancy amount of one image unit given by the buffer occupancy amount detecting means 22 is checked. And the buffer occupancy S14
If increases, lower the distribution function parameter γ,
When the buffer occupancy S14 decreases, the accuracy of prediction is improved by increasing the parameter γ of the distribution function so that the theoretical distribution function always approaches the distribution of actual image data.

As an example of a method of changing the distribution function parameter γ depending on the buffer occupancy of each image, there is a method of selecting the corresponding parameter γ on the curve of the buffer occupancy and the parameter γ shown in FIG.

As described above, the various embodiments of the present invention have the following advantages.

(I) FIGS. 10 (a) and 10 (b) are diagrams showing the effect of the embodiment of the present invention. FIG. 10 (a) shows a buffer occupation amount, and FIG. 10 (b) shows It is a figure which shows the state of image quality (SN ratio).

As shown in FIG. 10A, although the buffer occupancy S14 changes locally, the code amount predicting means 23 predicts that the buffer occupancy S14 will be a stable and constant value for each image. Therefore, it is not necessary to change the quantizer following a local change in the code amount. Therefore, stable image quality (S
N ratio) is obtained.

FIG. 10B is a diagram showing the state of the SN ratio. Compared with FIG. 5B showing the characteristics of the conventional system, the system of this embodiment has a more stable image quality (SN ratio). It can be seen that

(Ii) Conventionally, since the code amount could be controlled only passively, the image quality locally changes due to the local change of the code amount, and as a result, the evaluation value or more is exceeded. The image quality deteriorates. On the other hand, in the present embodiment, since the code amount of the image can be predicted in advance by the code amount predicting means 23, the quantizer suitable for the image can be actively selected according to the transfer bit rate and the like. Therefore, as shown in (i) above, a stable transmission bit rate and stable high image quality can be maintained for the entire device.

The present invention is not limited to the above embodiment,
For example, the constituent blocks in FIG. 1 may be configured by individual circuits using integrated circuits or the like, or may be configured by program control using a computer. Further, in order to improve the compression coding accuracy, it is possible to add other functional blocks or the like to the device of FIG.

[0069]

As described in detail above, according to the first invention, the control data such as the quantization step width suitable for the property of each image is predicted by the code amount predicting means, and the control data is calculated. Since the quantizing means is controlled based on the above, the quantizing suitable for the image can be performed according to the transmission bit rate and the like. Therefore, it is possible to prevent the local variation in the quality of the compressed image, and to maintain a stable transmission bit rate and a stable high image quality as a whole.

According to the second invention, since the code amount predicting means is tabulated, the control data generation process for the actual image can be realized at a high speed in the form of a lookup table.

According to the third invention, since the code amount predicting means is provided with the function of approximating the predicted distribution characteristic of the image to the distribution characteristic of the actual image based on the occupied amount of the buffer, the prediction accuracy is improved, As a result, it is possible to perform compression encoding with more stable image quality and bit rate.

[Brief description of drawings]

FIG. 1 is a configuration block diagram of an image compression encoding apparatus showing an embodiment of the present invention.

FIG. 2 is a configuration block diagram showing a conventional image compression encoding device.

FIG. 3 is a diagram for explaining a Midread type linear quantizer.

FIG. 4 is a diagram showing a conventional method of determining a quantization step width.

FIG. 5 is an explanatory diagram of a conventional problem.

FIG. 6 is a diagram showing a relationship between an average prediction code amount and a normalized quantization step width in the case of Huffman distribution according to the present embodiment.

FIG. 7 is a diagram showing a relationship between an average prediction code amount and a normalized quantization step width in the case of Laplace distribution according to the present embodiment.

FIG. 8 is a diagram showing a relationship between an average prediction code amount and a normalized quantization step width in the case of a Gaussian distribution according to the present embodiment.

FIG. 9 is a diagram showing a method of changing a distribution function parameter according to the present embodiment.

FIG. 10 is a diagram showing an advantage of this embodiment.

[Explanation of symbols]

 11 information amount compression means 12 quantization means 13 encoding means 14 buffer 21 statistic amount calculation means 22 buffer occupancy detection means 23 code amount prediction means

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Yoko Harada 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd.

Claims (3)

[Claims] 1. An information amount compression means for removing redundancy of an input image, a quantization means for quantizing the image data of which redundancy has been removed by the information amount compression means, and a quantization output data of the quantization means. In an image compression encoding device including an encoding unit for encoding and a buffer for temporarily storing the image data encoded by the encoding unit, a statistical quantity is calculated by calculating the statistical quality of the input image. , A buffer occupancy amount detecting unit for detecting an occupancy amount of the buffer, and a compression bit amount of the image by the quantizing unit and the encoding unit based on the statistic amount is a transmission bit rate or a setting. And a code amount predicting unit for predictively controlling the quantizing unit so that the amount of code becomes the occupied amount of the buffer and does not exceed a predetermined upper and lower limit. Compression encoding device. 2. The image compression coding apparatus according to claim 1, wherein the code amount prediction unit has a relationship between a predicted compression code amount of the image data normalized by the information amount compression unit and a quantization unit. Is calculated in advance and tabulated,
An image compression coding apparatus configured to read control data according to an input signal and control the quantizing means. 3. The image compression encoding apparatus according to claim 1, wherein the code amount predicting unit predicts an image based on an occupied amount of the buffer obtained from a difference between a predicted code amount and an actual code amount. An image compression encoding device having a function of bringing distribution characteristics closer to those of an actual image.