JP2934342B2 - Image compression coding device - Google Patents

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 a still image or a moving image, and more particularly to an image compression encoding apparatus for controlling a signal-to-noise ratio (hereinafter, referred to as an SN ratio) of an image, that is, quality. It is.

[0002]

2. Description of the Related Art Conventionally, techniques in such a field include:
For example, there is one described in the following literature.

[0003] IEEE TRANSACTIONS ON COMMUNICATIONS
COMMUNICATIONS) COM-25 [II] (1977-1
1) (US) ALI HABIBI, “Survey Off
Survey of Adaptive Image Coding Techniques ”
P. Conventionally, as described in the above-mentioned document, an image compression / encoding device performs efficient compression of an image by removing spatial or temporal redundancy (correlation) of the image. ing. Hereinafter, the configuration will be described with reference to the drawings.

FIG. 2 is a block diagram showing an example of the configuration of a conventional image compression encoding apparatus described in the above-mentioned document.

[0005] This image compression encoding apparatus uses an input image Si
Is input. The information amount compression means 11 removes the temporal or spatial redundancy of the input image Si and outputs the image signal S11. For example, a discrete cosine transform (DCT) means, a prediction (DPCM)
It is composed of means and the like. An encoding unit 13 is connected to an output side of the information amount compression unit 11 via a quantization unit 12.

The quantizing means 12 quantizes the image data S11 from which the redundancy has been removed and outputs quantized data S12, and is constituted by a Midread-type linear quantizer or the like. The encoding means 13 generates the quantized data S
12 and outputs encoded data So compressed by compressing the compressed data S12.
(Variance), or variable length coding means for allocating codes according to the appearance probability of the quantized data S12 and coding.

Next, the operation of FIG. 2 will be described with reference to FIG. FIG. 3 is a diagram showing a conventional Midread-type linear quantizer. X on the horizontal axis is the quantization input, y on the vertical axis
Is the quantized output. h is the quantization step width.

[0008] The information image compression means 11 removes spatial or temporal redundancy from the input image Si. The data amount (the number of bits) of the image signal S11 from which the redundancy has been removed is compressed by the quantization means 12 and the coding means 13, and the coded data So is output.

Here, when the quantization means 12 is constituted by, for example, a Midread-type linear quantizer shown in FIG.
In the linear quantizer, the image signal S11 is quantized and quantized.
The data S12 is output to the encoding means 13. This quantization
The data S12, that is, the quantized output y shown in FIG. 3 can be expressed by the following equation (1).

[0010]

(Equation 1)

In the equation (1), INT [x] is x
Means taking the integer part of Then, the quantized output y is encoded by the encoding means 13, and compressed encoded data So is output.

[0012]

However, in the apparatus having the above configuration, the quantization step width h in the quantization means 12 is not limited to the above.
By changing the compression ratio, the compression ratio of the image can be controlled, but it is difficult to control the SN ratio of the image, that is, the quality. Therefore, when image compression of a certain quality (SN ratio) is required, the conventional apparatus cannot cope with the problem and the problem of unstable image quality occurs, and it is difficult to solve the problem. Was.

An object of the present invention is to provide an image compression encoding apparatus which solves the problem of the prior art as to the instability of the quality of a compressed image.

[0014]

Means for Solving the Problems] To solve the above problems
In the first aspect of the present invention, the result of quantization is
Feed Forward method for predicting and controlling quantization
An image compression encoding apparatus of the type comprising the following means:
I have.

That is, in the first invention, when an input image is
Outputs image signals without intermittent or spatial redundancy
Using information compression means and quantization means control data
Quantizing the image signal output from the information amount compression means
Quantizing means for converting and outputting quantized data;
Statistics that calculate the statistical properties of an image and output the statistics
Amount calculating means, and the S / N ratio of a predetermined compressed image
Predict the relationship with the characteristics of the quantization means using the statistics
And the quantization step is performed so that the SN ratio becomes a predetermined level.
The quantization means control data for controlling the stage
And an output from the quantization means.
Encoding the input quantized data to obtain encoded data.
Encoding means for outputting.

In the second invention, the S / N ratio predicting means of the first invention obtains the result of the prediction in advance by calculation, and
And control of the quantization means according to the statistics.
It is configured to read and output data.

In a third aspect of the present invention, the SN ratio predicting means of the first or second aspect of the present invention comprises the step of:
Predicted quantization error power determined according to the ratio and the statistic ,
The distribution function of the image is adjusted by comparing the measured quantization error power, which is an error between the image signal output from the information amount compression unit and the actually measured quantization signal obtained by measuring the quantization result of the image signal. Configuration.

[0018]

According to the first aspect of the present invention, since the image compression encoding apparatus is configured as described above, when an input image is input to the statistic calculation means and the information compression means, the statistic calculation means: Calculate the statistical properties of the input image and calculate the statistic as SN
Give to ratio prediction means. The information amount compression unit removes the redundancy of the input image and supplies the image signal from which the redundancy has been removed to the quantization unit. The SN ratio prediction means is provided in advance.
The S / N ratio of the compressed image and the characteristics of the quantization means (for example, quantization
Prediction using the input image statistics
In order to satisfy this requirement, the S / N ratio (image quality)
The quantizer control data for controlling the stage is output to the quantizer.
Power.

In the quantizing means, the quantizing means control data is
To quantize the image signal and encode the quantized data
Give to the column. The encoding means encodes the quantized data and
Output encoded data.

According to the second invention, the SN ratio predicting means includes:
For example, the calculation is performed in advance by assuming a correlation coefficient or the like, and the calculation result is tabulated .
If you enter a statistic that is al outputted, according to the statistic
The quantization means control data is read to control the quantization means.

According to the third invention, the SN ratio predicting means comprises:
Comparing predicted quantization error power and measured quantization error power
The prediction is performed by adjusting the distribution function of the image.

[0022]

FIG. 1 is a block diagram showing the configuration of an image compression encoding apparatus according to an embodiment of the present invention. Elements common to those in FIG. 2 are denoted by the same reference numerals.

In this image compression encoding apparatus, as in the prior art, the information amount compression unit 11 for removing the redundancy of the input image Si and outputting the image signal S11, and the quantization unit control data S23 (for example, the quantization step width) image signal on the basis of an equal)
Issue S11 and quantization means 12 for outputting the quantized data S12 to quantize, code by encoding the quantized data S12
Encoding means 13 for outputting encoded data So. This device differs from the conventional one in that the input image Si
, A quantization error measurement unit 22 connected to the input / output side of the quantization unit 12, and connected to the output side of the quantization error measurement unit 22 and the statistic calculation unit 21. That is, an SN ratio prediction unit 23 is added.

The statistic calculation means 21 obtains the statistical properties (for example, average, variance, correlation, etc.) of one input image Si, and sends the statistic S21 to the SN ratio prediction means 23. It has the function of giving. The quantization error measuring means 22 calculates an actually measured quantization error signal from the input image signal S11 before and after the quantization of one image and the quantized data S12.
It has a function of measuring the force S22 and giving it to the SN ratio prediction means 23. The SN ratio prediction unit 23 calculates the relationship between the quantization step width and the compressed SN ratio when a predetermined quantization method (for example, quantization step width) is used, based on the image statistics S21. The control data S23 (for example, a quantization step width) is calculated so that the predicted compression S / N ratio matches an externally requested SI or an internal predetermined S / N ratio, and the quantization step width is given to the quantization means 12. Has a function. Further, the SN ratio prediction means 23
Is the actual compressed image measured by the quantization error measuring means 22.
It has a function of improving the prediction accuracy of the compressed SN ratio from the ratio between the measured quantization error power S22 and the predicted quantization error power .

Next, the operation will be described. When the input image Si is input to the statistic calculation means 21 and the information amount compression means 11, the statistic calculation means 21 outputs one input M ×
For N input images Si = g (x, y), the following equations (2) to
From (5), the average value μ of the g (x, y), the variance σ ² ,
The horizontal correlation coefficient ρ _H and the vertical correlation coefficient ρ _V are obtained.

[0026]

(Equation 2)

Then, the statistics S21 = (μ, σ)
² , ρ _H , ρ _V ) to the SN ratio prediction means 23.

In the information amount compression means 11, for example, after dividing the input image Si into small blocks (n × n), a discrete cosine transform (DCT) is performed to remove the correlation. In this DCT, an image signal S is applied to an input image Si = g _b (x, y).
11 = y _b (u, v) is _expressed by the following equations (6) to (8).

[0029]

(Equation 3)

The quantization means 12 uses the control data S23 given by the SN ratio prediction means 23, for example, the quantization step width h, and uses the input image signal S11 = y _b (u, u
v) perform 3 and (1) quantization relative to quantization
Data _{S12 = Z b (u, v} ) is output in accordance with the following equation (10) a.

[0031]

(Equation 4)

The coding means 13 assigns the quantized data S12, a code preset, the encoded data
Data So is output.

In the quantization error measuring means 22, the image signal S11 of one image before quantization and the quantized data S
Mean square error ( actually measured quantization error power ) S12 with S12
(= MSE) and its measured quantization error power S22
(= MSE) to the SN ratio prediction means 23.

The image signal S11 before quantization is converted to y (x,
y), assuming that the quantized data S12 after quantization is z (x, y), the measured quantization error power S22 (= MSE) is expressed by the following equation (11).

[0035]

(Equation 5)

In the SN ratio prediction means 23, first, based on the correlation coefficient of the image in the statistic S21 given from the statistic calculation means 21, the variance of the image signal S11 from which the information amount compression means 11 has removed redundancy is used. Predict. For example, assuming that the information amount compressing unit 11 uses block cosine transform, the variance σ _N ² (of the components y _b (u, v) of the normalized image signal S11 (σ ² = 1) after block cosine transform is used. u , v)
Can be expressed by the following equation (12). _{n-1 n-1 n-1 n-1 | x-x1 |} σ ² _N (u, v) = Σ Σ Σ Σ ρ _H ^{x = 0 y = 0 x1 = 0 y1 = 0} _{| y-y1 |} × ρ _V · φ ^{(u, v)} (x, y) · φ ^{(u, v)} (x1, y1) (12) Then, the distribution of each component after DCT is the probability density function f
(X, σ, r), which is used to predict the quantization error power of the image by quantization. Here, the probability density function f (x, σ, r) can be generally represented by the following equation (13).

[0037]

(Equation 6)

R is a parameter indicating a distribution characteristic.
A Gaussian distribution often used as an image distribution is r = 2
In the Laplace distribution, r = 1.

When the quantizing means 12 uses, for example, the quantizer shown in FIG. 3 and equation (1), the quantization step width is h and the number of quantization steps is sufficiently large (N is sufficiently large).
Then, the quantization error power σ _EN ² (u, v) of each component of the normalized image data after DCT can be expressed by the following equation (16). _{N (l + 1/2) h} σ _EN ² (u, v) = Σ ∫ [x-lh] ² × ^{l = -N (l-1 / 2) h} f (x, σ _N (u, v) , R) dx (16) The average quantization error power σ _EN ² of the normalized image is given by the following equation (17).
become that way.

[0040]

(Equation 7)

Then, the predicted quantization error power (σ _E ² ) and SN ratio (SNR) of an image having a variance of σ ² are given by the following equations (18),
It becomes like (19).

[0042]

(Equation 8)

From the above calculations, it is assumed that the distribution of the image data is known, and if the correlation coefficient of a certain image is known, the normalized quantization error power of the image and the relationship between the compression SN ratio and the quantization step width are obtained. Can be predicted. That is, assuming that a compression S / N ratio required from the outside is given, the quantization step width can be predicted therefrom.

The above calculation can be performed by assuming a correlation coefficient in advance. Therefore, if the calculation results are tabulated, the calculation processing on the actual image can be performed at a very high speed in the form of a lookup table.

FIGS. 4 to 6 are diagrams showing the relationship between the normalized quantization step width h calculated in advance as the correlation coefficient ρ _H = ρ _v = ρ and the normalized quantization error power σ ² _EN . . FIG. 4 shows the case of the gamma (1/2) distribution (r = 1/2), FIG. 5 shows the case of the Laplace distribution (r = 1), and FIG.
This is the case of 2).

Further, the SN ratio prediction means 23 takes the ratio SS = MSE / σ _E ² (20) between the predicted quantization error power σ ² _E and the measured quantization error power MSE. Then, if S> 1, the distribution coefficient r is slightly increased, and if S <1, r is slightly decreased, and the accuracy of prediction can be improved. The following equation (21) shows an example of a method of changing the distribution coefficient r.

[0047]

(Equation 9)

As described above, in the present embodiment, it is possible to respond to the request SI for the compression S / N ratio from the outside, thereby enabling image compression of stable image quality.

[0049]

As described in detail above, according to the first aspect, the S / N ratio of a given compressed image and the quantization
Predict the relationship with the step characteristics using the statistics of the input image,
Control the quantization means so that this SN ratio becomes a predetermined level
SNR prediction means for outputting quantization means control data
Provided, a compressed image of a certain quality (SN ratio) is required
If the request is
Quality image compression is possible.

According to the second invention, the SN ratio prediction means is
Was calculated by assuming a pre-correlation coefficient, etc., since the calculation result so that the table by using a memory or the like, quantization means control data based on the statistical quantity <br/> is read, the read Since the quantization means is controlled by the control data obtained, high-speed processing can be realized.

According to the third aspect, the SN ratio predicting means includes:
Compare the predicted quantization error power with the measured quantization error power ,
Since the configuration is such that the distribution function of the image is adjusted,
The prediction accuracy of the N-ratio predicting means can be improved, whereby a desired compressed S / N ratio (image quality) can be more accurately obtained.

[Brief description of the drawings]

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

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

FIG. 3 is a diagram illustrating a Midstream-type linear quantizer.

FIG. 4 is a diagram illustrating a relationship between a normalized quantization step width h and a normalized quantization error power σ _EN ² in the embodiment when the information-compressed image has a gamma distribution.

FIG. 5 is a diagram illustrating a relationship between a normalized quantization step width h and a normalized quantization error power σ _EN ² in an embodiment in a case where an information amount compressed image has a Laplace distribution.

FIG. 6 is a diagram showing a relationship between a normalized quantization step width h and a normalized quantization error power σ _EN ² in the embodiment in the case where the information-compressed image has a Gaussian distribution.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 Information amount compression means 12 Quantization means 13 Encoding means 21 Statistical amount calculation means 22 Quantization error measurement means 23 SN ratio prediction means

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Koji Sakurada 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (72) Inventor Yoko Harada 1-712 Toranomon, Minato-ku, Tokyo No. Oki Electric Industry Co., Ltd. (56) Reference JP-A-63-222593 (JP, A) IEICE Technical Report, IE 88-88 53−60

Claims (3)

(57) [Claims] 1. An information amount compression unit for outputting an image signal from which a temporal redundancy or a spatial redundancy of an input image has been removed, and the image signal output from the information amount compression unit using quantization unit control data. Quantizing means for quantizing and outputting quantized data; statistic calculating means for calculating a statistical property of the input image and outputting a statistic thereof; and a signal-to-noise ratio of a given compressed image and The relation between the characteristic of the quantizing means and the statistical quantity is predicted, and the quantizing means control data for controlling the quantizing means so that the signal-to-noise ratio becomes a predetermined level is output to the quantizing means. An image compression encoding apparatus comprising: a signal-to-noise ratio prediction unit that performs encoding; and an encoding unit that encodes the quantized data output from the quantization unit and outputs encoded data. 2. The signal-to-noise ratio prediction means calculates a result of the prediction in advance and forms a table, and reads and outputs the quantization means control data according to the statistic. The image compression encoding apparatus according to claim 1. 3. The signal-to-noise ratio predictor includes a signal-to-noise ratio of the predetermined compressed image and the statistical value.
The predicted quantization error power determined according to the amount, and the measured quantization error power which is the error between the image signal output from the information amount compression unit and the measured quantization signal obtained by actually measuring the quantization result of the image signal. 3. The image compression encoding apparatus according to claim 1, wherein the image compression function is adjusted by comparing the following.