CN1494323A - Method of proceeding quantization/reverse quantization using produced aynamic meter and device - Google Patents

Abstract

As fast development of Internet, high resolution TV, and visible phone, content of information is changed from character as center to visual cartoon and image as center. It is unreasonable to use general method for storing and transferring image data. The main points of the technique are that the invention includes phases: phase of picking up complexity from input any image as well as phase of generating given quantization table based on said complexity and discrete cosine transform, and phase of quantization processing based on generated quantization table. The invention generates nost suitable quantization table for any image inputted so as to eliminate high frequency field effectively, and to reduce quantity of bit.

Description

The dynamic meter that utilize to generate quantizes/method and the device of inverse quantization

[technical field]

The present invention relates to utilize the dynamic meter of generation to quantize/method and the device of inverse quantization.Specifically, it is relevant to quantize by generating the dynamic meter that is suitable for importing video most/a kind of method of inverse quantization and device.

[background technology]

Recently, medium (mebia) industry that begins to develop from the type medium, along with the high speed development of the develop rapidly of internet and HDTV (high definition TV), visual telephony, be the visual content at center by changing into animation, video etc. as the literal center of supertext.This pictorial data, because of its data volume very huge, by common storage, transfer approach, no matter still be price from function on, all very unreasonable.

Fig. 1 shows the conventional codec block diagram; And Fig. 2 shows INTERA (interactive mode) quantization table of traditional MPEG (mobile motion picture expert group version).

As shown in Figure 1, traditional cartoon compressing method is that when any video is transfused to, complexity calculating part (110) will calculate the complexity of input video, be sent to quantization unit (140) then.Meanwhile, discrete cosine transform handling part (120) is sent to quantization unit (140) after also the video information conversion of input being become discrete cosine.

At this moment, in quantization unit (140), the index information according to from complexity calculating part (110) and discrete cosine transform handling part (120) acquisition will detect the video compression ratio, as shown in Figure 2 on the quantization table of making in advance.

Fig. 3 illustrates the quantizer block diagram that utilizes multiple quantization table.

As shown in Figure 3, in Korean Patent bulletin 1019980132895, record a kind of method.This method is on the basis of having analyzed diversified video, has tentatively generated several representative tables, and goes respectively on encoder and decoder after the numerical value of agreement table, the index of going to transmit selecteed tabular value when coding.

Said method changes so can adapt to various video because of utilizing the table of Duoing than traditional standard scale number.But under above-mentioned situation, can not fine show other video, and when the table number increases, will cause that the bit number of the index that is transmitted increases, thereby produce the shortcoming that code efficiency descends.

Second method is the table that uses in encoder, covers in the bit ordered series of numbers and transmits; At this moment unengaged table also can restore.

But the bit quantity of transmission table will increase with geometric progression, the problem that can cause code efficiency to reduce greatly.

In addition, above-mentioned certain methods on technique of expression, the standard that neither one is clear and definite.And because of the leap ahead of crossing of bit quantity, the improvement that resembles matter is also little.And lay stress on usually tabular value is fixed on the recommended value, and it is used for improving quantization step, to be suitable for quantizer.

But these methods, in frequency field, high frequency separated with low frequency and treat, so though separated out the parameter of utilizing human visual system, because tabular value is fixed, during actual the use low frequency and high frequency there are identical reaction, thereby its characteristic is reduced by half.

[summary of the invention]

To the objective of the invention is the problem that exists in the conventional art in order going effectively to solve, the video of current input to be generated a kind of suitable quantization table, and provide a kind of and generate suitable dynamic meter with quantification/quantification method and device.

In order to finish above-mentioned technical task, the present invention is provided with following several stages: promptly, be provided with the stage of the complexity of extracting from image arbitrarily;

Be provided with the stage that generates certain quantization table by by the complexity of being extracted and discrete cosine transform;

Be provided with the stage of carrying out quantification treatment according to the quantization table that generates.

In order to finish above-mentioned technical task, the present invention also will be provided with following several stages on the other hand: promptly, be provided with according to quantization step and go in the prediction error code that produces from coding generating capacity controller stage of inverse quantization in addition;

Be provided with and generate the stage of regeneration rub-out signal then being transformed into anti-discrete cosine by the image data of inverse quantization;

Be provided with and utilize regeneration rub-out signal and complexity, the stage of the image that goes to regenerate.

In order to finish above-mentioned technical task, the present invention also will be provided with following several means again on the other hand: promptly, be provided with for predict and compensate input any image dynamically, the complexity of handling that performs calculations is calculated means;

Be provided with the discrete cosine transform processing means of any image data of input being carried out the orthogonal conversion;

Be provided with to the memory data output that makes buffer maintains a specified quantitative, be used for controlling the symbol generating capacity control device that quantization step is used;

Be provided with by the complexity of in complexity calculating means, calculating and generate a kind of quantization table that is fit to any image characteristics, and utilize the quantization table of scan mode and generation arbitrarily, will be the certain image data that provides from discrete cosine transform processing means, the quantification means that quantize;

The image data that is provided with quantizing carries out information encoded entropy coding means.

In order to finish above-mentioned technical task, the present invention also will be provided with following several means again on the other hand: promptly, be provided with by quantizing step-length, the prediction error code carried out the inverse quantization means of inverse quantization from the quantification means;

Be provided with from the inverse quantization means and will become anti-discrete cosine to signal transformation, so as to generating the discrete cosine transform processing means that a regeneration rub-out signal is used;

Be provided with and utilize regeneration rub-out signal and prediction signal, generate the anti-comentropy coding means of reconstructed images.

As previously mentioned, the dynamic meter quantification/quantification method and the device that generate according to the present invention, any video to input generates a kind of only quantization table, and goes to remove the high frequency field effectively with it, and bit quantity is reduced; And can be easily remove the formation table by separately frequency field.

[description of drawings]

Fig. 1 illustrates the conventional codec block diagram.

Fig. 2 illustrates INTERA (interactive mode) quantization table of traditional MPEG (mobile motion picture expert group version).

Fig. 3 illustrates the quantizer block diagram that utilizes multiple quantization table.

Fig. 4 illustrates the encoder section block diagram of the present invention's one example.

Fig. 5 illustrates the decoder part block diagram of the present invention's one example.

The explanation of symbol among the figure

400: complexity is calculated means

410: means are handled in discrete cosine transform

420: the quantification means

430: symbol generating capacity control device

440: the comentropy coding means

510: the inverse quantization means

520: anti-comentropy coding means

530: means are handled in discrete cosine transform

[embodiment]

With reference to the accompanying drawings 4 and Fig. 5, quantize by generating a kind of dynamic meter with regard to utilizing/example of quantification method, be elaborated.

Fig. 4 illustrates the part block diagram of the encoder of the present invention's one example.

As shown in Figure 4, in order to realize generating the quantization method of dynamic meter, be provided with complexity and calculate means (400), means (410), quantification means (420), symbol generating capacity control device (430), comentropy coding means (440) are handled in discrete cosine transform.

Complexity is calculated means (400), and the characteristic parameter for any image of input of extracting will carry out the calculating of complexity.

Means (410) are handled in discrete cosine transform, will be divided into square block to any image data pixel of input; And as shown in Figure 2, in the data of pixel block unit, the image data of low frequency composition, biasing transforms to the left upper end part; And the image data of high frequency composition, biasing transforms to the bottom righthand side part.At this moment,, be distributed in the unnecessary radio-frequency component of bottom righthand side, can under the situation of losing the image data amount hardly, compress so if remove in the image data.

Quantification means (420) will be utilized in complexity and calculate the complexity of calculating in the means (400), go to generate the quantization table that is fit to any image characteristics of input, and any image data is quantized.

Symbol generating capacity control device (430) will remove to control quantization step, make the memory space of buffer (not shown) maintain specified quantitative.

Comentropy coding means (440) will go to quantize to handle the signal that means (410) provide, generation forecast error code by discrete cosine transform according to the quantization step that obtains from symbol generating capacity control device (430).

Fig. 5 illustrates the part block diagram of the decoder of the present invention's one example.

As shown in fig. 5, in order to realize generating the quantization method of dynamic meter, be provided with by anti-comentropy coding means (510), the decoder that means (530) are formed is handled in inverse quantization means (520) and discrete cosine transform.

Inverse quantization means (520), according to quantization step, to the prediction error code decipher, carry out the calculating identical by decoder with the video recuperation after, restore the identical inferior image signal of decoder.

Anti-comentropy coding means (510), the signal with inverse DCT (520) restores is transformed into anti-discrete cosine, and regeneration error code signal is offered adder.In this case, utilize regeneration error code signal and complexity, add and calculate prediction signal, constitute new recovery video, so that remove to predict next input video.

To describe in detail below with a kind of quantization method that utilizes the quantification means that generates dynamic meter.

At first, from any image data of input, if import any one pixel, promptly a code word according to the spatial frequency characteristic of image data, is handled means (410) by discrete cosine transform so, with low frequency composition and high frequency composition, bias profile is to left upper end and bottom righthand side respectively.

And, for the complexity of any video of extracting, decision utilize the examination of following mathematics,

[mathematics examination 1]

${\mathrm{\σ}}^{\′}=\frac{1}{10}\sqrt{\frac{\mathrm{n\Σf}{\left(x\right)}^2-{\left(\mathrm{\Σf}\left(x\right)\right)}^2}{n(n-1)}}$

Above-mentioned complexity is to come value with 1/10th ratio of standard deviation; It is to calculate according to pixel value (x) in the pixel piece and pixel number (n).In this case, the value of gained is bigger, and expression changes strong high frequency composition the more; Its value little unconverted low frequency composition of then representing of healing.At this moment, above-mentioned complexity can replace being waited by dispersion or general standard deviation.

In addition, the video of input can utilize the image that is restored in the decoder.The phenomenon on the super top of bit quantity in this case, there is no need to transmit the value that calculates by above-mentioned mathematical expression 1, so can not take place.

Then, with reference to complexity, remove to generate given quantization table.At this moment the mathematical expression of generating quantification is as follows.

[mathematical expression 2]

$q(u,v)=\frac{1}{1+{\mathrm{\σ}}^{\′}{e}^{-r(\sqrt{u^2+v^2}-\mathrm{center})}}$

In the formula, center represents the center of pixel piece.For example, under 8 * 8 situation, be to be ∑ 24.5.R represents the gradient of low frequency composition and high frequency composition intersection.At this moment, the optimal value of r is 0.5 ~ 1.2.

According to the quantization table of mathematical expression 2 generations, after following mathematical expression 3 converts, as seeing in the standard scheme, being quantized.

[mathematical expression 3]

$\hat{F}(u,v)=\mathrm{round}\left(\frac{F(u,v)}{Q(u,v)\mathrm{xmquant}}\right)$

Here, (u v) is the coefficient value after the transition coding to F; And Q (u v) is the quantization table that utilizes mathematical expression 2 to generate.The value of using in order to generate above-mentioned quantization table will be transmitted behind the process comentropy coding for inverse quantization in decoder again.

The mathematical expression of using above 2, the monobasic formula that can be transformed into following mathematical expression 4 is used.

[mathematical expression 4]

$q^{\′}\left(u\right)=\frac{1}{1+{\mathrm{\σ}}^{\′}{e}^{-r(u-\mathrm{center})}}$

In the formula, σ ' the decision high frequency that in mathematical expression 2, occurs and the boundary of low frequency; The increase of σ ' will be played the effect that the distribution that makes the quantization table content is moved to the high frequency aspect; On the whole, only to being that decipher in the narrow field of benchmark with the DC (direct current) in the picture piece.And, make the major part of radio-frequency component become 0, improved code efficiency, reduced bit rate.

In addition, r is illustrated in the gradient in the boundary.Therefore, the r value is littler, more can obtain satisfactory gradient; And suitable low numerical value, can go to reduce quantized value contingent error on boundary section.In mathematical expression 4, if r 0, mathematical expression 2 will become 1/ (1+ σ ') so, will play the effect that resembles linear quantizer.Gradient is more satisfactory, and the radio-frequency component that is comprised will be the more, and the reflection human vision property just tails off.

Then, in order to remove the radio-frequency component in the image data of being partial to different field separately, generating quantification table in quantification means (420); In the process that quantizes by quantization table, it is transformed into a quantification image data that only has the low frequency composition, and then with its output.

The image data that only has the low frequency composition will carry out encoding process by the comentropy coding means.

The present invention though just an illustrated example is illustrated, can both know as long as possess the people of the common knowledge in present technique field as a reference, can derive thus diversified, and other impartial example.For this reason, to technical protection scope of the present invention, the technological thought according to claim should be determined.

Claims (8)

1, a kind ofly utilize the dynamic meter of generation to quantize/method and the device of inverse quantization, it is characterized in that: it is provided with the stage of the complexity of extracting from any video that is transfused to; Be provided with by by the complexity of being extracted and the conversion of discrete cosine, generate the given quantization table stage; And be provided with the stage of carrying out quantification treatment according to the quantization table that generates.

2, the dynamic meter that generates according to the described utilization of claim 1 quantizes/method and the device of inverse quantization, and it is characterized in that: quantization table actively generates by following formula.

$q(u,v)=\frac{1}{1+{\mathrm{\σ}}^{\′}{e}^{-r(\sqrt{u^2+v^2}-\mathrm{center})}}$

3, the dynamic meter that generates according to the described utilization of claim 2 quantizes/method and the device of inverse quantization, and it is characterized in that: the span of described r value is 0.5-1.2.

4, the dynamic meter that generates according to the described utilization of claim 1 quantizes/method and the device of inverse quantization, it is characterized in that: be to generate dynamic meter, utilize the following formula will be the border of the quantization parameter value of high and low frequency, that can design be variable.

$q(u,v)=\frac{1}{1+{\mathrm{\σ}}^{\′}{e}^{-r(\sqrt{u^2+v^2}-\mathrm{center})}}$

5, the dynamic meter that generates according to the described utilization of claim 1 quantizes/method and the device of inverse quantization, it is characterized in that: for generating dynamic meter, utilize following formula will go to adjust the whole quantization parameter level and the gradient of high frequency and low frequency.

$q(u,v)=\frac{1}{1+{\mathrm{\σ}}^{\′}{e}^{-r(\sqrt{u^2+v^2}-\mathrm{center})}}$

6, a kind ofly utilize the dynamic meter of generation to quantize/method and the device of inverse quantization, it is characterized in that: it is provided with according to quantization step, the prediction error code in the coding generating capacity controller is carried out the stage of inverse quantization; Be provided with an inverse quantization image data and be transformed into anti-discrete cosine, go to generate a kind of stage of the rub-out signal of regenerating then; And utilize regeneration rub-out signal and complexity to regenerate stage of video.

7, a kind ofly utilize the dynamic meter of generation to quantize/method and the device of inverse quantization, it is characterized in that: it is provided with complexity and calculates means, and it is the processing that dynamically performs calculations of any video of prediction and compensation input; Be provided with the discrete cosine transform processing means of any image data of input being carried out the orthogonal conversion; Be provided with symbol generating capacity control device, it is to make the data savings amount of buffer maintain a specific amount, removes to control quantization step; By the complexity of in complexity calculating means, being calculated, generate a kind of quantization table that is suitable for importing any image characteristics, and utilize the quantization table of scan mode and generation arbitrarily, will be the certain image data that provides from discrete cosine exchange processing means, the quantification means that quantized; And the image data that quantizes carried out information encoded entropy coding means.

8, a kind ofly utilize the dynamic meter of generation to quantize/method and the device of inverse quantization, it is characterized in that: it is provided with one from the quantification means, the prediction error code is carried out the inverse quantization means of inverse quantization by quantizing step-length; Be provided with from the inverse quantization means, information conversion is become anti-discrete cosine, so as to generating the discrete cosine transform processing means that a regeneration rub-out signal is used; And be provided with utilization regeneration error message and information of forecasting, generate the anti-comentropy coding means that restores video.

Images