CN1286061C - Image quality precision gradual regulating method - Google Patents

Abstract

The present invention relates to an image quality refined gradual regulation method. Firstly, the gridded division of DCT coefficients of an image is carried out; the boundary of the number of grid reserved rows is determined according to the required image quality variation range, and the rough grade division of image quality is completed through the progressive increase or the descending of rows; in this way, the rational selection and the rational rejection of the DCT coefficients are carried out from low frequency components to high frequency components through the use of grid columns in each divided range, and thereby, the purpose of the further refining of image quality is achieved. The method has the advantages of simple software realization, simple hardware realization and flexible application, and is suitable for multimedia specially effect manufacture application field, such as movies, videos, slide shows, etc.

Description

Image quality precision gradual regulating method

Technical field

The invention belongs to digital image processing field, is a kind of method that can make the digital video image quality produce meticulous gradual change.

Background technology

Along with the arrival of multimedia era, people are more and more higher to the requirement of audio frequency and video mass effect.And video quality has wherein occupied bigger proportion.Usually, what people pursued is the visual effect of high definition, but in some special application scenarios, as film, television production, magic lantern etc., we then need digital picture is carried out some special processings, make it to produce indistinct, either overt or covert or the like artistic effect.Therefore, the high flexibility that realizes the video digital images quality changes the target that is the multimedia making person pursued, and the picture quality gradual change then is one of form of expression of changing of elasticity.

Digital picture can be represented in spatial domain or frequency field.In spatial domain, a width of cloth digital picture can be represented with RGB, also can represent or the like with YUV.In frequency field, be that unit carries out DCT (or class DCT) conversion with 8*8 or 4*4 piece normally with the spatial domain image.Heterogeneity after the conversion has different influences for the mass effect of video image, and low-frequency component has been concentrated most of energy of image, shows the framework of image, and radio-frequency component has then embodied the detail section of image.

When weighing the picture quality quality, at present more common method is weighed with Y-PSNR (PSNR), and it is defined as follows:

$\mathrm{PSNR}=10\mathrm{LO}{G}_{10}\frac{X*Y*{255}^2}{\mathrm{\Σ}{({\mathrm{PLX}}_S-{\mathrm{PLX}}_D)}^2},$ Formula-

Wherein, X, Y is the ranks number of pixels of representative image respectively, PLX _S, PLX _DThe value of representing the pixel of information source and processing back image respectively, the 255th, the maximal value that 8 bits of every pixel can be represented.

International standard in the past video coding, as H.261, H.263, H.264, among MPEG-1, MPEG-2, the MPEG-4, can adjust picture quality by changing the QP value, its method is with behind the residual signals dct transform behind the coding, the signal after being obtained quantizing by look-up table according to the QP value.Such as in H.264, the QP value can change between 0-51, the every increase by 1 of QP value, its PSNR with regard to corresponding decline a bit can realize the gradual change of picture quality like this, but needs corresponding look-up table, and the QP value is at a two field picture, is difficult for realizing the flexible adjustment of topography's quality.Thereby adjust picture quality with changing the QP value, have need look-up table, dirigibility not high, realize complicated shortcoming.

Thereby needing a kind of realization simple, the picture quality that can make of applying flexible produces meticulous gradual change method.

Summary of the invention

The object of the present invention is to provide a kind of image quality precision gradual regulating method, realize the meticulous gradual change of picture quality.It is simple that this method has realization, and advantage flexibly should be arranged.Be suitable for the occasion that needs produce the picture quality gradual change, as making in the application in multimedia special efficacys such as film, television production and magic lanterns.

For achieving the above object, the present invention adopts following technical scheme:

A kind of image quality precision gradual regulating method is characterized in that at first image DCT coefficient being carried out gridding divides; According to requirement picture quality variation range, determine that network keeps the bound of line number, by the increasing or decreasing of row, finish the coarse level of picture quality is divided; During the further refinement of quality, determine DCT coefficient brachymemma position; Change requirement by progressive precision, determine DCT coefficient brachymemma step-length; Concrete steps are as follows:

A) gridding of image DCT coefficient

The DCT coefficient is arranged from left to right by the order from the low frequency to the high frequency, again each coefficient is launched into bit from top to bottom by binary number, form grid;

B) the rough division of image quality level:

In grid, each provisional capital has different importance, and from bottom to up, importance rate rises successively, and the change of reservation line number can realize the rough division to picture quality.According to requirement mass change scope, adopt dichotomy to determine that grid keeps the bound of the number of plies;

C) the initial brachymemma of DCT coefficient position:

The DCT coefficient will change the lower limit that the DCT coefficient that cuts is determined in accuracy requirement by picture quality by reducing from low frequency to high frequency importance.The following formula condition should be satisfied in the initial brachymemma of DCT coefficient position:

PSNR _{(ROW，COLUMN_0)}＝PSNR _(ROW-1)+Q

The condition that the DCT coefficient does not cut:

|PSNR _(ROW)-PSNR _(ROW-1)|＜Q

PSNR _(ROW): when the reservation line number is ROW, the image Y-PSNR.

Q: institute's picture quality Y-PSNR that requires changes precision.

The initial brachymemma of COLUMN_0:DCT coefficient position.

D) exponential model of brachymemma step-length

Under the prerequisite of determining initial brachymemma position, adopt the index variation model, determine the brachymemma step-length of DCT coefficient.

Characteristics of the present invention and effect

The present invention proposes a kind of method that makes picture quality produce meticulous gradual change, and the DCT coefficient grid of requirement region of variation is carried out suitable brachymemma, has realized the picture quality variation of (or from carefully to thick) from coarse to fine.Have applying flexible, software and hardware is realized simple advantage.Be suitable for the occasion that needs produce the picture quality gradual change, as making in the application in multimedia special efficacys such as film, television production and magic lanterns.

Description of drawings:

Fig. 1 system realizes block diagram

The gridding of Fig. 2 DCT coefficient

Fig. 3 grid keeps determining of line number up-and-down boundary

The initial brachymemma of Fig. 4 DCT coefficient position

Fig. 5 DCT coefficient step-length is chosen

Concrete implementation step

Details are as follows for a preferred embodiment of the present invention: referring to Fig. 1, Fig. 2, Fig. 3, Fig. 4 and Fig. 5,1, the gridding of image DCT coefficient: referring to Fig. 1, is that unit carries out dct transform with original image with the 8*8 piece, 8*8 piece after the conversion according to " it " font scanning ordering, is formed the one-dimension array that 1*64 arranges from the low frequency to the high frequency.Again each element of this one-dimension array is launched from top to bottom by binary number, form grid, referring to Fig. 2.The line number of grid depends on the maximal value in this 64 number.As maximal value is 100, then 100=2 ⁶+ 2 ⁵+ 2 ², then have seven row, be followed successively by 1,1,0,0,1,0,0 from top to bottom.

2, the rough division of image quality level

According to the zone of require variation, demarcate corresponding 8*8 piece, and the PSNR bound of require variation is represented with B, A respectively.The total line number of hypothetical trellis is C, and the line number of Bao Liuing is ROW from top to bottom.The coboundary ROW that keeps line number so _OnWith the lower limit A correspondence of PSNR, keep the lower boundary ROW of line number _DownWith the upper limit B correspondence of PSNR referring to Fig. 3.

1) at first according to the definite coboundary ROW that keeps line number of A, B _OnWith lower boundary ROW _Down

Can determine ROW with dichotomy _On, step is as follows:

A) initialization.ROW _MAX＝C，R0W _MIN＝0；

B) calculate ROW _On, ROW _On=(ROW _MAX+ ROW _MIM)/2;

C) calculate this moment PSNR, with the PSNR and the A comparison of gained, if PSNR＞A forwards (d) to; If PSNR=A forwards (f) to; If PSNR＜A forwards (e) to;

D) ROW _MAX=ROW _On, forward (b) to;

E) ROW _MIN=ROW _On, forward (b) to;

F) finish.

In like manner, can obtain ROW according to the upper limit B of PSNR _Down, referring to Fig. 3.

2) coarse variations of picture quality realizes

When keeping line number is ROW _OnThe time, PSNR reaches minimum value A.Increase the grid line number line by line, so resultant PSNR is also progressively increasing, and is increased to ROW when keeping the grid line number _DownThe time, PSNR also reaches upper limit B.But can only realize the variation of the big step-length of PSNR like this, further realize the meticulous variation of quality, then need consider the brachymemma of DCT coefficient.

3, DCT coefficient brachymemma

Grid from left to right, the DCT coefficient is by the series arrangement from the low frequency to the high frequency, suitable brachymemma high frequency coefficient can be realized the meticulous gradual change of picture quality.Defining meticulous variation yardstick is Q, has realized the coarse variations of picture quality by back, and the quality fine setting that keeps between different line numbers will lean on brachymemma DCT coefficient to realize.

1) judges whether the DCT coefficient needs brachymemma

If | PSNR _(ROW)-PSNR _(ROW-1)|＜Q need not just can reach the mass change requirement by brachymemma DCT coefficient so.If | PSNR _(ROW)-PSNR _(ROW-1)|＞Q, can realize PSNR by brachymemma DCT coefficient so _(ROW), PSNR _(ROW-1)Between quality fine setting.Wherein, PSNR _(ROW)It is image Y-PSNR when the grid reservation number of plies is ROW.

2) the initial brachymemma of DCT coefficient position is determined

For the meticulous linking of the best quality that can reach with previous row, must determine the initial position of brachymemma.Making initial brachymemma position is COLUMN_0, referring to Fig. 4.Then COLUMN_O should meet the following conditions:

PSNR _{(ROW，COLUMN_0)}＝PSNR _(ROW-1)，+Q

The exponential model of 4, brachymemma step-length

The brachymemma position whenever pushes away further backward, has just in time realized the raising of PSNR.But because the heterogeneity of frequency for the contribution difference of PSNR, improves so will realize the linearity of PSNR, the brachymemma step-length can not be linear increasing, and uses the index variation model herein, referring to Fig. 5.The index law that is the brachymemma step-length changes, and can realize preferably that the linearity of PSNR increases.Available following formula is determined:

Step (n)=COLUMN_0*eq ^(n-1)0＜step (n)＜64, q are and the relevant constant of DCT coefficient distribution.Experiment showed, the variation that q=1/2 can well match brachymemma step-length, realize the linear increase of PSNR simultaneously.

The concrete operations step is as follows:

1) the selected zone that will change should all be set at 8 multiple by the zone length and width usually, and this zone is divided into the fritter of 8*8, be that unit carries out dct transform with the fritter.

2) the DCT coefficient after the conversion is scanned by " it " font, form the one-dimension array of 1*64, this array is by arranging from the low frequency coefficient to the high frequency coefficient.

3) the downward step-by-step power of this array relation is launched into binary number, and then forms grid.The row of grid is defined as ROW, and row are defined as COLUMN.

4) calculate the bound ROW of corresponding grid line number roughly according to border A, the B of requirement mass change _On, ROW _Down, adopt dichotomy to determine here.When the pairing PSNR of up-and-down boundary can't reach A, B accurately, get ROW _On, make it pairing PSNR and be slightly larger than A; Get ROW _Down, make it corresponding PSNR and be slightly larger than B, can accurately reach A, B by the brachymemma of DCT coefficient again.

5) determine the bound of grid line number by back, in this scope, increased the grid line number successively, can realize the coarse variations that picture quality increases progressively.

6) when each specific grid line number, the brachymemma position that all will progressively advance the DCT coefficient to be realizing the meticulous variation of picture quality, for the meticulous linking of the best quality that can reach with previous row, and initial position that must definite brachymemma.In a single day initial position is determined, then can obtain remaining brachymemma position successively with exponential relationship.

7) convert the DCT grid after the brachymemma to decimal system one-dimension array again, carry out contrary " it " font ordering, inverse dct transform obtains reconstruct 8*8 image block again.(5) are once carried out once in every like this brachymemma--and (7) circulation, can realize the picture quality gradual change.

Such system that can realize the meticulous variation of image just realized, more than all to increase progressively with picture quality be the example explanation; In like manner, meticulous the successively decreasing of picture quality can realize that too as long as (5), (6) step part omitted adjusts and gets final product.

The present invention repeatedly by my experimental verification, proves and can produce a desired effect, and occupying system resources is few, and the flexibility ratio height is realized simple.

Claims (1)

1. an image quality precision gradual regulating method is characterized in that at first image DCT coefficient being carried out gridding divides; According to requirement picture quality variation range, determine that grid keeps the bound of line number, by the increasing or decreasing of row, finish the coarse level of picture quality is divided; During the further refinement of quality, determine DCT coefficient brachymemma position; Change requirement by progressive precision, determine DCT coefficient brachymemma step-length; Concrete steps are as follows:

A) gridding of image DCT coefficient:

The DCT coefficient is arranged from left to right by the order from the low frequency to the high frequency, again each coefficient is launched into bit from top to bottom by binary number, form grid;

B) the rough division of image quality level:

In grid, each provisional capital has different importance, from bottom to up, importance rate rises successively, the change of reservation line number can realize the rough division to picture quality, according to requirement mass change scope, adopts dichotomy to determine that grid keeps the bound of line number;

C) the initial brachymemma of DCT coefficient position:

The DCT coefficient will change the lower limit that the DCT coefficient that cuts is determined in accuracy requirement by picture quality by reducing from low frequency to high frequency importance; The following formula condition should be satisfied in the initial brachymemma of DCT coefficient position:

PSNR _{(ROW，COLUMN_0)}＝PSNR _(ROW-1)+Q

The condition that the DCT coefficient does not cut:

|PSNR _(ROW)-PSNR _(ROW-1)|＜Q

PSNR _(ROW): when the reservation line number is ROW, the image Y-PSNR;

Q: institute's picture quality Y-PSNR that requires changes precision;

The initial brachymemma of COLUMN_0:DCT coefficient position;

D) exponential model of brachymemma step-length:

Under the prerequisite of determining initial brachymemma position, adopt the index variation model, determine the brachymemma step-length of DCT coefficient.

Images