CN106934768A

CN106934768A - A kind of method and device of image denoising

Info

Publication number: CN106934768A
Application number: CN201511033689.9A
Authority: CN
Inventors: 王微
Original assignee: Spreadtrum Communications Tianjin Co Ltd
Current assignee: Spreadtrum Communications Tianjin Co Ltd
Priority date: 2015-12-30
Filing date: 2015-12-30
Publication date: 2017-07-07
Anticipated expiration: 2035-12-30
Also published as: CN111915534B; CN111915534A; CN106934768B; CN111915535A; CN111915535B

Abstract

A kind of method and device of image denoising, recursive filtering is carried out to the pixel in described image, adopts the filtering strength for obtaining currently pending pixel with the following method：The marginal information on the chromatic component both horizontally and vertically of center pixel is calculated, maximum is taken, as first edge information maximization value；Calculate the variance of described image block；Calculate the first difference maximum of the center pixel and the filtered pixel selected in the row of kth -1 in described image block/row of kth -1；Calculate the marginal information on the luminance component of the center pixel；Judge whether the frequency field residing for the center pixel is flat site；When the frequency field residing for the center pixel is flat site, the present intensity of the center pixel is calculated；According to the present intensity, the first filtering strength is obtained, as the filtering strength of the center pixel.The conditions of streaking for avoiding image after denoising can be taken into account while the denoising effect of noise is ensured using such scheme.

Description

A kind of method and device of image denoising

Technical field

The present invention relates to image processing field, more particularly to a kind of method and device of image denoising.

Background technology

Whether digital camera, mobile phone camera, or computer camera, are carrying out the mistake of image collection Cheng Zhong, more or less can be disturbed by color noise.In general, in order to be carried out to color noise Suppress, most common method is exactly to design wave filter to filter it.And it is directed to the high frequency of color noise Characteristic, it will usually design a series of low pass filter to suppress it.This kind of wave filter uses one The template of sizing.The color noise relatively low in order to effectively remove frequency, can usually use larger template, And recursion filter (Infinite Impulse Response, IIR) is commonly used to its distinctive transmission characteristic Simulate the noise removal function compared with large form.

In order to filter color noise, at present, during using IIR, denoising is carried out using following steps：According to figure The center pixel and upper one of level, the marginal information of vertical direction, variance and described image block as block The difference of the chromatic component of the pixel selected in row or column judges whether described image block is flat region Domain.When described image block is flat site, according in the brightness of image block, and described image block The difference of the imago element chromatic component of neighboring pixel therewith determines the filtering strength to described image block, Then described image block is filtered using the filtering strength.

But, image denoising treatment being carried out using the above method, if filtering strength is excessively weak, may lead It is poor to the denoising effect of noise to cause, or if filtering strength is too strong, image goes out after may result in denoising Existing conditions of streaking.

The content of the invention

The problem that the present invention is solved is how while the denoising effect of noise is ensured, to take into account and avoid denoising The conditions of streaking of image afterwards.

To solve the above problems, the embodiment of the invention provides and a kind of method of image denoising is provided, to institute The pixel stated in image carries out recursive filtering, adopt obtain with the following method currently pending pixel filtering it is strong Degree, including：

Obtain using the pending pixel as center pixel image block, the center pixel is located at described The row k kth row of image block；

The marginal information on the chromatic component both horizontally and vertically of the center pixel is calculated, and takes institute The maximum of the marginal information on chromatic component both horizontally and vertically is stated, as first edge information most Big value；

Calculate the variance of described image block；

After calculating the filtering in the center pixel and described image block selected in the row of kth -1/row of kth -1 Pixel difference maximum, as the first difference maximum；

Calculate the marginal information on the luminance component of the center pixel；

According to the first edge information maximization value, the variance of described image block, the first difference maximum and Marginal information on the luminance component of the center pixel, judges the frequency field residing for the center pixel Whether it is flat site；

When the frequency field residing for the center pixel is flat site, working as the center pixel is calculated Preceding brightness；

According to the present intensity of the center pixel, the first filtering strength is obtained, as the center pixel Filtering strength.

Alternatively, it is described according to the first edge information maximization value, the variance of described image block, first Marginal information on the luminance component of difference maximum and the center pixel, judges the center pixel institute Whether the frequency field at place is flat site, including：

When following all conditions are met, the frequency for determining the center pixel is flat site：

The first edge information maximization value is less than default first threshold；

The variance is less than default Second Threshold；

The first difference maximum is less than default 3rd threshold value；

Marginal information on the luminance component of the center pixel is less than default 4th threshold value.

Alternatively, the value of the first threshold, Second Threshold, the 3rd threshold value and the 4th threshold value is with institute Brightness section is different and different residing for the present intensity of the center pixel for stating image block.

Alternatively, methods described also includes：

When the present intensity of the center pixel is less than default dark space luminance threshold, the center is determined Pixel be in dark space, and to the recursive filtering after the center pixel carry out at the saturation degree suppression of dark space Reason.

Alternatively, it is described to the recursive filtering after the center pixel carry out at the saturation degree suppression of dark space Reason, including：

Judge the component of the center pixel after the recursive filtering whether more than described before recursive filtering The component of center pixel；

Described middle imago before the component of the filtered center pixel is more than the recursive filtering Element component when, using the component of the center pixel before recursive filtering as the recursive filtering after The component of the center pixel.

Alternatively, it is described to the recursive filtering after the center pixel carry out at the saturation degree suppression of dark space Reason, also includes：

The center pixel before by the component of the center pixel after the recursive filtering, recursive filtering point Amount takes difference with default 5th threshold value respectively, obtains the first difference and the second difference；

Judge first difference with the product of the second difference whether less than zero；

When the product of first difference and the second difference is less than zero, the 5th threshold value is passed as described Return the component of the filtered center pixel.

Alternatively, the 5th threshold value is 128.

Alternatively, methods described also includes：

When the marginal information on the luminance component of the center pixel is less than four threshold value, institute is judged Whether the marginal information on the luminance component of center pixel is stated more than default first edge threshold value；

When the marginal information on the luminance component of the center pixel is more than the first edge threshold value, root According to the marginal information on the luminance component of the center pixel, first filtering strength is modified, Obtain the second filtering strength；

By second filtering strength, as the filtering strength of the center pixel.

Alternatively, methods described also includes：

When the first difference maximum is more than default 6th threshold value and is less than default 7th threshold value, and When 7th threshold value is less than three threshold value；

According to the first difference maximum, second filtering strength is modified, obtains the 3rd filter Intensity of wave, using the 3rd filtering strength as the center pixel filtering strength.

Alternatively, methods described also includes：

Before judging whether the frequency field residing for the frequency of the center pixel be flat site, to institute First row/column pixel of the image block of acquisition carries out noise-removed filtering treatment.

Alternatively, it is corresponding when the present intensity of the center pixel is less than default eight threshold value Filtering strength is reduced as the present intensity is reduced.

Alternatively, methods described also includes：

When the frequency field non-planar regions residing for the center pixel, the center pixel is carried out Value filtering.

A kind of device of image denoising is the embodiment of the invention provides, described device includes：

Image block acquiring unit, be suitable to obtain using pending pixel as center pixel image block, it is described The row k kth that center pixel is located at described image block is arranged；

First computing unit, on the chromatic component both horizontally and vertically for being suitable to calculate the center pixel Marginal information, and take the maximum of marginal information on the chromatic component both horizontally and vertically, As first edge information maximization value；

Second computing unit, is suitable to calculate the variance of described image block；

3rd computing unit, is suitable to calculate the center pixel with the row of kth -1/kth -1 in described image block The difference maximum of the filtered pixel selected in row, as the first difference maximum；

4th computing unit, is suitable to calculate the marginal information on the luminance component of the center pixel；

First judging unit, be suitable to according to the first edge information maximization value, the variance of described image block, Marginal information on the luminance component of the first difference maximum and the center pixel, judges the middle imago Whether the frequency field residing for element is flat site；

5th computing unit, is suitable to when the frequency field residing for the center pixel is flat site, meter Calculate the present intensity of the center pixel；

Filtering strength acquiring unit, is suitable to the present intensity according to the center pixel, obtains first and filters Intensity, as the filtering strength of the center pixel；

Filter unit, is suitable to, using the filtering strength of the center pixel, pass the center pixel Return filtering.

Alternatively, first judging unit, is suitable to when following all conditions are met, determine it is described in The frequency of imago element is flat site：

The variance is less than default Second Threshold；

The first difference maximum is less than default 3rd threshold value；

Alternatively, the value of the first threshold, Second Threshold, the 3rd threshold value and the 4th threshold value is with institute State the different and different of brightness section residing for the present intensity of center pixel.

Alternatively, described device also includes：

Saturation degree processing unit, is suitable to be less than default dark space brightness when the present intensity of the center pixel During threshold value, determine the center pixel be in dark space, and to the recursive filtering after the center pixel Carry out dark space saturation degree suppression treatment.

Alternatively, the saturation degree processing unit, including：

First judgment sub-unit, be suitable to judge the center pixel after the recursive filtering component whether The component of the center pixel before more than recursive filtering；

First computation subunit, is suitable to institute after first judgment sub-unit determines the recursive filtering When the component for stating center pixel is more than the component of the center pixel before the recursive filtering, using recurrence The component of the center pixel before filtering as the center pixel after the recursive filtering component.

Alternatively, the saturation degree processing unit, including：

Second computation subunit, is suitable to the component of the center pixel after the recursive filtering, recursive filtering The component of the preceding center pixel takes difference with default 5th threshold value respectively, obtains the first difference and second Difference；

Whether second judgment sub-unit, be suitable to judge the product of first difference and the second difference less than zero；

3rd computation subunit, is suitable to determine first difference and second when second judgment sub-unit When the product of difference is less than zero, using the 5th threshold value as the center pixel after the recursive filtering Component.

Alternatively, the 5th threshold value is 128.

Alternatively, described device also includes：

Second judging unit, is suitable to when the marginal information on the luminance component of the center pixel is less than described During four threshold values, judge marginal information on the luminance component of the center pixel whether more than default the One edge threshold；

First amending unit, is suitable to when the marginal information on the luminance component of the center pixel is more than described During first edge threshold value, the marginal information on luminance component according to the center pixel, to described first Filtering strength is modified, and obtains the second filtering strength, by second filtering strength, in described The filtering strength of imago element.

Alternatively, described device also includes：

3rd judging unit, is suitable to judge whether to meet following condition：The first difference maximum is more than Default 6th threshold value is simultaneously less than default 7th threshold value, and the 7th threshold value is less than the 3rd threshold value；

Second amending unit, is suitable to when the first difference maximum more than default 6th threshold value less than pre- If the 7th threshold value, and the 7th threshold value be less than three threshold value when, according to first difference most Big value, is modified to second filtering strength, obtains the 3rd filtering strength, by the described 3rd filtering Intensity as the center pixel filtering strength.

Alternatively, described device also includes：

Noise-removed filtering unit, is suitable to judging whether the frequency field residing for the center pixel is flat region Before domain, the first row/column pixel to acquired image block carries out noise-removed filtering treatment.

Alternatively, described device also includes：

Mean filter unit, is suitable to when the frequency field non-planar regions residing for the center pixel, right The center pixel carries out mean filter.

Compared with prior art, technical scheme has advantages below：

By when whether judge the frequency field residing for center pixel is flat site, more than considering One edge information maximization value, the variance of the center pixel and the first difference maximum, it is also contemplated that described Marginal information on the luminance component of center pixel, and the marginal information on luminance component is more accurate, therefore Can make the frequency region information of the center pixel of acquisition can be more accurate, such that it is able to in described Imago element determines more accurate filtering strength, therefore can cause that filtered image retains more details, While the denoising effect of noise is ensured, the conditions of streaking for avoiding image after denoising is taken into account.

Because at different brightnesses, human eye is different to the sensitivity of color distortion, accordingly by root According to the center pixel of described image block present intensity residing for brightness section, adjustment first threshold, the second threshold Value, the 3rd threshold value and the value of the 4th threshold value, can avoid the color noise in flat site from being mistaken for Color boundaries, or color boundaries are mistaken for flat, therefore can further avoid dragging for image after denoising Tail phenomenon.

Further, by when the center pixel present intensity be less than default dark space luminance threshold when, Determine the center pixel be in dark space, to the recursive filtering after the center pixel carry out dark space satisfy Processed with degree suppression so that in especially dark place, even if color boundaries are not detected at, it is also possible to keep away Remove the conditions of streaking of image after making an uproar from.

Further, when the marginal information on the luminance component of the center pixel is more than the first edge threshold During value, the marginal information on luminance component according to the center pixel is entered to first filtering strength Row amendment, can avoid filtered image from lamination occur in some regions, such that it is able to further Improve picture quality.

Further, when the first difference maximum is less than default 7th threshold more than default 6th threshold value Value, and the 7th threshold value be less than three threshold value when, by the first difference maximum, to institute State the second filtering strength to be modified, filtered image can be avoided lamination occur in some regions, Such that it is able to further improve picture quality.

Brief description of the drawings

Fig. 1 is a kind of schematic flow sheet of the method for the image denoising in the embodiment of the present invention；

Fig. 2 is a kind of schematic diagram of iir filter in the prior art；

Fig. 3 is the schematic flow sheet of the method for another image denoising in the embodiment of the present invention；

Fig. 4 is a kind of U components of the expression center pixel in the embodiment of the present invention, and N is 5 image block；

Fig. 5 be a kind of size in the embodiment of the present invention for 3 × 3 image block；

Fig. 6 be another size in the embodiment of the present invention for 3 × 3 image block；

Fig. 7 be another size in the embodiment of the present invention for 3 × 3 image block；

Fig. 8 be another size in the embodiment of the present invention for 3 × 3 image block；

Fig. 9 is the relation curve of a kind of threshold value in the embodiment of the present invention and brightness；

Figure 10 is that a kind of filtering strength in the embodiment of the present invention is bent with the relation of the present intensity of center pixel Line；

Figure 11 is marginal information edge and the filter of a kind of luminance component of the center pixel in the embodiment of the present invention The relation curve of intensity of wave；

Figure 12 is the relation curve of a kind of first difference maximum in the embodiment of the present invention and filtering strength；

Figure 13 is a kind of structural representation of the device of the image denoising in the embodiment of the present invention；

Figure 14 is the structural representation of the device of another image denoising in the embodiment of the present invention；

Figure 15 is the structural representation of the device of another image denoising in the embodiment of the present invention.

Specific embodiment

In order to filter color noise, at present, during using IIR, denoising is carried out by following steps：According to figure The center pixel and upper one of level, the marginal information of vertical direction, variance and described image block as block The difference of the chromatic component of the pixel selected in row or column judges whether described image block is flat region Domain.When described image block is flat site, according in the brightness of image block, and described image block The difference of the imago element chromatic component of neighboring pixel therewith determines the filtering strength to described image block, Then described image block is filtered using the filtering strength.

But, image denoising treatment is carried out using the above method, if filtering strength control is bad, going The phenomenon of hangover can be produced while except noise.By taking color noise as an example, this hangover shows as image Middle color is down dragged, and as having grown tail, is allowed originally and is not had coloured place to there has also been color, is influenceed Picture quality.Particularly in dark space, if filtering strength is too strong, IIR downward transitivity can become strong, then Hangover can be produced；And if filtering strength is excessively weak, downward transitivity dies down, though hangover will not be produced, But it is clean that noise is difficult to removal.

In order to solve the above problems, the method that the embodiment of the invention provides image denoising, due to brightness point Marginal information in amount can be relatively more accurate, and methods described is by judging the frequency field residing for center pixel When whether being flat site, first edge information maximization value, the variance of the center pixel are more than considered And the first difference maximum, it is also contemplated that the marginal information on the luminance component of the center pixel, and it is bright Marginal information on degree component is more accurate, therefore can believe the frequency field of the center pixel of acquisition Breath can be more accurate, such that it is able to determine more accurate filtering strength for the center pixel, therefore can To cause that filtered image retains more details, while the denoising effect of noise is ensured, take into account and keep away Remove the conditions of streaking of image after making an uproar from.

It is understandable to enable the above objects, features and advantages of the present invention to become apparent, below in conjunction with the accompanying drawings Specific embodiment of the invention is described in detail.

The method for the following providing a kind of image denoising in the embodiment of the present invention, as shown in figure 1, below Methods described is specifically introduced with reference to Fig. 1：

S11：Obtain using the pending pixel as center pixel image block.

In specific implementation, due to the pixel in adjacent domain, the pixel value that can be interacted to each other, Therefore can obtain using the pending pixel as center pixel image block, to obtain on the center Pixel more accurately information.For ease of description, the center pixel can be set positioned at described image block Row k kth is arranged.

In an embodiment of the present invention, denoising can be carried out to the first row/column pixel of acquired image block Filtering process, then reuses filtered image block, performs S12.So, in larger filter Under intensity of wave a, if the pixel of the first row/column has larger color noise, would not be in the mistake of downward transmission The color in image boundary is caused to be trailed in journey.

S12：The marginal information on the chromatic component both horizontally and vertically of the center pixel is calculated, and Take the maximum of marginal information on the chromatic component both horizontally and vertically, the center pixel Variance, the center pixel and in described image block the row of kth -1 or kth -1 row selected in filtering after The difference maximum of pixel and the luminance component of the center pixel on marginal information.

For purposes of illustration only, can be by the maximum of the marginal information both horizontally and vertically, referred to as First edge information maximization value, can be by the row of kth -1 in the center pixel and described image block or the The difference maximum of the filtered pixel selected in k-1 row, referred to as the first difference maximum.

In order that obtain filtered image detail more being retained, the center pixel can be detected Marginal information.In an embodiment of the present invention, can be by calculating the horizontal and vertical of the center pixel Marginal information on the chromatic component in direction, and take the side on the chromatic component both horizontally and vertically The maximum of edge information, the variance of the center pixel, the center pixel and kth -1 in described image block The brightness point of the difference maximum and the center pixel of the filtered pixel selected in the row of row/kth -1 Marginal information in amount, and the edge of the center pixel is known according to the result that these are calculated Information.

In an embodiment of the present invention, the iir filter of selection vertical direction, therefore selected Filtered Picture Element can be the pixel of the lastrow of the center pixel.

In specific implementation, when by judging to determine the first edge information maximization value less than default the One threshold value, and the variance is less than default Second Threshold, and the first difference maximum less than default The 3rd threshold value, and marginal information on the luminance component of the center pixel is less than default 4th threshold value When, it may be determined that the frequency of the center pixel of the input is flat site.

Because under different brightness, human eye is different to the sensitivity of color distortion.Brightness is smaller, human eye It is just very sensitive to less color distortion；And brightness is bigger, human eye can be more sensitive to big colour-difference exceptional talents. Therefore if all setting same first to fourth threshold value in all of brightness, then when setting, if threshold Value is too high, then in dark region, inspection does not measure real color boundaries；If threshold value is too low, brighter Region, the color noise in flat site can be misjudged into color boundaries.And if by flat site Color noise misjudges into color boundaries, and color noise can be remained in the picture；And if color boundaries missed Sentence into flat, filtering strength is excessive, and color " hangover " occurs in the region in causing image.Therefore, at this Invent in an embodiment, the first threshold, Second Threshold, the 3rd threshold value and the 4th threshold value can be set Value, with the center pixel of described image block present intensity residing for brightness section it is different and different.

S13：It is maximum according to the first edge information maximization value, the variance of described image block, the first difference Marginal information on the luminance component of value and the center pixel, judges the frequency residing for the center pixel Whether region is flat site.

In specific implementation, because the end value for calculating can to a certain degree reflect the center Frequency field residing for pixel, therefore can be according to the first edge information maximization value, described image block Marginal information on the luminance component of variance, the first difference maximum and the center pixel, judges described Whether the frequency field residing for center pixel is flat site.Wherein, the luminance component of the center pixel On marginal information it is often more accurate, as sentencing for the frequency field described in the center pixel One of disconnected factor, can know the affiliated result of more accurate frequency field.

When the frequency field residing for the center pixel is flat site, S14 is performed, conversely, performing S16。

S14：Calculate the present intensity of the center pixel.

Under different brightness cases, the color noise distribution on image block is also different.Therefore in specific implementation, When the frequency field residing for the center pixel is flat site, working as the center pixel can be calculated Preceding brightness, so, can be different according to present intensity, using different noise processed parameters.

S15：According to the present intensity of the center pixel, the first filtering strength is obtained, as the center The filtering strength of pixel, recursive filtering is carried out to the center pixel.

Because brightness can influence distribution of the noise in image block, and at different brightnesses, human eye pair The sensitivity of color distortion is different, therefore can obtain first according to the present intensity of the center pixel Filtering strength, as the filtering strength of the center pixel, and utilizes first filtering strength, to institute Stating center pixel carries out recursive filtering.

In general, in the dark place of brightness ratio, it is easier to color noise occur, therefore when the center When the present intensity of pixel is less than default eight threshold value, the corresponding filtering of the present intensity can be set Intensity is reduced as the present intensity lowers.Such that it is able to avoid in especially dark region, there is figure As conditions of streaking.

Due to the influence of noise, if two neighboring pixel, one has been judged into flat site, one It has been judged to, into fringe region, filtered image can be caused in some regions, such as face and hair , there is lamination in intersection.In order to avoid this lamination, in an embodiment of the present invention, when When marginal information on the luminance component of the center pixel is less than four threshold value, it can be determined that described Whether the marginal information on the luminance component of center pixel is more than default first edge threshold value, if described When marginal information on the luminance component of center pixel is more than the first edge threshold value, then according in described Marginal information on the luminance component of imago element, is modified to first filtering strength, obtains second Filtering strength, then by second filtering strength, as the filtering strength of the center pixel, and root The center pixel is filtered according to second filtering strength.

In order to further suppress lamination, in an alternative embodiment of the invention, when first difference Maximum is less than default 7th threshold value, and the 7th threshold value less than described more than default 6th threshold value During three threshold values, second filtering strength can also be repaiied according to the first difference maximum Just, obtain the 3rd filtering strength, using the 3rd filtering strength as the center pixel filtering strength, Then the center pixel is filtered using the 3rd filtering strength.

In order to prevent streaking phenomenon in the dark, in an embodiment of the present invention, when the middle imago When the present intensity of element is less than default dark space luminance threshold, it may be determined that the center pixel is in dark space, And to the recursive filtering after the center pixel carry out dark space saturation degree suppression treatment.

Specifically, the first step：May determine that the component of the center pixel after the recursive filtering is It is no more than recursive filtering before the center pixel component, when the filtered center pixel When component is more than the component of the center pixel before the recursive filtering, described in before recursive filtering The component of center pixel can thus be protected as the component of the center pixel after the recursive filtering The saturation degree for demonstrate,proving filtered center pixel is not higher than the saturation degree of the center pixel before filtering.

Second step：Can be by described in before component, the recursive filtering of the center pixel after the recursive filtering The component of center pixel takes difference with default 5th threshold value respectively, obtains the first difference and the second difference, connects Whether the product for judging first difference and the second difference less than zero, if the product is less than zero, Can using the 5th threshold value as the center pixel after the recursive filtering component, by this Step, it is possible to ensure the colourity of filtered center pixel without reverse.In sum, by this two step Saturation degree suppression treatment, in especially dark place, even if color boundaries are not detected at, will not also send out Raw streaking phenomenon.

In an embodiment of the present invention, the component of the center pixel can be colourity U components, described the Five threshold values can be 128.

It is understood that for colourity V component and luminance Y component, it would however also be possible to employ similar side Method is processed accordingly, be will not be repeated here.

S16：Mean filter is carried out to the center pixel.

In specific implementation, if the frequency field residing for the center pixel is not flat site, can Mean filter is carried out with to the center pixel, such that it is able to avoid the color to the center pixel in itself Information causes to damage.

Explanation is needed, the method for image denoising involved in the present invention goes for face in yuv space The removal of coloured noise, it can also be used to the removal of the noise of brightness Y, is also applied for other color spaces such as Lab The denoising in space, for the removal of color noise in yuv space, is not only only applicable to colourity U components Denoising, apply also for the denoising of colourity V component.And the direction of IIR filtering is not also to the present invention Any limitation is constituted, those skilled in the art according to actual needs, can select the IIR filtering of vertical direction, The IIR of horizontal direction can also be selected to filter.In order to those skilled in the art more fully understand and realize this hair It is bright, when herein all to be filtered using the IIR of vertical direction, the side of the noise on removal colourity U components The implementation steps of method described image denoising as an example.The IIR filtering modes in other directions, Yi Jiqi Noise remove mode on its component, may be referred to the embodiment that this text provides and is implemented, herein not Repeat one by one again.

In general, the larger template of the color noise needs of low frequency could be removed, and large form corresponds to Complicated calculating, and hardware costs higher.However, IIR filtering is with its distinctive transmission characteristic, letter Single quickly to calculate, the cost of low cost can simulate the noise removal function of large form, can effectively remove Low frequency color noise.One typical iir filter as shown in Fig. 2 such as representing Fig. 2 with formula in show The operation relation for going out, then for：Y (n)=a*y (n-1)+(1-a) * x (n), x (n) represent current former Beginning pixel, y (n) represents the pixel after current filter, and y (n-1) represents previous filtered pixel, and a is represented Filtering strength.The transitivity of IIR filtering shows：Y (n) depends on y (n-1) ... y (n-N), i.e. x's (n) is defeated Go out to depend on x (n-1) ... ..x (n-N), wherein N<n.

To cause that those skilled in the art more fully understand and realize the present invention, the present invention also provided below The schematic flow sheet of the method for another image denoising in embodiment, may be referred to Fig. 3, methods described tool Body can be comprised the step of：

S31：Image block to being input into carries out noise pre-filtering.

By taking the iir filter of vertical direction as an example, it can be seen from introduction of the above to the iir filter, In the picture, the output of the pixel of the second row is largely dependent upon the value of the first row pixel, and the 3rd Output in row pixel is largely dependent upon the value of the second row pixel, is also to be largely dependent upon The value of the first row pixel.Therefore, if the pixel value of the first row has larger color noise, if filtering is strong A is bigger for degree, can more cause the color in image boundary to be trailed during transmission downwards.And one As for, particularly under dark environment, the brightness value in image boundary is lower, and color noise can be more Greatly, if filtering strength is too small, color noise removes unclean, once filtering strength is excessive, image boundary On color hangover will be more serious.

Therefore in an embodiment of the present invention, the face that template size is 1*N can be carried out to the pixel of the first row Coloured noise pre-filtering, with the U components of center pixel illustrated in fig. 4, as a example by N is 5 image block, Can be filtered with following formula, the formula is as follows：

Wherein：U0i meets：Abs (U0i-U02) ＜ thr10, i.e. U0i is to meet the image with input The absolute difference of the colourity U components U02 of the center pixel of block is less than default tenth threshold value thr10's Pixel, andI.e. M is：In the image block of input The absolute difference of the colourity U components U02 of imago element is less than the individual of the pixel of default tenth threshold value thr10 Number.

Then saturation degree suppression can be carried out to the filtered center pixel U02 ' using formula below Treatment：

U02 "=U02 ' * ratio；

Wherein, ratio is saturation degree inhibiting factor, and ratio is bigger, and saturation degree suppresses weaker；Conversely, Saturation degree suppresses stronger.

After simple filtering and saturation degree suppress, the color noise in image the first row pixel can weaken Many, can greatly suppress the color conditions of streaking in image boundary.

S32：Rim detection is carried out to center pixel, judges the center pixel whether in flat site.

In order that filtered image details retains, in specific implementation, colourity U components can be carried out Rim detection, is only filtered just now on the ground for being detected as flat site.Can specifically utilize n × m's Template comes the edge in calculated level and vertical the two directions.In an embodiment of the present invention, shown with Fig. 5 As a example by the template of 3 × 3 for going out, the center pixel is U11, on the chromatic component in its calculated level direction Marginal information h and vertical direction chromatic component on marginal information v calculation such as formula (1) And under (2)：

H=| U00+U01+U02-U20-U21-U22 | (1)

V=| U00+U10+U20-U02-U12-U22 | (2)

Then a maximum is selected from both direction value, as first edge information maximization value max_direc=max (h, v), it is described if this first edge information maximization value is less than first threshold th1 Central pixel point may be in flat site.

In an embodiment of the present invention, by taking shown in Fig. 53 × 3 template as an example, the image block is calculated Variance, computing formula is such as under (3)：

If this variance is not less than the Second Threshold th2, it is believed that the center pixel is not in flat region In domain.

Because final filter result can be got by weighted average calculation, the filtering of the pixel of lastrow End value can cause very big influence to the pixel of one's own profession afterwards, therefore in an embodiment of the present invention, can be with Image block according to Fig. 6 calculates the first difference maximum of the center pixel U11, Wherein, U00 ' represents the filtered values of U00, and U01 ' represents the filtered values of U01, and U02 ' represents U02 Filtered value, in an embodiment of the present invention, above three points U00 ', U01 ', U02 ' can be endowed Larger weight.The first difference maximum, it is possible to use formula (4) is calculated to (7), i.e.,：

Max_u '=max (U00 ', U01 ', U02 ') (4)

Min_u '=min (U00 ', U01 ', U02 ') (5)

u_diff1=| max_u '-u11 |, u_diff2=| min_u '-u11 | (6)

Umax_diff=max (u_diff1,u_diff2) (7)

If the first difference maximum Umax_diff is less than the 3rd threshold value th3, then it is assumed that in described Imago have may be in flat site.

It can be seen that, three extractions of marginal information of the above are carried out in colourity U planes, and are typically come Say, the marginal information on brightness Y plane is often more accurate, be also to be worth reference.Therefore in this hair In a bright embodiment, the edge letter on the luminance component of the center pixel that can calculate brightness Y plane Breath.Specifically can be by taking the image block shown in Fig. 7 as an example, the corresponding luminance component of the center pixel is Y11.

Before marginal information on the luminance component for calculating the center pixel, in order to reduce noise opposite side The influence that edge is extracted, can simply be filtered to 3 × 3 luminance component shown in Fig. 7 first, Filtering mode can be with varied, and specific filtering mode does not constitute any limitation to the present invention, herein By taking simplest mean filter as an example, the luminance Y component of the center pixel is carried out using formula (8) Filtering：

It is understood that other pixels can also be filtered using same method, its is obtained right The filtered luminance component answered, this is no longer going to repeat them, the image block in Fig. 7 after filtering after, The image block shown in Fig. 8 can be generated, then according to the image block shown in described Fig. 8, using formula (9) The marginal information edge on the luminance component of the center pixel is calculated to (11)：

edge_x=| Y00 '+Y01 '+Y02 '-Y20 '-Y21 '-Y22 ' | (9)

edge_y=| Y00 '+Y10 '+Y20 '-Y02 '-Y12 '-Y22 ' | (10)

Edge=edge_x*edge_x+edge_y*edge_y (11)

If the marginal information edge on the luminance component of this center pixel is not less than the 4th threshold Value y_thr, then think the center pixel corresponding to it not in flat site.

Understand in sum, if conditions above is satisfied by, i.e., described first edge information maximization value max_direc<The first threshold th1, the variance u of described image block_var<The Second Threshold th2, institute State the first difference maximum Umax_diff<The 3rd threshold value th3, on the luminance component of the center pixel Marginal information edge<The 4th threshold value y_thr, it may be determined that the center corresponding to four threshold values Pixel is in flat site, and it can be filtered.

Because under different brightness, human eye is different to the sensitivity of color distortion.Brightness is smaller, human eye It is all very sensitive to less color distortion；And brightness is bigger, human eye can be sensitive to larger colour-difference exceptional talents. If all setting same threshold value in all of brightness, threshold value is too high, then in dark region, detection Real color boundaries are not gone out；Threshold value is too low, in brighter region, then can be the color in flat site Noise misjudges into color boundaries.If the color noise of flat site is misjudged into color boundaries, color Noise can be remained in the picture；And if color boundaries are judged by accident into it is flat, filtering strength again it is larger, then There is color hangover in the region during image can be caused.

Therefore in an embodiment of the present invention, in order to solve color hangover in this case, can be according to Brightness changes to set first to fourth above-mentioned threshold value.Briefly, when brightness is dark, can be by It is smaller that threshold value is set, to detect real color boundaries；When brightness is larger, color noise is smaller, Can without being filtered, therefore can by threshold value set it is also smaller；When in intermediate luminance, threshold value Can set larger, to prevent the color noise by flat site from misjudging into color boundaries.

By substantial amounts of practice and it is demonstrated experimentally that in an embodiment of the present invention, it is possible to use Fig. 9 shows A kind of Chroma threshold and brightness curve, present intensity according to the center pixel sets described first To the 4th threshold value.Curve in Fig. 9 gives a kind of Chroma threshold segmentation method, and Chroma threshold is divided into Multistage, and by Chroma threshold as brightness change is also divided into multistage, wherein y_curRepresent working as center pixel Preceding brightness, for ease of understanding, if the brightness y of the center pixel_curPosition in fig .9, then correspond to Chroma threshold u_thrCalculation such as formula (12) shown in：

It should be noted that the parameter Y in Fig. 9_thr1、Y_thr2、Y_thr3、Y_thr4、u_thr1And u_thr2 According to actual needs, voluntarily to be set by those skilled in the art.And first to fourth above-mentioned threshold value The all applicable above-mentioned Fig. 9 of setting shown in curve, only for each threshold value selection when, the horizontal stroke The parameter of axle and the longitudinal axis voluntarily changes, such as such as to set first threshold, then the value of the longitudinal axis is all First threshold parameter, certainly, those skilled in the art can also according to actual needs, by the curve Transverse axis or the longitudinal axis are divided into multistage, and specific segmentation method is not limited the invention.As long as can be anti- Reflect the mapping relations of the parameter and brightness.

When the center pixel is in flat site, S33 is performed；Conversely, then performing S39.

S33：The present intensity of the center pixel is calculated, according to the present intensity of the center pixel, is obtained The first filtering strength is taken, as the filtering strength of the center pixel.

In general, it is easier color noise occur in the dark place of brightness ratio, thus it is real in the present invention one Apply in example, the size of filtering strength is determined by brightness.Within a certain range, brightness is smaller, and intensity is got over By force；After more than this scope, intensity keeps constant.Figure 10 shows the one kind in the embodiment of the present invention The relation schematic diagram that filtering strength changes with the change of brightness, according to the present intensity of the center pixel, First filtering strength a can be obtained according to Figure 10_y, relation shown in Figure 10 also can be by equation below (13) Represent：

Wherein：Luminance parameter y_th, filtering strength parameter a1 and a2 can be carried out according to actual needs Relative set.In order to prevent streaking phenomenon, for especially dark region, filtering strength can be weakened, That is, when the present intensity of the center pixel is less than the 8th threshold value Y_min_th, institute is right The filtering strength answered is reduced as the present intensity is reduced, specifically may be referred to brightness in Figure 10 from 0 to Y_min_th sections of filtering strength change.

S34：To the first filtering strength a_yIt is modified, obtains the second filtering strength a__yy。

If two neighboring, because the effect of noise, one is judged into flat, and one is judged into edge. Then filtered result is in some regions, it may appear that lamination, such as face and hair intersection.And In order to avoid this phenomenon, in an embodiment of the present invention, the first edge threshold value can be set Y_thr1, and the first edge threshold value y_thr1<The 4th threshold value y_thr, for the center pixel Luminance component on marginal information edge fall [y_thr1, y_thr] interval in value, filtering can be corrected Intensity is flat excessively more natural and edge between to make.

In an embodiment of the present invention, it is possible to use fair curve shown in Figure 11 corrects first filter Intensity of wave, so as to obtain the second filtering strength a__yy, wherein, a in Figure 11_yIt is according to present intensity The first filtering strength calculated by above-described formula, the curved line relation in Figure 11 is described with formula, Can be such as (14)：

According to being described above, the setting of the 4th threshold value y_thr can be with as shown in figure 9, according to brightness not Set different together, so, the setting of the first edge threshold value y_thr1 similar with the 4th threshold value y_thr Described can also be set using the relation curve shown in Fig. 9 according to the different and different of brightness One edge threshold y_thr1, simply sets different threshold parameters on the longitudinal axis, and specific establishing method can Implement with reference to above description, will not be repeated here.

S35：To the second filtering strength a__yyIt is modified, obtains the 3rd filtering strength a.

In order to further avoid lamination, in the first difference maximum Umax_diff ∈ [th6, th7], When i.e. described first difference maximum is less than the 7th threshold value th7 more than the 6th threshold value th6, can be with root According to the first difference maximum, using the fair curve shown in Figure 12, to the second filtering strength a__yy Be modified, obtain the 3rd filtering strength a, using the 3rd filtering strength a as the center pixel filter Intensity of wave, the curve in Figure 12 can be equally described as follows with formula (15)：

According to mentioned above, the setting of the 3rd threshold value th3 may be referred to the curved line relation shown in Fig. 9, i.e., Can according to the different and different of brightness, so similar with the 3rd threshold value th3, the 6th threshold value th6 and The setting of the 7th threshold value th7 can also be according to the different and different of brightness.Similarly, the 6th threshold Value th6 can be as the method for the 3rd threshold value th3 settings, simply with the curve of the 7th threshold value th7 Different threshold parameters are set on the longitudinal axis in fig .9, be will not be repeated here.

S36：According to the 3rd filtering strength a, recursive filtering is carried out to the pixel in described image.

In an embodiment of the present invention, by taking the image block shown in Fig. 6 as an example, filtered institute is specifically calculated The formula such as (16) for stating center pixel U11 ' is shown：

S37：Judge the present intensity of the center pixel whether less than the default dark space luminance threshold.

When the present intensity of the center pixel is less than the default dark space luminance threshold, S38 is performed, instead It, then terminate flow.

S38：Saturation degree suppression treatment is carried out to filtered center pixel.

In order to prevent color in the dark from trailing, in an embodiment of the present invention, the middle imago can be worked as When the present intensity of element is less than default dark space luminance threshold, determine that the center pixel is in dark space, can Suppressed with the saturation degree to color, and in particular to two steps：

The first step：If the saturation degree of U11 ' is higher than U11, U11 '=U11, that is, ensure filtered full Before not higher than being filtered with degree；

Second step：If (U11 '-th5) * (U11-th5)<0, then U11 '=th5, that is, ensure filtered color Degree is without reverse；

Wherein：U11 ' represents the filtered center pixel, and U11 represents the center pixel before filtering, The 5th threshold value th5 is 128.

It should be noted that the saying of the first step and second step, simply to illustrate that and understand, and The specific execution sequence for not suppressing treatment to above-mentioned saturation degree constitutes any limitation.

There are the limitation of the two conditions, that is, the saturation degree suppression treatment of two steps above, especially dark Place, even if color boundaries are not detected well, will not also occur color hangover.

S39：Mean filter is carried out to the center pixel.

By taking the image block shown in table 3 as an example, it is assumed that have the absolute value of difference of N number of point and U11 less than default 9th threshold value th9, then final filtering such as formula (17) is described：

Wherein：Uij is satisfaction：The pixel of | Uij-U11 | ＜ th9, i.e. Uij is satisfaction and center pixel U11 Difference pixel of the absolute value less than the 9th threshold value th9, and N is to meet and the difference of center pixel U11 Number of pixels of the absolute value of value less than the 9th threshold value th9, the relation that N is met, such as formula (18) It is shown：

Wherein：Abs () is represented and the parameter in bracket is taken absolute value.

Understand in sum, because the marginal information on luminance component can be relatively more accurate, therefore by judging When whether the frequency field residing for center pixel is flat site, first edge information maximization is more than considered Value, the variance of the center pixel and the first difference maximum, it is also contemplated that the brightness of the center pixel Marginal information on component, can cause that the frequency region information of the center pixel can be more accurate, from And more accurate filtering strength can be determined for the center pixel, therefore filtered image can be caused Retain more details, while the denoising effect of noise is ensured, take into account the hangover for avoiding image after denoising Phenomenon.

To cause that those skilled in the art more fully understand and realize the present invention, it is also provided below can be real The device of the method for existing above-mentioned image denoising, as shown in figure 13, described device can include：Image block is obtained Take unit 81, the first computing unit 82, the second computing unit 83, the 3rd computing unit 84, the 4th calculating Unit 85, the first judging unit 86, the 5th computing unit 87, filtering strength acquiring unit 88 and filtering are single Unit 89, wherein：

Described image block acquiring unit 81, be suitable to obtain using pending pixel as center pixel image block, The row k kth that the center pixel is located at described image block is arranged；

First computing unit 82, is suitable to calculate the colourity both horizontally and vertically of the center pixel Marginal information on component, and take the marginal information on the chromatic component both horizontally and vertically most Big value, as first edge information maximization value；

Second computing unit 83, is suitable to calculate the variance of described image block；

3rd computing unit 84, be suitable to calculate the center pixel and the row of kth -1 in described image block/ The difference maximum of the filtered pixel selected in the row of kth -1, as the first difference maximum；

4th computing unit 85, is suitable to calculate the marginal information on the luminance component of the center pixel；

First judging unit 86, is suitable to according to the first edge information maximization value, described image block Variance, the first difference maximum and the center pixel luminance component on marginal information, judge institute State whether the frequency field residing for center pixel is flat site；

5th computing unit 87, is suitable to when the frequency field residing for the center pixel is flat site When, calculate the present intensity of the center pixel；

The filtering strength acquiring unit 88, is suitable to the present intensity according to the center pixel, obtains the One filtering strength, as the filtering strength of the center pixel；

The filter unit 89, is suitable to using the filtering strength of the center pixel, to the center pixel Carry out recursive filtering.

In specific implementation, first judging unit 86 is suitable to when following all conditions are met, really The frequency of the fixed center pixel is flat site：The first edge information maximization value is less than default the One threshold value；The variance is less than default Second Threshold；The first difference maximum is less than default the Three threshold values；Marginal information on the luminance component of the center pixel is less than default 4th threshold value.

In specific implementation, the value of the first threshold, Second Threshold, the 3rd threshold value and the 4th threshold value With the center pixel present intensity residing for brightness section it is different and different.

Figure 14 shows the device of another image denoising in the embodiment of the present invention, except image block is obtained Unit 81, the first computing unit 82, the second computing unit 83, the 3rd computing unit the 84, the 4th calculate single First 85, first judging unit 86, the 5th computing unit 87, filtering strength acquiring unit 88 and filter unit Outside 89, described device also includes：Saturation degree processing unit 91, is suitable to work as the current bright of the center pixel When degree is less than default dark space luminance threshold, determine that the center pixel is in dark space, and to the recurrence The filtered center pixel carries out dark space saturation degree suppression treatment.

In specific implementation, the saturation degree processing unit 91, including：

First judgment sub-unit 911, the component for being suitable to judge the center pixel after the recursive filtering is It is no more than recursive filtering before the center pixel component；

First computation subunit 912, be suitable to when first judgment sub-unit 911 determine it is described filtered The component of the center pixel more than the center pixel before the recursive filtering component when, using passing The component of the center pixel returned before filtering divides as the center pixel after the recursive filtering Amount.

In specific implementation, the saturation degree processing unit 91 also includes：

Second computation subunit 913, is suitable to filter the component of the center pixel after the recursive filtering, recurrence The component of the center pixel of wavefront takes difference with default 5th threshold value respectively, obtains the first difference and Two differences；

Second judgment sub-unit 914, is suitable to judge whether first difference is less than with the product of the second difference Zero；

3rd computation subunit 915, is suitable to determine that the product is less than when second judgment sub-unit 914 When zero, using the 5th threshold value as the center pixel after the recursive filtering component.

In an embodiment of the present invention, the component of the center pixel can be colourity U components, the described 5th Threshold value can be 128.

To cause that those skilled in the art more fully understand and realize the present invention, the present invention also provided below The schematic device of another image denoising in embodiment, as shown in figure 15, except above-mentioned image block Acquiring unit 81, the first computing unit 82, the second computing unit 83, the 3rd computing unit the 84, the 4th are counted Calculate unit 85, the first judging unit 86, the 5th computing unit 87, filtering strength acquiring unit 88, filtering Outside unit 89 and saturation degree processing unit 91, described device also includes that the second judging unit 101, first is corrected The filter of unit 102, the 3rd judging unit 103, the second amending unit 104, noise-removed filtering unit 105 and average Ripple unit 106, wherein：

Second judging unit 101, is suitable to when the marginal information on the luminance component of the center pixel is small When four threshold value, judge the marginal information on the luminance component of the center pixel whether more than pre- If first edge threshold value；

First amending unit 102, is suitable to when the marginal information on the luminance component of the center pixel is big When the first edge threshold value, the marginal information on luminance component according to the center pixel, to institute State the first filtering strength to be modified, obtain the second filtering strength, by second filtering strength, as The filtering strength of the center pixel.

In specific implementation, the 3rd judging unit 103 is suitable to judge whether to meet following condition：Institute The first difference maximum is stated more than default 6th threshold value less than default 7th threshold value, and the 7th threshold Value is less than the 3rd threshold value；

Second amending unit 104, is suitable to be more than default 6th threshold value when the first difference maximum When being less than three threshold value less than default 7th threshold value, and the 7th threshold value, according to described first Difference maximum, is modified to second filtering strength, obtains the 3rd filtering strength, by described Three filtering strengths as the center pixel filtering strength.

In specific implementation, the noise-removed filtering unit 105 is suitable to judging residing for the center pixel Before whether frequency field is flat site, the first row/column pixel to acquired image block carries out denoising Filtering process

In specific implementation, when the present intensity of the center pixel is less than default eight threshold value, institute Corresponding filtering strength is reduced as the present intensity is reduced.

In specific implementation, the mean filter unit 106 is suitable to when the frequency residing for the center pixel When region is not flat site, mean filter is carried out to the center pixel.

One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment Rapid to can be by program to instruct the hardware of correlation to complete, the program can be stored in can with computer Read in storage medium, storage medium can include：ROM, RAM, disk or CD etc..

Although present disclosure is as above, the present invention is not limited to this.Any those skilled in the art, Without departing from the spirit and scope of the present invention, can make various changes or modifications, therefore guarantor of the invention Shield scope should be defined by claim limited range.

Claims

1. a kind of method of image denoising, it is characterised in that recursive filtering is carried out to the pixel in described image, The filtering strength for obtaining currently pending pixel with the following method is adopted, including：

Obtain using the pending pixel as center pixel image block, the center pixel be located at the figure As the row k kth of block is arranged；

The marginal information on the chromatic component both horizontally and vertically of the center pixel is calculated, and is taken described The maximum of the marginal information on chromatic component both horizontally and vertically, as first edge information most Big value；

Calculate the variance of described image block；

Calculate filtered selected in the center pixel and the row of kth -1 in described image block/row of kth -1 The difference maximum of pixel, as the first difference maximum；

According to the first edge information maximization value, the variance of described image block, the first difference maximum and institute The marginal information on the luminance component of center pixel is stated, the frequency field residing for the center pixel is judged Whether it is flat site；

When the frequency field residing for the center pixel is flat site, the current of the center pixel is calculated Brightness；

2. the method for image denoising according to claim 1, it is characterised in that described according to described first Marginal information maximum, the variance of described image block, the first difference maximum and the center pixel Marginal information on luminance component, judges whether the frequency field residing for the center pixel is flat region Domain, including：

The variance is less than default Second Threshold；

The first difference maximum is less than default 3rd threshold value；

3. the method for image denoising according to claim 2, it is characterised in that the first threshold, The value of two threshold values, the 3rd threshold value and the 4th threshold value is current bright with the center pixel of described image block Brightness section is different and different residing for degree.

4. the method for image denoising according to claim 1, it is characterised in that also include：

When the present intensity of the center pixel is less than default dark space luminance threshold, the middle imago is determined Element be in dark space, and to the recursive filtering after the center pixel carry out at the saturation degree suppression of dark space Reason.

5. the method for image denoising according to claim 4, it is characterised in that described to be filtered to the recurrence The center pixel after ripple carries out dark space saturation degree suppression treatment, including：

Judge the component of the center pixel after the recursive filtering whether more than in described before recursive filtering The component of imago element；

The center pixel before the component of the filtered center pixel is more than the recursive filtering Component when, using the component of the center pixel before recursive filtering as the recursive filtering after The component of the center pixel.

6. the method for image denoising according to claim 4, it is characterised in that described to be filtered to the recurrence The center pixel after ripple carries out dark space saturation degree suppression treatment, also includes：

The component of the center pixel before by the component of the center pixel after the recursive filtering, recursive filtering Difference is taken with default 5th threshold value respectively, the first difference and the second difference is obtained；

When the product of first difference and the second difference is less than zero, using the 5th threshold value as the recurrence The component of the filtered center pixel.

7. the method for image denoising according to claim 6, it is characterised in that the 5th threshold value is 128.

8. the method for image denoising according to claim 2, it is characterised in that also include：

When the marginal information on the luminance component of the center pixel is less than four threshold value, judge described Whether the marginal information on the luminance component of center pixel is more than default first edge threshold value；

When the marginal information on the luminance component of the center pixel is more than the first edge threshold value, according to Marginal information on the luminance component of the center pixel, is modified to first filtering strength, Obtain the second filtering strength；

By second filtering strength, as the filtering strength of the center pixel.

9. the method for image denoising according to claim 8, it is characterised in that also include：

When the first difference maximum is more than default 6th threshold value and is less than default 7th threshold value, and institute When stating the 7th threshold value less than three threshold value；

According to the first difference maximum, second filtering strength is modified, obtains the 3rd filtering Intensity, using the 3rd filtering strength as the center pixel filtering strength.

10. the method for image denoising according to claim 1, it is characterised in that also include：

Before judging whether the frequency field residing for the frequency of the center pixel be flat site, to being obtained First row/column pixel of the image block for taking carries out noise-removed filtering treatment.

The method of 11. image denoisings according to claim 1, it is characterised in that when the center pixel When present intensity is less than default eight threshold value, corresponding filtering strength drops with the present intensity It is low and reduce.

The method of 12. image denoisings according to claim 1, it is characterised in that also include：

When the frequency field non-planar regions residing for the center pixel, average is carried out to the center pixel Filtering.

A kind of 13. devices of image denoising, it is characterised in that including：

Image block acquiring unit, be suitable to obtain using pending pixel as center pixel image block, it is described in Imago element is arranged positioned at the row k kth of described image block；

First computing unit, on the chromatic component both horizontally and vertically for being suitable to calculate the center pixel Marginal information, and the maximum of marginal information on the chromatic component both horizontally and vertically is taken, As first edge information maximization value；

3rd computing unit, is suitable to calculate the center pixel with the row of kth -1 in described image block/row of kth -1 Selected in filtered pixel difference maximum, as the first difference maximum；

First judging unit, be suitable to according to the first edge information maximization value, the variance of described image block, Marginal information on the luminance component of the first difference maximum and the center pixel, judges the center Whether the frequency field residing for pixel is flat site；

5th computing unit, is suitable to, when the frequency field residing for the center pixel is flat site, calculate The present intensity of the center pixel；

Filtering strength acquiring unit, is suitable to the present intensity according to the center pixel, obtains the first filtering strong Degree, as the filtering strength of the center pixel；

Filter unit, is suitable to, using the filtering strength of the center pixel, recurrence be carried out to the center pixel Filtering.

The device of 14. image denoisings according to claim 13, it is characterised in that first judging unit, The frequency for being suitable to determine the center pixel when following all conditions are met is flat site：

The variance is less than default Second Threshold；

The first difference maximum is less than default 3rd threshold value；

The device of 15. image denoisings according to claim 14, it is characterised in that the first threshold, The value of two threshold values, the 3rd threshold value and the 4th threshold value with the center pixel present intensity residing for it is bright Spend the different and different of interval.

The device of 16. image denoisings according to claim 13, it is characterised in that also include：

Saturation degree processing unit, is suitable to be less than default dark space luminance threshold when the present intensity of the center pixel During value, determine the center pixel be in dark space, and to the recursive filtering after the center pixel Carry out dark space saturation degree suppression treatment.

The device of 17. image denoisings according to claim 16, it is characterised in that the saturation degree treatment is single Unit, including：

First judgment sub-unit, is suitable to judge whether the component of the center pixel after the recursive filtering is big In the component of the center pixel before recursive filtering；

First computation subunit, is suitable to described after first judgment sub-unit determines the recursive filtering When the component of center pixel is more than the component of the center pixel before the recursive filtering, using recurrence The component of the center pixel before filtering as the center pixel after the recursive filtering component.

The device of 18. image denoisings according to claim 16, it is characterised in that the saturation degree treatment is single Unit, including：

Second computation subunit, is suitable to before the component of the center pixel after the recursive filtering, recursive filtering The component of the center pixel take difference with default 5th threshold value respectively, obtain the first difference and second Difference；

3rd computation subunit, is suitable to determine that first difference is poor with second when second judgment sub-unit When the product of value is less than zero, using the 5th threshold value as the center pixel after the recursive filtering Component.

The device of 19. image denoisings according to claim 18, it is characterised in that the 5th threshold value is 128。

The device of 20. image denoisings according to claim 14, it is characterised in that also include：

Second judging unit, is suitable to marginal information on the luminance component when the center pixel less than described the During four threshold values, judge marginal information on the luminance component of the center pixel whether more than default the One edge threshold；

First amending unit, is suitable to marginal information on the luminance component when the center pixel more than described the During one edge threshold, the marginal information on luminance component according to the center pixel, to described first Filtering strength is modified, and obtains the second filtering strength, by second filtering strength, as described The filtering strength of center pixel.

The device of 21. image denoisings according to claim 20, it is characterised in that also include：

3rd judging unit, is suitable to judge whether to meet following condition：The first difference maximum is more than pre- If the 6th threshold value and be less than default 7th threshold value, and the 7th threshold value be less than the 3rd threshold value；

Second amending unit, is suitable to when the first difference maximum more than default 6th threshold value less than default The 7th threshold value, and the 7th threshold value be less than three threshold value when, according to first difference most Big value, is modified to second filtering strength, obtains the 3rd filtering strength, by the described 3rd filter Intensity of wave as the center pixel filtering strength.

The device of 22. image denoisings according to claim 13, it is characterised in that also include：

Noise-removed filtering unit, is suitable to judging whether the frequency field residing for the center pixel is flat site Before, the first row/column pixel to acquired image block carries out noise-removed filtering treatment.

The device of 23. image denoisings according to claim 13, it is characterised in that when the center pixel When present intensity is less than default eight threshold value, corresponding filtering strength drops with the present intensity It is low and reduce.

The device of 24. image denoisings according to claim 13, it is characterised in that also include：

Mean filter unit, is suitable to when the frequency field non-planar regions residing for the center pixel, to institute Stating center pixel carries out mean filter.