CN102045487B - Method for noise suppression by using multiple digital pictures - Google Patents

Abstract

The invention discloses a method for noise suppression by using multiple digital pictures, which is used for denoising by virtue of multiple digital pictures, wherein the digital pictures are firstly subject to a feature weighting processing program and the picture feature compensation of target pixels, and then noise suppression is carried out on the target pixels by carrying out interactive reference on multiple continuous or similar digital pictures. The method for noise suppression by using multiple digital pictures, which is provided by the invention, comprises the following steps: acquiring multiple digital pictures; carrying out a first pixel compensation program on each digital picture; and carrying out a second pixel compensation program on a comparison picture in the digital pictures. In the noise suppression method disclosed by the invention, pixels of which the picture features are similar and the degree of similarity is high are used to repair, thus not destroying the digital pictures, so as to obtain better picture quality.

Description

Utilize many numbered image to suppress the method for noise

Technical field

The present invention is particularly to a kind of method of utilizing many numbered image to suppress noise about a kind of processing method of digital picture.

Background technology

Along with the fast development of digital camera, for the also significantly increase of demand of digital picture.Digital camera utilizes the plane formula photosensory assembly to catch one whole image, or therefrom captures a square zone.In case accomplish the action of exposure, control circuit can make the electric charge on the capacitor cell pass to adjacent next unit, when arriving last unit, edge, charge signal imports amplifier into, is transformed into current potential.So go round and begin again, all change into current potential, deposit internal memory after sampling and the digitlization in up to entire image.

But photosensory assembly can make the surface charge of photosensory assembly produce disturbance through after the long exposure actions.Because the relation of electric charge disturbance makes the imaging of digital picture some noises can occur.Except the time for exposure, the yield value of heightening photosensory assembly also can make the luminous energy power that receives of photosensory assembly increase, but also can make the easier disturbance of electric charge.

The noise that on hardware handles, is produced, the software processes process also can produce noise.When for instance, digital picture being carried out edge sharpening (sharp) processing.The more unconspicuous different colored pixels of script in the digital picture can be drawn high the color range of different color because sharpening handles.The particle that so, will occur different colours in the digital picture.

So many camera manufacturers or image processing manufacturer be in order to improve the quality of digital picture, thereby the method that many kinds are eliminated noises has been proposed.With regard to most method of eliminating audible noise, be mostly to utilize digital filtering technique to carry out the action that it eliminates noise.For instance, eliminating noise can be undertaken by the mode of similar color correction or fuzzy (blur) processing.Though these methods can reduce the noise in the digital picture, can have influence on the whole image quality of digital picture.For instance, through the digital picture of Fuzzy Processing, though can make the similar color pixel further reduce difference therebetween.But relative, can cause also that original border thickens and identification easily in the digital picture.

Summary of the invention

In view of above problem, main purpose of the present invention is to provide a kind of method of utilizing many numbered image to suppress noise, by the pixel of adjacent domain in many numbered image, in order to a pixel is carried out color correction process.

For reaching above-mentioned purpose, the disclosed method of utilizing many numbered image to suppress noise of the present invention may further comprise the steps: obtain many numbered image; Carry out the first pixel compensation program of each numbered image; And the second pixel compensation program of carrying out a comparison image in these a little digital pictures.

The first pixel compensation program may further comprise the steps: utilize the characteristic weighing handling procedure, convert each original pixels in the digital picture into the characteristic weighing pixel, with the output characteristic weighted image; According to the position of an object pixel in digital picture, the opposite position selected characteristic weighted pixel in the characteristic weighing image is as benchmark pixel; Carry out similarity calculation program, calculate the similarity of the benchmark pixel further feature weighted pixel outer with respect to benchmark pixel; Use the characteristic weighing pixel by way of compensation according to choosing a characteristic weighing pixel the further feature weighted pixel of the similarity that calculates outside benchmark pixel; Carry out the pixel compensation program, compensate according to the characteristics of image of corresponding similarity weights object pixel according to the original image value of the compensation that selects with the characteristic weighing pixel.

The second pixel compensation program may further comprise the steps: the similarity of calculating comparison image other digital picture outer with respect to the comparison image; And, carry out the weighted average of the pairing feature compensation pixel of object pixel according to those similarities of comparison image other digital picture outer with respect to the comparison image, with the second feature compensation pixel of the object pixel that obtains comparing image.

Wherein, weighted average can be calculated based on formula 6.

${\mathrm{PixelWeightValue}}^{\′}\left(i\right)=\frac{\underset{k=1:N}{\mathrm{\Σ}}{\mathrm{\ω}}_k\×\mathrm{PixelWeightVa}{\mathrm{Lue}}_k\left(i\right)}{\underset{k=1:N}{\mathrm{\Σ}}{\mathrm{\ω}}_k}$ ... ... ... .... formula 6

Wherein, i represents object pixel, and arbitrary in outer other digital picture of k representative comparison image, PixelWeightValue ' (i) are the pixel value of the second feature compensation pixel of object pixel in the comparison image, PixelWeightValue _k(i) for comparing the pixel value of the first feature compensation pixel of target in the digital picture beyond the image, ω _kBe the similarity of other digital picture that the comparison image is outer with respect to the comparison image, N representes the number of digital picture, i be 1-N in arbitrary positive integer, and k then is the arbitrary positive integer beyond the k in 1-N.

Moreover the first pixel compensation program can be carried out according to formula 1, formula 2, formula 3 and formula 4.

Diff (i, j)=‖ PixelValue (Neighbor _i)-PixelValue (Neighbor _j) ‖ ... ... .. formula 1

W (i, j)=f (Diff (i, j)) ... ... ... ... ... ... ... .. formula 2

$\mathrm{PixelWeightValue}\left(i\right)=\underset{j\∈R}{\mathrm{\Σ}}w(i,j)\×\mathrm{PixelValue}\left(j\right)$ ... ... .... formula 3

$\underset{j\∈R}{\mathrm{\Σ}}w(i,j)=1$ ... ... ... ... ... ... ... ... formula 4

Wherein, i represents object pixel, and j represents the original pixels of the characteristic weighing pixel around the corresponding benchmark pixel; (i j) is the diversity factor of benchmark pixel with respect to the further feature weighted pixel to Diff, and PixelValue (Neighbori) is the pixel characteristic weighted value of i; PixelValue (Neighborj) is the pixel characteristic weighted value of j, and (i j) is the similarity of benchmark pixel with respect to the further feature weighted pixel to w; F () is the conversion function of diversity factor to similarity, and PixelWeightValue (i) carries out the pixel value after the feature compensation program for i, and PixelValue (j) is the pixel value of j; R representes the size of digital picture, and R is M * N, and M and N are respectively the positive integer more than or equal to 1; I be 1-M * N in arbitrary positive integer, j then is the arbitrary positive integer beyond the i in 1-M * N.

The noise suppressing method of digital picture of the present invention utilizes the close and pixel that similarity is high of characteristics of image to repair, thus can not destroy digital picture, thereby can obtain better picture quality.

About feature of the present invention and real the work, existing conjunction with figs. is described in detail as follows as most preferred embodiment.

Description of drawings

Fig. 1 is for to utilize many numbered image to suppress the flow chart of the method for noise according to an embodiment of the invention;

Fig. 2 A is the sketch map of the digital picture of an embodiment;

Fig. 2 B is the sketch map of the characteristic weighing image of an embodiment;

Fig. 3 is the flow chart of the characteristic weighing handling procedure of an embodiment;

Fig. 4 is the sketch map that the digital picture of an embodiment converts the characteristic weighing image into;

Fig. 5 A is the sketch map of selected directions of the chosen area of first embodiment;

Fig. 5 B is the sketch map of selected directions of the chosen area of second embodiment;

Fig. 5 C is the sketch map of selected directions of the chosen area of the 3rd embodiment;

Fig. 6 A is the sketch map of the digital picture of another embodiment;

Fig. 6 B is the sketch map of the characteristic weighing handling procedure of corresponding diagram 6A;

Fig. 6 C is the sketch map of the similarity calculation program of corresponding diagram 6A.

Wherein, Reference numeral:

210 digital pictures, 210 ' characteristic weighing image

220 chosen area, 221 intermediate pixels

Pixel arround 221 ' the characteristic weighing pixel 222

223 benchmark pixel, 322 characteristic weighing pixels

323 characteristic weighing pixel P original pixels

P ' characteristic weighing pixel Pn locatees pixel

Embodiment

The method of utilizing many numbered image to suppress noise according to the present invention can be applicable to a computing electronics, uses through this computing electronics the digital picture in the inputing to computing electronics is carried out the noise color correction process.In other words; According to of the present invention utilize method that many numbered image suppresses noise can software or the firmware program be stored in the storage element (for example: internal memory or hard disk etc.) of computing electronics, carry out the software that stores or firmware program by the processor of computing electronics again and realize.

Please refer to shown in Figure 1ly, it is operation workflow sketch map according to an embodiment of the invention.This method of utilizing many images to suppress the noise of digital picture may further comprise the steps.

At first, step S102: obtain many continuous or similar digital pictures.

And, carry out the pixel compensation of each numbered image, promptly carry out the first pixel compensation program.

Step S110: utilize the characteristic weighing handling procedure, convert each the original pixels P in the digital picture 210 into characteristic weighing pixel P ', to produce characteristic weighing image 210 ', shown in Fig. 2 A and 2B.

Step S120: according to the position of object pixel in digital picture 210, the opposite position selected characteristic weighted pixel in characteristic weighing image 210 ' is with as benchmark pixel.Wherein, object pixel is the arbitrary pixel among all original pixels P of digital picture 210.

Step S130: benchmark pixel is carried out similarity with respect to the further feature weighted pixel P ' beyond the benchmark pixel and is calculated.

Step S140: from characteristic weighing pixel P ', the high or the most higher characteristic weighing pixel P ' of benchmark pixel similarity is used the characteristic weighing pixel respectively by way of compensation according to setting to choose.In addition, also can choose similarity to benchmark pixel greater than threshold value or meet the characteristic weighing pixel that imposes a condition and use the characteristic weighing pixel by way of compensation.In other words, can be on demand in advance this step being set at chooses the highest characteristic weighing pixel P ' of benchmark pixel similarity or being set to choose the higher characteristic weighing pixel P ' of benchmark pixel similarity or be set at chooses the similarity of benchmark pixel greater than the characteristic weighing pixel P ' of threshold value or being set at the similarity of choosing benchmark pixel meets the characteristic weighing pixel P ' that imposes a condition and use the characteristic weighing pixel by way of compensation.

Step S150: according to corresponding this similarity object pixel is carried out the compensation of characteristics of image with the pairing original pixels P of characteristic weighing pixel P according to the compensation that selects, to obtain the first feature compensation pixel.

Then, carry out the pixel compensation between the digital picture, promptly carry out the second pixel compensation program.

Step S160: from many numbered image 210, choose a digital picture 210 as the comparison image.Wherein, 210 of the digital picture of comparison beyond the image are as reference picture.

Step S170: calculate the similarity of comparison image with respect to each reference picture.

Step S180: carry out the weighted average of the pairing feature compensation pixel of object pixel according to the similarity that calculates, with the second feature compensation pixel of the object pixel that obtains comparing image.

Wherein, can more may further comprise the steps in the characteristic weighing handling procedure in step S110, and please also refer to Fig. 3 and Fig. 4.

Step S111: in digital picture 210, set chosen area 220.Wherein, topography's block in the chosen area 220 index word images 210, it can have the specific dimensions size or be single pixel.In this, chosen area 220 can be a * b pel array, and a and b are respectively the positive integer more than or equal to 1.Wherein, a can be identical numerical value with b, or is numerical value inequality.

Step S112: pixel 222 is carried out the characteristic weighing handling procedure arround utilizing the intermediate pixel 221 be arranged in chosen area 220 and being arranged in each of chosen area 220, to produce the characteristic weighing pixel 221 ' of intermediate pixel 221.Wherein, intermediate pixel 221 is the original pixels P of the centre that is positioned at chosen area 220, and arround pixel 222 for being positioned at the original pixels P arround the intermediate pixel 221 in the chosen area 220.

Step S113: execution in step S111 and step S112 each original pixels P in digital picture 210 all converts characteristic weighing pixel P ' into repeatedly.

Please refer to Fig. 5 A, 5B and 5C, because the order of choosing of chosen area 220 depends on selected location pixel Pn.In other words, when carrying out the setting of chosen area 220, can select a pixel as the location pixel, be that benchmark forms chosen area 220 with location pixel Pn at every turn again.Wherein, can set the original pixels that is arranged in chosen area 220 arbitrary positions (for example: the upper right corner, the upper left corner, the lower right corner, the lower left corner or center etc.) as location pixel Pn.

In the present embodiment, the setting of chosen area 220 can zigzag (like the represented direction of dotted line among Fig. 5 A) selection in regular turn be located pixel Pn, promptly positions the choosing of pixel Pn by direction left-to-right, from top to bottom.Therefore, chosen area 220 can be in regular turn and the setting that overlaps, so that choose each image block of whole numbered image, and shown in Fig. 5 B.Can certainly utilize different directions to carry out the order of choosing in proper order to location pixel Pn and chosen area 220.In addition, chosen area 220 also can nonoverlappingly be set, shown in Fig. 5 C.

Because utilize chosen area 220 can cause the marginal portion of digital picture 210 can't produce corresponding characteristic weighing pixel P ', make the size of the characteristic weighing image 210 ' after the weighting can be slightly less than the size of digital picture 210.For example, if digital picture 210 is the image of 100*100 (pel array), and chosen area 220 is the 3*3 pel array.Then the characteristic weighing image 210 ' after the output is the 98*98 pel array.

Therefore, can utilize the edge of digital picture 210 to repair for the marginal portion of characteristic weighing image 210 ', the size that makes characteristic weighing image 210 ' is big or small identical with digital picture 210.

Perhaps the marginal portion of characteristic weighing image 210 ' is not repaired, but characteristic weighing pixel 210 ' is adjusted for the corresponding relation on the digital picture 210.For instance, go up the characteristic weighing pixel of position (1,1), its correspondence original pixels of position (3,3) on digital picture 210 at characteristic weighing image 210 '.In like manner, also can carry out the foundation of the pixel corresponding relation between characteristic weighing image 210 ' and the digital picture 210 for the digital picture 210 and the chosen area 220 of other different sizes according to this mode.

And please cooperate with reference to shown in Figure 4, it is the sketch map of each original pixels P characteristic of correspondence weighted pixel P ' of chosen area 220.Be convenient explanation, being defined in location pixel selected in the digital picture 210 at this is intermediate pixel 221.After selecting out intermediate pixel 221, be one a * b is set at the center arround intermediate pixel 221 pel array with intermediate pixel 221.In this, this a * b pel array is defined as chosen area 220.Intermediate pixel 221 respectively to all the other original pixels of chosen area 220 (that is, and arround pixel 222, as filling up the original pixels of oblique line among Fig. 4) carry out the characteristic weighing routine processes respectively, use the characteristic weighing pixel 221 ' that produces corresponding intermediate pixel 221.

For instance, if chosen area 220 is the 5*5 pel array, the original image that can select to be positioned at location of pixels (3,3) is usually as intermediate pixel 221.Select the position of intermediate pixel 221 to determine, so do not enumerate one by one at this according to different enforcement aspects.

Then, please refer to following formula 1 to formula 4, its pixel value to pixel carries out the calculating of the similarity and the first feature compensation pixel.

Diff (i, j)=‖ PixelValue (Neighbor _i)-PixelValue (Neighbor _j) ‖ ... ... formula 1

W (i, j)=f (Diff (i, j)) ... ... ... ... ... .... formula 2

$\mathrm{PixelWeightValue}\left(i\right)=\underset{j\∈R}{\mathrm{\Σ}}w(i,j)\×\mathrm{PixelValue}\left(j\right)$ ... ... .. formula 3

$\underset{j\∈R}{\mathrm{\Σ}}w(i,j)=1$ ... ... ... ... ... ... ... ... ... ... ... formula 4

Wherein, i represents object pixel (that is, i original pixels), and j represents reference pixel (that is j original pixels).Wherein, reference pixel refer to around the benchmark pixel can reference the pairing original pixels of characteristic weighing pixel.For instance, if digital picture 210 is M * N pel array, and M and N are the positive integer more than or equal to 1.At this moment, i is any positive integer in 1-M * N, and j then is any positive integer beyond the i in 1-M * N.

PixelValue (Neighbor _i) be the pixel characteristic weighted value of object pixel i, i.e. benchmark pixel.

PixelValue (Neighbor _j) be the pixel characteristic weighted value of the chosen area 210 internal reference pixel j at object pixel i place, i.e. further feature weighted pixel beyond the benchmark pixel.

F (.) is that characteristic weighing pixel diversity factor is to similarity weights conversion function.

(i is the diversity factor of object pixel i to the characteristic weighing value of reference pixel j j) to Diff, and promptly benchmark pixel is with respect to the diversity factor of further feature weighted pixel.

PixelWeightValue (i) is the pixel value of the correction of object pixel i, and promptly object pixel i carries out the pixel value (the first feature compensation pixel) after characteristics of image compensates.

PixelValue (j) is the preceding original pixel value of the processing of reference pixel j.

(i j) is the similarity weights of object pixel i to reference pixel j to w, and promptly benchmark pixel is with respect to the similarity of the further feature weighted pixel beyond the benchmark pixel.

R representes the size of this digital picture.

The method of utilizing many numbered image to suppress noise according to the present invention is carried out the similarity calculating except being directed against color of pixel (pixel characteristic), more can carry out similarity according to the different images characteristic value and calculate.For instance, characteristics of image can be the material of color space, image, the texture of image and/or the brightness of image, but is not limited thereto.With color space, (i j) then can change following formula 5 into to the similarity weight w.

Diff (i, j)=‖ ColorIndexValue (Neighbor _i)-ColorIndexValue (Neighbor _j) ‖ ... ... ... ... ... ... ... ... ... ... ... ... ... .. formula 5

ColorIndexValue (Neighbor _i) be the characteristic weighing value of the color space of object pixel i, i.e. benchmark pixel.

PixelValue (Neighbor _j) be the characteristic weighing value of the color space of the chosen area 210 internal reference pixel j at object pixel i place.

Based on the relation of these characteristics of image (or its combination), further calculate the similarity degree between neighborhood pixels again.Human eye vision is not only to be the linear relationship that directly the mutual addition of pixel can be obtained for the different images characteristic.For instance, in the chosen area with complex texture 220, the similarity of each pixel is understood the relation because of texture, and causes the similarity between each pixel to reduce.

After the characteristic weighing pixel of all pixels, the genesis sequence according to each characteristic weighing pixel produces a characteristic weighing image again in accomplishing digital picture 210.Then, in the characteristic weighing image, selecting the characteristic weighing pixel one by one, in order to carry out calculation of similarity degree, is benchmark pixel (that is, the position is in the characteristic weighing pixel of object pixel correspondence position) at this to selecteed characteristic weighing pixel definition.Obtain object pixel to the similarity of the characteristics of image of reference pixel and the similarity weight w (i, j).From characteristic weighing pixel 221 ', choose the characteristic weighing pixel the highest or higher relatively, use the characteristic weighing pixel to be defined as compensation to the benchmark pixel similarity.According to the original pixels of compensation with characteristic weighing pixel position, again object pixel is carried out the feature compensation program, and the new object pixel in output compensation back.

The feature compensation program of wherein, carrying out the compensation of characteristics of image can be but the mode such as replacement or average that is not defined as pixel value is carried out.If the mode with average is an example; It averages the value of object pixel and the value of calculating the back gained through weights; In order to export new revised object pixel; The calculating of similarity weights simultaneously look selectedly be decided to be that similarity is the highest, all pixel similarities are calculated in higher relatively or all chosen area of certain condition, look different demands and application scenario and decide.

The similarity of in this enforcement aspect, respectively each pixel in the digital picture being carried out is to each other calculated.For operation workflow of the present invention can clearly be described, thus at this with the explanation of a face image as present embodiment.

At first, select an original pixels in the digital picture 210 and set corresponding chosen area 220, shown in Fig. 6 A.The size of supposing chosen area 220 is the 5*5 pel array, and intermediate pixel 221 is the pixel of the position (3,3) in the chosen area 220.Then, utilize chosen area 220 interior pixels that middle pixel 221 is carried out the characteristic weighing handling procedure, and produce characteristic of correspondence weighted pixel 221 ', shown in Fig. 6 B.In other words, utilize chosen area 220 in order to produce the characteristic weighing value of intermediate pixel 221 exactly.Digital picture is repeated above-mentioned action, use the characteristic weighing pixel that produces each original pixels in the digital picture 210.

Fig. 6 C is the parts of images of characteristic weighing image 210 '.In Fig. 6 C, select a characteristic weighing pixel, this is defined as benchmark pixel 223 at this.And set a comparison scope, benchmark pixel 223 respectively with the comparison scope in all the other characteristic weighing pixels carry out calculation of similarity degree.The big I setting of comparison scope is big or small consistent with chosen area 220, can also do different restrictions to the comparison scope.For example: in less digital picture 210, can set the comparison scope for whole numbered image 210; If in bigger digital picture 210, then can set the comparison scope in a big way pel array, use and accelerate its computational speed.

In this embodiment, respectively benchmark pixel 223 is carried out calculation of similarity degree with the second characteristic weighing pixel 322 and the 3rd characteristic weighing pixel 323.In other implements aspect, can carry out the comparison of similarity according to the characteristic weighing pixel that right quantity is chosen in different settings.

With regard to color distortion, from the characteristic weighing image 210 ' of Fig. 6 C, can find out the difference of the difference of 223 pairs second characteristic weighing pixels 322 of benchmark pixel less than 223 pairs the 3rd characteristic weighing pixels 323 of benchmark pixel.In other words, the second characteristic weighing pixel 322 than the 3rd characteristic weighing pixel 323 more near benchmark pixel 223.So when being noise pixel, can preferentially choosing the second characteristic weighing pixel 322 and repair as if benchmark pixel 223.

Please refer to following formula 6 again, its weighted average degree to the pixel value of the first feature compensation pixel calculates.

${\mathrm{PixelWeightValue}}^{\′}\left(i\right)=\frac{\underset{k=1:N}{\mathrm{\Σ}}{\mathrm{\ω}}_k\×\mathrm{PixelWeightVa}{\mathrm{Lue}}_k\left(i\right)}{\underset{k=1:N}{\mathrm{\Σ}}{\mathrm{\ω}}_k}$ ... ... .... formula 6

PixelWeightValue ' (i) is the pixel value of the correction after the final processing of object pixel i in the comparison image, promptly compares the pixel value of the second feature compensation pixel of the object pixel i of image.

PixelWeightValue _k(i) be the pixel value of the first feature compensation pixel of target i among the reference picture k.

ω _kFor reference picture k compares to the similarity of image.

N is total number of digital picture, and wherein N is the positive integer more than 2.

In the present invention, be a plurality of chosen area 220 from digital picture 210 cuttings, and from each chosen area 220, produce the characteristic weighing pixel P ' that represents each chosen area 220.Choose one the pixel of weighting and itself and remaining characteristic weighing image are carried out similarity calculate as benchmark pixel 223.Therefrom calculate the compensation correction weight that each characteristic weighing pixel P ' revises benchmark pixel 223 more respectively, utilize the original pixel value at these characteristic weighing pixels P ' place that target pixel value is revised at last.Then, many continuous or similar digital pictures are being carried out mutual reference, to reach better picture quality.Because the present invention repairs object pixel according to pixel close and that similarity is high, thus can not cause the destruction of former digital picture, thereby have better picture quality.

Claims (5)

1. a method of utilizing many numbered image to suppress noise is carried out denoising by many numbered image and is handled, and it is characterized in that the method for utilizing many numbered image to suppress noise may further comprise the steps:

Obtain many numbered image;

Carry out one first pixel compensation program of each this digital picture, this first pixel compensation program may further comprise the steps:

Convert each original pixels in this digital picture into a characteristic weighing pixel, to produce a characteristic weighing image;

According to the position of an object pixel in this digital picture, the opposite position in this characteristic weighing image one of is chosen in this characteristic weighing pixel as a benchmark pixel, and wherein this object pixel is arbitrary in all these original pixels of this digital picture;

Calculate the similarity of this benchmark pixel other this characteristic weighing pixel outer with respect to this benchmark pixel;

According to this benchmark pixel this similarity of other these characteristic weighing pixels is selected a compensation from other this characteristic weighing pixel and use the characteristic weighing pixel; And

Compensate according to the characteristics of image of corresponding this similarity according to this compensation this original pixels, to obtain one first feature compensation pixel of this object pixel this object pixel with characteristic weighing pixel place; And

Carrying out one second pixel compensation program of a comparison image in those digital pictures, should the comparison image be the arbitrary digital picture in this digital picture wherein, and this second pixel compensation program may further comprise the steps:

Calculate the similarity of this comparison image other those digital pictures outer with respect to this comparison image; And

Those similarities according to this comparison image other those digital pictures outer with respect to this comparison image; Carry out the weighted average of pairing this feature compensation pixel of this object pixel, with one second feature compensation pixel of this object pixel of obtaining this comparison image.

2. the method for utilizing many numbered image to suppress noise as claimed in claim 1 is characterized in that, the step that produces this characteristic weighing image comprises:

In this digital picture, set a chosen area;

Pixel is carried out a characteristic weighing handling procedure arround utilizing an intermediate pixel and in this chosen area each in this chosen area; To produce to this characteristic weighing pixel that should intermediate pixel; Wherein this intermediate pixel is this original pixels that is positioned at the centre of this chosen area, and should arround pixel for being positioned at this original pixels arround this intermediate pixel in this chosen area; And

This original pixels of step each in this digital picture of carry out setting the step of this chosen area repeatedly and producing this characteristic weighing pixel that should intermediate pixel converts this characteristic weighing pixel into.

3. the method for utilizing many numbered image to suppress noise as claimed in claim 2 is characterized in that, this chosen area is one a * b pel array, and a and b are respectively the positive integer more than or equal to 1.

4. the method for utilizing many numbered image to suppress noise as claimed in claim 1; It is characterized in that this first pixel compensation program is carried out according to formula 1, formula 2, formula 3 and formula 4, wherein i represents this object pixel; J representative is to this original pixels of this characteristic weighing pixel around should benchmark pixel; (i j) is the diversity factor of this benchmark pixel with respect to other this characteristic weighing pixel to Diff, PixelValue (Neighbor _i) be the pixel characteristic weighted value of this i, PixelValue (Neighbor _j) be the pixel characteristic weighted value of this j, (i j) is this benchmark pixel this similarity with respect to other this characteristic weighing pixel to w; F () is the conversion function of this diversity factor to this similarity; PixelWeightValue (i) is the pixel value of this first feature compensation pixel of this i, and PixelValue (j) is the pixel value of this j, and R representes the size of this digital picture; R is M * N; M and N are respectively the positive integer more than or equal to 1, i be 1-M * N in arbitrary positive integer, j then is the arbitrary positive integer beyond the i in 1-M * N;

Diff(i，j)＝

‖ PixelValue (Neighbor _i)-PixelValue (Neighbor _j) ‖ ... .... formula 1

W (i, j)=f (Diff (i, j)) ... ... ... ... ... ... .. formula 2

$\mathrm{PixelWeightValue}\left(i\right)=\underset{j\∈R}{\mathrm{\Σ}}w(i,j)\×\mathrm{PixelValue}\left(j\right)$ ... ... ... ... formula 3

$\underset{j\∈R}{\mathrm{\Σ}}w(i,j)=1$ ... ... ... ... ... ... ... ... .... formula 4.

5. the method for utilizing many numbered image to suppress noise as claimed in claim 1 is characterized in that, this characteristics of image is a pixel color, a color space, a texture or a brightness.

Images