CN102547160B - Noise point suppression method - Google Patents
Noise point suppression method Download PDFInfo
- Publication number
- CN102547160B CN102547160B CN201110094256.XA CN201110094256A CN102547160B CN 102547160 B CN102547160 B CN 102547160B CN 201110094256 A CN201110094256 A CN 201110094256A CN 102547160 B CN102547160 B CN 102547160B
- Authority
- CN
- China
- Prior art keywords
- pixel
- image
- source
- value
- intend
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 230000001629 suppression Effects 0.000 title claims abstract description 18
- 238000001914 filtration Methods 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 40
- 238000003475 lamination Methods 0.000 claims description 31
- 230000005764 inhibitory process Effects 0.000 claims description 15
- 238000004040 coloring Methods 0.000 claims description 8
- 230000008030 elimination Effects 0.000 abstract description 2
- 238000003379 elimination reaction Methods 0.000 abstract description 2
- 230000000452 restraining effect Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000003672 processing method Methods 0.000 description 4
- 230000000295 complement effect Effects 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000002250 progressing effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
- Processing Of Color Television Signals (AREA)
Abstract
The invention provides a noise point restraining method, which is used for filtering noise points in a source image by utilizing a reference image. Wherein, the arrangement mode of each pixel of the source image and the reference image accords with the color arrangement rule of the Bell chart. In the method, a first pseudo-image corresponding to a source image and a second pseudo-image corresponding to a reference image are generated. A global motion vector between the first and second pseudo-images is calculated. The method comprises the steps of carrying out dynamic compensation processing on a reference image according to a global motion vector to obtain a plurality of dynamic compensation results, carrying out time superposition processing on a source image by utilizing the dynamic compensation results to generate a first-order noise suppression image, and carrying out space noise elimination processing on the first-order noise suppression image to generate a second-order noise suppression image.
Description
Technical field
The present invention relates to a kind of image treatment method, relate in particular to a kind of for eliminating the noise point suppression method of image noise.
Background technology
Progressing greatly of multimedia technology makes modern day by day improve the requirement of high image quality image.And the quality of image quality has sizable associated with the noise amount of supervening in pick-up image, signal conversion and transmitting procedure.In order effectively to remove noise to improve image quality, the research of eliminating about noise in image processing field also more and more comes into one's own.Common noise cancellation can be divided into denoising point processing method and the time-based denoising point processing method based on space.
Denoising point processing method based on space is mainly to process for individual image.But owing to often can destroy the details such as edge or texture in raw video in processing procedure, therefore removing the fuzzyyer result of easy generation after noise.
Time-based denoising point processing method is to utilize multiple images to process, although can retain the details in image, but also easily carry out when filtering is processed producing the situation of mobile ghost at the moving article in image, thereby easily in the time watching, cause discomfort.In addition the complexity of, utilizing multiple images to process is high and operand is large, the therefore more difficult demand that meets immediate removal noise.
Summary of the invention
In view of this, the invention provides a kind of noise point suppression method, process more effectively eliminate the noise in the image of source by two stage denoising point.
The present invention proposes a kind of noise point suppression method, to utilize with reference to the noise in the image of image filtering source.Wherein originate image with comprise respectively n 2 × 2 block of pixels with reference to image, and each pixel in each 2 × 2 block of pixels is to have one of them pixel value of corresponding red channel (R color channel), the first green channel (Gr colorchannel), the second green channel (Gb color channel) and blue channel (B colorchannel) according to Bel figure (Bayer pattern) color queueing discipline, and n is positive integer.The method comprises that producing first of corresponding source image intends like image and correspondingly intend like image with reference to second of image.Wherein, first intends comprising that like image n source plan is like pixel, and intend like 2 × 2 block of pixels in the image of the corresponding source of pixel difference in each source.And the second plan comprises that like image n with reference to intending like pixel, and each is corresponding to 2 × 2 block of pixels in image respectively like pixel with reference to plan.The method also comprises that calculating first intends intending like the universe motion-vector between image like image and second, according to universe motion-vector to carrying out dynamic compensation processing with reference to image to obtain multiple dynamic compensation results, and utilize above-mentioned dynamic compensation result to carry out the time superimposed (temporal blending) to source image and process to produce the first rank noise inhibition image, and the first rank noise inhibition image is carried out to space denoising point (spatial noise reduction) processing and suppress image to produce second-order noise.
In one embodiment of this invention, wherein according to universe motion-vector to carry out with reference to image dynamic compensation processing with obtain the step of multiple dynamic compensation results comprise order with reference to image according to universe motion-vector alignment source image, to produce the first dynamic compensation result, the second dynamic compensation result, the 3rd dynamic compensation result of corresponding red channel, the first green channel, the second green channel and blue channel respectively, and the 4th dynamic compensation result.
In one embodiment of this invention, wherein utilize above-mentioned dynamic compensation result to carry out time lamination process to source image and also comprise for each seemingly pixel of plan of originating to produce the step of the first rank noise inhibition image, intend seemingly in pixel, finding out with the plan reference plan that seemingly pixel has correspondence position of originating like pixel in all references.Intend, like pixel and with reference to intending like color and luminance difference distance between pixel, judging whether to intend carrying out time lamination process like corresponding 2 × 2 block of pixels of pixel to source according to source.If the time of carrying out lamination process, will originate and intend like the each pixel value in corresponding 2 × 2 block of pixels of pixel and in the first dynamic compensation result, the second dynamic compensation result, the 3rd dynamic compensation result, and the 4th is superimposed in the pixel value of correspondence position in dynamic compensation result, to produce a local superimposed result.If do not carry out time lamination process, using source intend like corresponding 2 × 2 block of pixels of pixel as part superimposed result.Last intend distinguish the superimposed result in corresponding part like pixel according to the color queueing discipline of Bel's graph coloring each source of recombinating again, thus generation the first rank noise inhibition image.
In one embodiment of this invention, wherein according to source intend like pixel with reference to intending like color and luminance difference distance between pixel, judge whether that the step that source is intended carrying out time lamination process like corresponding 2 × 2 block of pixels of pixel comprises: the two the absolute difference of pixel value of source being intended intending like the pixel value of pixel and source to corresponding blue channel in corresponding 2 × 2 block of pixels of pixel is seemingly defined as seemingly chroma blue of a source plan.Source is intended to be defined as a source with source plan like the absolute difference of the pixel value of corresponding red channel in corresponding 2 × 2 block of pixels of pixel like the pixel value of pixel intends like red color.Be defined as one with reference to intend seemingly chroma blue with reference plan like the absolute difference of the pixel value of corresponding blue channel in corresponding 2 × 2 block of pixels of pixel with reference to the pixel value of intending like pixel.Be defined as one with reference to intend seemingly red color with reference plan like the absolute difference of the pixel value of corresponding red channel in corresponding 2 × 2 block of pixels of pixel with reference to the pixel value of intending like pixel.And, obtain corresponding source and intend brightness threshold value, chroma blue threshold value and the red color threshold value like pixel.Intend the seemingly absolute difference of the pixel value of pixel like the pixel value of pixel with reference when source plan and be less than brightness threshold value, source intends intending being less than chroma blue threshold value like the absolute difference of chroma blue like chroma blue and reference, and when source intends intending being less than red color threshold value like the absolute difference of red color like red color and reference, judgement will be intended seemingly corresponding 2 × 2 block of pixels of pixel to source and carry out time lamination process.
In one embodiment of this invention, wherein obtain corresponding source plan and be included in the first plan like in image like the step of the brightness threshold value of pixel, intend obtaining one m × m block of pixels like centered by pixel to originate, wherein m is positive integer.Calculate the mean absolute error (Mean Absolute Error, MAE) of the each pixel value in m × m block of pixels.Obtain a brightness initial value according to mean absolute error, and intend obtaining a luminance gain value like the pixel value of pixel according to source, then intend brightness threshold value like pixel with the product of luminance gain value as corresponding source using brightness initial value.
In one embodiment of this invention, wherein obtaining corresponding source plan comprises according to source and intends obtaining a chroma blue yield value like chroma blue like the step of the chroma blue threshold value of pixel.Intend chroma blue threshold value like pixel with the product of chroma blue yield value as corresponding source using brightness initial value again.
In one embodiment of this invention, wherein obtain corresponding source plan and comprise according to source and intend obtaining a red color yield value like red color like the step of the red color threshold value of pixel, and intend red color threshold value like pixel with the product of red color yield value as corresponding source using brightness initial value.
In one embodiment of this invention, wherein produce corresponding source plan and more comprise definition the first weight preset value and the second weight preset value like the step of the superimposed result in part of pixel., intend obtaining m × m block of pixels like centered by pixel to originate like in image in the first plan, wherein m is positive integer, and calculates the mean absolute error of the each pixel value in m × m block of pixels.Obtain a superimposed weighted value of time according to mean absolute error, and produce the part superimposed result of corresponding source plan like pixel according to superimposed weighted value of time.
In one embodiment of this invention, wherein producing corresponding source plan according to superimposed weighted value of time comprises with formula (1) and calculating like the step of the superimposed result in part of pixel:
Wherein BR represents the part superimposed result of corresponding source plan like pixel.Wc represents that superimposed weighted value of time, C_base represent default coefficient.Rs, Grs, Gbs, Bs are illustrated respectively in source and intend the pixel value like corresponding red channel, the first green channel, the second green channel and blue channel in corresponding 2 × 2 block of pixels of pixel.And Ra, Gra, Gba, Ba are illustrated respectively in the first dynamic compensation result, the second dynamic compensation result, the 3rd dynamic compensation result and the 4th dynamic compensation result, intend the pixel value in correspondence position like the each pixel in corresponding 2 × 2 block of pixels of pixel with source.
Based on above-mentioned, the present invention is in the time carrying out noise elimination to the source image that meets the color queueing discipline of Bel's graph coloring, utilize at least one with reference to image to obtain the dynamic compensation result of corresponding different color passage, and utilize above-mentioned dynamic compensation result first source image to be carried out to time lamination process one time, then again the result being produced by time lamination process is carried out to space denoising point and process, effectively eliminate accordingly the negative effect that noise causes source image.
For above-mentioned feature and advantage of the present invention can be become apparent, special embodiment below, and coordinate accompanying drawing to be described in detail below.
Accompanying drawing explanation
Fig. 1 is the flow chart according to the shown noise point suppression method of one embodiment of the invention.
Fig. 2 is the schematic diagram according to the shown source of one embodiment of the invention image.
Fig. 3 is the schematic diagram of intending like image according to one embodiment of the invention shown first.
Fig. 4 is the flow chart that suppresses image according to shown generation the first rank noise of one embodiment of the invention.
Fig. 5 intends the schematic diagram like the block of pixels centered by pixel according to one embodiment of the invention are shown by source.
Fig. 6 is the graph of relation according to the shown mean absolute error of one embodiment of the invention and brightness initial value.
Fig. 7 intends like the pixel value of pixel and the graph of relation of luminance gain value according to the shown source of one embodiment of the invention.
Fig. 8 is the graph of relation according to the shown mean absolute error of one embodiment of the invention and superimposed weighted value of time.
Main element symbol description:
S110～S150: each step of the noise point suppression method described in one embodiment of the invention
200: source image
210_a, 220_a: block of pixels
Intend like image at 300: the first
210_b, 220_b: intend like pixel in source
S410～S480: generation the first rank noise described in one embodiment of the invention suppresses each step of image
A: the first preset value
B: the second preset value
M1, M2, M3, M4: mean absolute error
C: the 3rd preset value
D: the 4th preset value
Y1, Y2: the pixel value like pixel is intended in source
E: the first weight preset value
F: the second weight preset value
Embodiment
Fig. 1 is the flow chart according to the shown noise point suppression method of one embodiment of the invention.Noise point suppression method described in the present embodiment is to utilize at least one to carry out the noise in the image of filtering source with reference to image, the method is applicable to have charge coupled cell (Charge Coupled Device, or complementary matal-oxide semiconductor (Complementary Metal-OxideSemiconductor CCD), the digital image pick-up of Image Sensor such as CMOS), for example digital camera or digital code camera etc.Because can only detecting the intensity of light, Image Sensor cannot distinguish the wavelength of light, therefore can not tell color, therefore need to install a Bel (Bayer) filter additional before Image Sensor, to filter out respectively ruddiness, blue light and green glow, and then just can obtain the intensity of red, blue, green three looks.In other words, will remove in the source image of noise at the present embodiment, each pixel is the pixel value according to Bel's figure (Bayer pattern) color queueing discipline with corresponding red channel (R color channel), the first green channel (Gr colorchannel), the second green channel (Gb color channel) and one of them color channel of blue channel (B colorchannel).
Take the source image 200 shown in Fig. 2 as example, source image 200 has 8 × 6 pixels, and those pixels can be divided into the block of pixels that 12 sizes are 2 × 2 (for example block of pixels 210_a, 220_a).Each block of pixels comprises four pixels, and wherein top left pixel has the pixel value (being denoted as Gr in Fig. 2) of corresponding the first green channel, pixel value (being denoted as R in Fig. 2) that upper right pixel has corresponding red channel, pixel value (being denoted as B in Fig. 2) that lower-left pixel has corresponding blue channel and bottom right pixel has the pixel value (being denoted as Gb in Fig. 2) of corresponding the second green channel.In other embodiments, the pixel value that included upper left, upper right, lower-left and the bottom right pixel of each block of pixels has also can be distinguished corresponding red channel, the first green channel, the second green channel, and blue channel.
In the present embodiment, there is identical size in order to reference image and the source image of auxiliary filtering noise.Suppose that source image comprises n 2 × 2 block of pixels (n is positive integer), also comprise n 2 × 2 block of pixels with reference to image so, and be also the pixel value according to the color queueing discipline of Bel's graph coloring with corresponding red channel, the first green channel, the second green channel and one of them color channel of blue channel with reference to each pixel in each block of pixels of image.
Below explanation is utilized to the detailed step with reference to the noise of image filtering source image.First as shown in step S210, first of the corresponding source of generation image is intended intending like image with reference to second of image like image and correspondence.If originate, image comprises n 2 × 2 block of pixels, and first intends seemingly image comprises that n source plan is like pixel, and each source intended seemingly pixel corresponding 2 × 2 block of pixels of originating in image respectively.And the second plan comprises that like image n with reference to intending seemingly pixel, and respectively with reference to intending like pixel difference correspondence with reference to 2 × 2 block of pixels in image.In other words, if source image be W × H with reference to the size of image, first to intend intending seemingly the size of image like image and second can be (W/2) × (H/2) so.
Take the source image 200 shown in Fig. 2 as example, the first plan that corresponding source image 200 produces like image 300 as shown in Figure 3.Please refer to Fig. 2 and Fig. 3, first to intend like to intend like pixel 210_b be that block of pixels 210_a, first in the image 200 of corresponding source intends intending like pixel 220_b like the source in image 300 to the source in image 300 be the block of pixels 220_a in the image 200 of corresponding source, by that analogy.In this enforcement profit, first to intend intending like the pixel value of pixel like each source of image 300 can be the average of four pixel values in corresponding block of pixels.For example, to intend like the pixel value of pixel 210_b be that the pixel value of the included upper left of block of pixels 210_a, upper right, lower-left and these four pixels of bottom right is average in source.
Due to source image with reference to each pixel in image be respectively corresponding blue, redness and wherein a kind of color of green, therefore cannot directly utilize source image and calculate needed motion-vector while carrying out after a while motion compensation processing with reference to image.To this, the present embodiment will utilize first and second to intend calculating motion-vector like image.As shown in step S120, calculate first and intend intending like the universe motion-vector between image like image and second.Then in step S130, to carrying out dynamic compensation processing with reference to image, and then obtain multiple dynamic compensation results according to universe motion-vector.
The present invention is not limited the mode of calculating universe motion-vector.And carrying out dynamic compensation while processing, be shilling with reference to image according to universe motion-vector alignment source image.Particularly, suppose that universe motion-vector is for (p, q), with reference to move right p pixel move down q pixel of image.Then, from alignment result, isolate the pixel value of all corresponding red channels, to produce the first dynamic compensation result of corresponding red channel.Similarly, from alignment result, isolate respectively the pixel value of all corresponding the first green channels, the second green channel and blue channel, just can produce the 3rd dynamic compensation result of the second dynamic compensation result of corresponding the first green channel, corresponding the second green channel, and the 4th dynamic compensation result of corresponding blue channel.
Next in step S140, utilize above-mentioned dynamic compensation result to carry out the time superimposed (temporal blending) to source image and process to produce the first rank noise inhibition image.In the present embodiment, the time of carrying out can consider when lamination process to originate feature details and the brightness of image, and then it is superimposed to adopt different weighted values to carry out.The detailed step that produces the first rank noise inhibition image will explain in rear cooperation accompanying drawing again.
Finally, as shown in step S150, further the first rank noise is suppressed to image and carry out space denoising point (spatial noise reduction) processing, suppress image thereby produce second-order noise.The mode that the present invention does not process space denoising point is limited.
As shown in Figure 1, the present embodiment is to process by carrying out time lamination process and space denoising point the effect that reaches three-dimensional removal noise.Base this, for the digital image pick-up with high frame rate (frame rate) and high-res, just can mat each step as shown in Figure 1 and more effectively eliminate the noise in its captured image, and then promote image quality.
In the step S140 of Fig. 1, utilize multiple dynamic compensation results to carry out time lamination process to source image by illustrating with Fig. 4 below, produce according to this detailed step that the first rank noise suppresses image.In brief, be first to intend seemingly pixel for each source to produce a superimposed result in part that size is 2 × 2 pixels in the present embodiment, then recycle each source and intend producing the first rank noise inhibition image like the superimposed result in the corresponding part of pixel.
Refer to Fig. 4, first, as shown in step S410, all sources that comprise like image from the first plan intend intending like pixel like obtaining a source pixel.For instance, the present embodiment is sequentially to obtain source according to location of pixels to intend like pixel.For convenience of description, below, represent that with (x, y) current source to be processed intends like pixel in the first plan like the location of pixels in image.
Then as shown in step S420, intend intending like pixel like finding out the reference that there is correspondence position like pixel with the plan of originating in pixel in all references.That is to say, seemingly in image, find out the reference plan that is positioned at (x, y) this location of pixels like pixel in the second plan.
Next in step S430, intend, like pixel and with reference to intending like color and luminance difference distance between pixel, judging whether to intend carrying out time lamination process like corresponding 2 × 2 block of pixels of pixel to source according to source.This step be for fear of intend in source like pixel with reference to intending like the color between pixel and luminance difference when excessive, carry out improper lamination process and make the situation of the first rank noise inhibition image generation ghosts (ghosting).
In detail, first calculate source and intend intending like red color CRs, with reference to intending like chroma blue CBr like chroma blue CBs, source, and with reference to intending like red color CRr.Suppose to represent that with Ys source plan is illustrated in source like pixel value, the Bs of pixel and intends the pixel value like corresponding blue channel in corresponding 2 × 2 block of pixels of pixel, Rs is illustrated in source and intends the pixel value like corresponding red channel in corresponding 2 × 2 block of pixels of pixel, the present embodiment is the absolute difference of Ys and Bs to be defined as to source intend like chroma blue CBs, and the absolute difference of Ys and Rs is defined as to source plan like red color CRs.In addition, if represent to be illustrated in reference to intending and to be illustrated in reference to the pixel value of intending like corresponding red channel in corresponding 2 × 2 block of pixels of pixel like pixel value, the Rr of corresponding blue channel in corresponding 2 × 2 block of pixels of pixel like pixel value, the Br of pixel with reference to intending with Yr, the present embodiment can be defined as the absolute difference of Yr and Br with reference to intending like chroma blue CBr, and the absolute difference of Yr and Rr is defined as with reference to intending like red color CRr.
Then obtain corresponding source and intend brightness threshold value, chroma blue threshold value and the red color threshold value like pixel.Must special instruction, above-mentioned threshold value is not changeless numerical value, and can intend like the image feature of pixel and brightness and different along with current handled source.And how to obtain applicable brightness threshold value, chroma blue threshold value and red color threshold value will explain again in rear cooperation accompanying drawing.
When source plan is intended with reference like the pixel value Ys of pixel, the seemingly absolute difference of the pixel value Yr of pixel is less than brightness threshold value, source is intended seemingly chroma blue CBs and intended being less than chroma blue threshold value like the absolute difference of chroma blue CBr with reference, and when source intends intending being less than red color threshold value like the absolute difference of red color CRr with reference like red color CRs, represent source intend like pixel with reference to intending like color between pixel with luminance difference apart within permissible range, just judge in the case will to source plan seemingly the corresponding block of pixels of pixel carry out time lamination process.
Be not difficult to infer, suppress image in order to produce preferably the first rank noise, how to judge whether that carry out time lamination process to block of pixels just seems especially important.And the numerical value of brightness threshold value, chroma blue threshold value and red color threshold value height is the important parameter of the accuracy of the above-mentioned judgment mechanism of impact.
Below explanation is to intend like pixel to obtain the mode of corresponding brightness threshold value for the source of current processing.First,, intend obtaining the block of pixels of size as m × m like centered by pixel to originate like in image in the first plan, wherein m is positive integer.Then calculate again the mean absolute error (Mean Absolute Error, MAE) of the each pixel value in m × m block of pixels.Take m as 3 as example, shown in Fig. 5 is to originate to intend like centered by pixel (x, y) and the schematic diagram of the size block of pixels that is 3 × 3.In this example, the formula of mean absolute error MAE is as follows:
Wherein, N is quantity (in the embodiment shown in fig. 5, N is 9), the Ys of the included pixel of m × m block of pixels
irepresent the pixel value of each pixel in m × m block of pixels, and
represent the pixel average of all pixels in m × m block of pixels.The present embodiment by according to mean absolute error to obtain corresponding brightness initial value.
Fig. 6 is the graph of relation according to the shown mean absolute error of one embodiment of the invention and brightness initial value.As shown in Figure 6, the size of the first preset value a and the second preset value b is pre-defined, and for example can be relevant with image feature.For example, in the time of the numerical value less (being less than M1) of mean absolute error, represent that it is region comparatively smooth in image that corresponding source is intended like pixel, generally speaking the flat site in image is beneficial to removal noise, therefore in the time that mean absolute error is less than M1, corresponding brightness initial value is the second preset value b that numerical value is higher, and this can improve the chance of the time of carrying out lamination process base.Otherwise, for example, when mean absolute error larger (being greater than M2), represent that source intends may having as image features such as hard-edges (hard edge) like pixel.Easily produce ghost situation owing to hard-edge place being carried out to lamination process improperly, therefore, in the time that mean absolute error is greater than M2, corresponding brightness initial value is the first preset value a that numerical value is lower, reduces according to this chance of carrying out time lamination process.And mean absolute error between M1 and M2 can linear correspond to the brightness initial value between the first preset value a and the second preset value b.
Fig. 7 intends like the pixel value of pixel and the graph of relation of luminance gain value according to the shown source of one embodiment of the invention.As shown in Figure 7, predefined the 3rd preset value c for example can be relevant with image brilliance with the 4th preset value d.And in the time that source is for example intended, like the pixel value less (being less than Y1) of pixel, represent that its brightness is lower.Because easily there is more noise in the dark portion in image, in order to eliminate as much as possible those noises, in the time that source intends being less than Y1 like the pixel value of pixel, its corresponding luminance gain value can be the 4th preset value d that numerical value is higher, so just can promote the chance of the time of carrying out lamination process.Otherwise, in the time that source is for example intended, like the pixel value larger (being greater than Y2) of pixel, represent that it has higher brightness, if therefore source intends being greater than Y2 like the pixel value of pixel, its corresponding luminance gain value can be the 3rd preset value c that numerical value is lower, to reduce the chance of the time of carrying out lamination process.And intend like the pixel value of pixel between between Y1 and Y2 in source, its corresponding luminance gain value can be the numerical value between the 3rd preset value c and the 4th preset value d.
Obtaining respectively after brightness initial value and luminance gain value, the present embodiment will be intended the brightness threshold value like pixel with the product of luminance gain value as corresponding source using brightness initial value.
The present embodiment can be intended obtaining a chroma blue yield value like chroma blue CBs according to the source being calculated in similar above-mentioned mode, and using the product of corresponding mean absolute error brightness initial value and chroma blue yield value as chroma blue threshold value.The acquisition mode of red color threshold value is similar to the acquisition mode of chroma blue threshold value, intending obtaining after a red color yield value like red color CRs according to source, using the product of corresponding mean absolute error brightness initial value and red color yield value as red color threshold value.
Go back to Fig. 4, if the judged result of step S430 is not carry out time lamination process, then, as shown in step S440, directly intend using source intending the superimposed result in part like pixel like corresponding 2 × 2 block of pixels of pixel as correspondence source.
If the time of carrying out lamination process, as shown in step S450, by the each pixel value in the source plan block of pixels that seemingly the corresponding size of pixel is 2 × 2 and in the first dynamic compensation result, the second dynamic compensation result, the 3rd dynamic compensation result, and the 4th is superimposed in the pixel value of correspondence position in dynamic compensation result, produce according to this corresponding source and intend the superimposed result in part like pixel.In one embodiment, superimposed mode is the average of two pixel values that calculating location is corresponding.In another embodiment, can obtain corresponding source of processing at present and intend the superimposed weighted value of time like pixel, and produce local superimposed result according to superimposed weighted value of time.
Fig. 8 is the graph of relation according to the shown mean absolute error of one embodiment of the invention and superimposed weighted value of time.As shown in Figure 8, for example, when mean absolute error larger (being greater than M4), represent that source plan may have as image features such as hard-edges like pixel, suppress image in order to produce well-defined the first rank noise, the former mode of stating obtains after mean absolute error, if mean absolute error is greater than M4, its superimposed weighted value of corresponding time is the second weight preset value f that numerical value is higher.If mean absolute error is less than M3, its superimposed weighted value of corresponding time is the first weight preset value e that numerical value is lower.And if mean absolute error is between M3 and M4, can linear correspond to the superimposed weight of time between the first weight preset value e and the second weight preset value f.
The present embodiment is for example to calculate corresponding source with formula (1) to intend the superimposed result BR in part like pixel:
Wherein, Wc represents that superimposed weighted value of time, C_base represent default coefficient (for example 128 or 512, the present invention is not limited).Rs, Grs, Gbs, Bs are illustrated respectively in source and intend the pixel value like corresponding red channel, the first green channel, the second green channel and blue channel in corresponding 2 × 2 block of pixels of pixel.And Ra, Gra, Gba, Ba are illustrated respectively in the first dynamic compensation result, the second dynamic compensation result, the 3rd dynamic compensation result and the 4th dynamic compensation result, intend the pixel value in correspondence position like the each pixel in corresponding 2 × 2 block of pixels of pixel with source.
Intend seemingly image 300 take first of Fig. 3 and, as example, suppose that at present handled is that the first plan is intended like pixel 210_b like the source in image 300.Judging and will intend carrying out the time lamination process in the situation that like the corresponding block of pixels 210_a of pixel 210_b to source, when carrying out when superimposed the top left pixel in block of pixels 210_a (it has the pixel value of corresponding the first green channel), suppose that the location of pixels of this top left pixel in the image 200 of source is (1, 1) and pixel value be Grs, the present embodiment is by the second dynamic compensation result of corresponding the first green channel, find out and belong to (1, 1) the pixel value Gra of this location of pixels, then by Grs and Gra substitution formula (Wc × Grs+ (C_base-Wc) × Gra)/C_base, in the hope of the superimposed result of the top left pixel of block of pixels 210_a.Also can try to achieve in a similar fashion the superimposed result of upper right, lower-left and the bottom right pixel of block of pixels 210_a.These four superimposed results are that intend like the superimposed result in the corresponding part of pixel 210_b in source.
The present embodiment will perform step S420 repeatedly to step S470, to intend obtaining respectively the superimposed result in corresponding part like pixel for each source.Finally as shown in step S480, intend distinguishing the superimposed result in corresponding part like pixel according to the color queueing discipline of Bel's graph coloring each source of recombinating, to produce the first rank noise inhibition image.Wherein, the first rank noise inhibition image is identical with the size of source image.
Again intend like image 300 as example take first of Fig. 3, if represents the location of pixels in the first corresponding rank noise inhibition image with (α, β), wherein α between between 1 to 8 and β between 1 to 6.If (1, 1) be the location of pixels that the first rank noise suppresses the upper left corner of image, and (8, 6) be the location of pixels that the first rank noise suppresses image last cell, source plan will be according to the color queueing discipline of Bel's graph coloring respectively as being positioned at (1 like the superimposed result in the corresponding part of pixel 210_b so, 1), (1, 2), (2, 1), and (2, 2) pixel value of these four pixels, and source plan will be according to the color queueing discipline of Bel's graph coloring respectively as being positioned at (1 like the superimposed result in the corresponding part of pixel 220_b, 3), (1, 4), (2, 3), and (2, 4) pixel value of these four pixels, by that analogy.
Capture continuously the digital image pick-up of multiple images for meeting, utilize the each step shown in Fig. 1 and the second-order noise that the produces reference image when suppressing image and can be used as the noise of removing next Zhang Laiyuan image at every turn.In detail, producing according to first and second Zhang Laiyuan image after first second-order noise inhibition image, in the time preparing to remove the noise of the 3rd source image, just can first second-order noise suppress image as with reference to image, suppress image thereby obtain second second-order noise.And second second-order noise suppress image also can be in the time the 4th source image removed to noise processing as with reference to image.Thus, just can be issued at limited hardware resource environment the three-dimensional removal noise effect of stacked (cascade) formula.
Must specify, although be that with reference to image, as example is come, the present invention will be described take one in the above-described embodiments, but in other embodiments, also can adopt multiple to remove the noise in the image of source with reference to image.When using multiple during with reference to image, to calculate a source with aforementioned formula (1) if judge and intend the superimposed result in part like pixel, need (C_base-Wc) this part weighted value to divide equally to each with reference to image.Because step and the previous embodiment of utilizing multiple to produce second-order noise inhibition image with reference to image are same or similar, therefore do not repeat them here.
In sum, noise point suppression method of the present invention be utilize source image with reference to the relevance between image, analyze color information and successively carry out time lamination process and space denoising point is processed the effect that reaches three-dimensional removal noise.Wherein, can consider details and the bright dark degree of image in the time of carrying out when lamination process, thereby can avoid the superimposed image of providing ghost situation.And denoising point processing in space can also be eliminated noise further, produce accordingly preferably noise and remove result.
Although the present invention discloses as above with embodiment, so it is not in order to limit the present invention, and the those of ordinary skill in any affiliated technical field, when doing a little change and retouching, and does not depart from the spirit and scope of the present invention.
Claims (5)
1. a noise point suppression method, to utilize a noise of originating in image with reference to image filtering one, wherein this source image and this comprise respectively n 2 × 2 block of pixels with reference to image, and respectively each pixel in those 2 × 2 block of pixels is to have one of them pixel value of a corresponding red channel, one first green channel, one second green channel and a blue channel according to the color queueing discipline of Bel's graph coloring, and n is positive integer, and the method comprises:
Produce and intend like image and to intending like image with reference to of image second one first of the image of should originating, wherein this first plan comprises that like image n source plan is like pixel, and respectively those source intend like pixel respectively to those 2 × 2 block of pixels that should originate in image one of them, and this second intends comprising that like image n with reference to intending like pixel, and respectively those with reference to intend seemingly pixel respectively to should with reference to those 2 × 2 block of pixels in image one of them;
Calculate this first plan like image and the seemingly universe motion-vector between image of this second plan;
According to this universe motion-vector, this is carried out to a dynamic compensation processing to obtain multiple dynamic compensation results with reference to image;
Utilize those dynamic compensation results to carry out a time lamination process to this source image and suppress image to produce one first rank noise, comprising:
Intend like pixels for each those sources, those with reference to intend like in pixel, find out intend thering is correspondence position like pixel with this source this with reference to intending like pixel; And
According to this source intend like pixel with this with reference to intending like color and luminance difference distance between pixel, judge whether to this source plan seemingly corresponding 2 × 2 block of pixels of pixel carry out this time lamination process, comprising:
This source is intended intending intending seemingly chroma blue like the absolute difference of pixel value that should blue channel being defined as to a source in corresponding 2 × 2 block of pixels of pixel like the pixel value of pixel and this source;
This source is intended intending intending seemingly red color like the absolute difference of pixel value that should red channel being defined as to a source in corresponding 2 × 2 block of pixels of pixel like the pixel value of pixel and this source;
By this with reference to intend like the pixel value of pixel with this with reference to intending like in corresponding 2 × 2 block of pixels of pixel, the absolute difference of pixel value that should blue channel being defined as to one with reference to intending seemingly chroma blue;
By this with reference to intend like the pixel value of pixel with this with reference to intending like in corresponding 2 × 2 block of pixels of pixel, the absolute difference of pixel value that should red channel being defined as to one with reference to intending seemingly red color;
Obtain and intend a brightness threshold value, a chroma blue threshold value and the red color threshold value like pixel to originating, comprising:
, intend obtaining one m × m block of pixels like centered by pixel by this source like in image in this first plan, wherein m is positive integer;
Calculate a mean absolute error of the each pixel value in this m × m block of pixels;
Obtain a brightness initial value according to this mean absolute error;
Intend obtaining a luminance gain value like the pixel value of pixel according to this source;
Using the product of this brightness initial value and this luminance gain value as intend this brightness threshold value like pixel to originating;
Intend obtaining a chroma blue yield value like chroma blue according to this source;
Using the product of this brightness initial value and this chroma blue yield value as intend this chroma blue threshold value like pixel to originating;
Intend obtaining a red color yield value like red color according to this source; And
Using the product of this brightness initial value and this red color yield value as intend this red color threshold value like pixel to originating; And
When this source plan is less than this brightness threshold value like the pixel value of pixel and this with reference to intending the seemingly absolute difference of the pixel value of pixel, this source intends being less than this chroma blue threshold value like chroma blue and this with reference to the absolute difference of intending like chroma blue, and this source intend like red color with this when intending being less than this red color threshold value like the absolute difference of red color, judgement will be intended seemingly corresponding 2 × 2 block of pixels of pixel to this source carry out this time lamination process; And
This first rank noise is suppressed to image and carry out a space denoising point processing to produce a second-order noise inhibition image.
2. noise point suppression method according to claim 1, wherein carries out this dynamic compensation processing with reference to image and comprises with the step that obtains those dynamic compensation results this according to this universe motion-vector:
Make this with reference to image according to this universe motion-vector align this source image with produce respectively to should red channel, one first dynamic compensation result, one second dynamic compensation result, one the 3rd dynamic compensation result of this first green channel, this second green channel and this blue channel, and one the 4th dynamic compensation result.
3. noise point suppression method according to claim 2, wherein utilizes those dynamic compensation results to carry out this time lamination process to this source image and also comprises with the step that produces this first rank noise inhibition image:
When judging, this source is intended carrying out this when lamination process like corresponding 2 × 2 block of pixels of pixel time, this source is intended like the each pixel value in corresponding 2 × 2 block of pixels of pixel and in this first dynamic compensation result, this second dynamic compensation result, the 3rd dynamic compensation result, and the 4th is superimposed in the pixel value of correspondence position in dynamic compensation result, intend the local superimposed result like pixel to originating to produce;
This source is not intended carrying out this when lamination process like corresponding 2 × 2 block of pixels of pixel when judgement time, intend using this source like corresponding 2 × 2 block of pixels of pixel as intend the superimposed result in this part like pixel to originating; And
Respectively intend distinguishing the corresponding superimposed result in this part like pixel in this source according to the color queueing discipline restructuring of Bel's graph coloring, suppress image to produce this first rank noise.
4. noise point suppression method according to claim 3, wherein produces the plan of originating is also comprised like the step of this superimposed result in part of pixel:
Obtain a superimposed weighted value of time according to this mean absolute error; And
Produce and intend the superimposed result in this part like pixel to originating according to superimposed weighted value of this time.
5. noise point suppression method according to claim 4, wherein produces the plan of originating is comprised with formula (1) and being calculated like the step of this superimposed result in part of pixel according to superimposed weighted value of this time:
Wherein BR represents to intend originating the superimposed result in this part like pixel, and Wc represents that superimposed weighted value of this time, C_base represent a default coefficient,
Rs, Grs, Gbs, Bs be illustrated respectively in this source intend like in corresponding 2 × 2 block of pixels of pixel to should red channel, the pixel value of this first green channel, this second green channel and this blue channel,
Ra, Gra, Gba, Ba are illustrated respectively in this first dynamic compensation result, this second dynamic compensation result, the 3rd dynamic compensation result and the 4th dynamic compensation result, intend the pixel value in correspondence position like the each pixel in corresponding 2 × 2 block of pixels of pixel with this source.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW99147316A TWI459795B (en) | 2010-12-31 | 2010-12-31 | Noise reduction method |
| TW099147316 | 2010-12-31 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102547160A CN102547160A (en) | 2012-07-04 |
| CN102547160B true CN102547160B (en) | 2014-07-02 |
Family
ID=46352994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110094256.XA Expired - Fee Related CN102547160B (en) | 2010-12-31 | 2011-04-13 | Noise point suppression method |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN102547160B (en) |
| TW (1) | TWI459795B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1466844A (en) * | 2001-06-29 | 2004-01-07 | 成津C&C株式会社 | Method of adaptive noise smoothing/restoration in spatio-temporal domain and high-definition image capturing device thereof |
| WO2008002491A2 (en) * | 2006-06-27 | 2008-01-03 | Marvell World Trade Ltd. | Systems and methods for a motion compensated picture rate converter |
| CN101175219B (en) * | 2006-11-03 | 2010-11-03 | 兆宏电子股份有限公司 | Color interpolation method and image processing apparatus using the same |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI423666B (en) * | 2008-12-31 | 2014-01-11 | Altek Corp | Image elimination method for image sequence |
-
2010
- 2010-12-31 TW TW99147316A patent/TWI459795B/en active
-
2011
- 2011-04-13 CN CN201110094256.XA patent/CN102547160B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1466844A (en) * | 2001-06-29 | 2004-01-07 | 成津C&C株式会社 | Method of adaptive noise smoothing/restoration in spatio-temporal domain and high-definition image capturing device thereof |
| WO2008002491A2 (en) * | 2006-06-27 | 2008-01-03 | Marvell World Trade Ltd. | Systems and methods for a motion compensated picture rate converter |
| CN101175219B (en) * | 2006-11-03 | 2010-11-03 | 兆宏电子股份有限公司 | Color interpolation method and image processing apparatus using the same |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102547160A (en) | 2012-07-04 |
| TWI459795B (en) | 2014-11-01 |
| TW201228346A (en) | 2012-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI484818B (en) | Techniques for synchronizing audio and video data in an image signal processing system | |
| US10205924B2 (en) | Method and system of lens shading color correction using block matching | |
| TWI433532B (en) | System and method for processing image data using an image signal processor having back-end processing logic | |
| TWI526066B (en) | System ,method and electronic device for processing an image signal | |
| TWI492609B (en) | Image signal processor line buffer configuration for processing raw image data | |
| US8295595B2 (en) | Generating full color images by demosaicing noise removed pixels from images | |
| CN103856767B (en) | Method and apparatus for processing image | |
| US7236191B2 (en) | Method and system for image processing with pixel interpolation using second order gradients | |
| CN101272503B (en) | Gridding noise elimination method and device for remaining image definition | |
| CN101069416B (en) | Artifact reduction in a digital video | |
| CN104380727B (en) | Image processing apparatus and image processing method | |
| CN102859987A (en) | Combining data from multiple image sensors | |
| CN102859989A (en) | Combining data from multiple image sensors | |
| US20120182452A1 (en) | Image processing device, image processing method, and program | |
| CN103873836B (en) | Image processing equipment and image processing method | |
| US7609300B2 (en) | Method and system of eliminating color noises caused by an interpolation | |
| CN103237158A (en) | Image processing method, image acquisition system and image processor | |
| JP4118906B2 (en) | Video signal processing apparatus and video signal processing method | |
| US20210076017A1 (en) | Image processing apparatus, imaging apparatus, image processing method, and program | |
| CN105678718B (en) | Image de-noising method and device | |
| US8385671B1 (en) | Digital camera and method | |
| CN102547160B (en) | Noise point suppression method | |
| CN103747222A (en) | Method and apparatus for suppressing color artifact of image | |
| US7885488B2 (en) | Image processing apparatus, method and medium | |
| US7714900B2 (en) | Color interpolation processing method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20161228 Address after: Chinese Taiwan Hsinchu Science Industrial Park Hsinchu City Road No. 12 5 floor Patentee after: Altek Semiconductor Corp. Address before: Chinese Taiwan Hsinchu City Science Park force Road No. 12 Patentee before: Altek Corp. |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20140702 |