CN103491280A - Bayer image united denoising interpolation method - Google Patents
Bayer image united denoising interpolation method Download PDFInfo
- Publication number
- CN103491280A CN103491280A CN201310463616.8A CN201310463616A CN103491280A CN 103491280 A CN103491280 A CN 103491280A CN 201310463616 A CN201310463616 A CN 201310463616A CN 103491280 A CN103491280 A CN 103491280A
- Authority
- CN
- China
- Prior art keywords
- passage
- interpolation
- delta
- denoising
- summation
- 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.)
- Granted
Links
Images
Landscapes
- Image Processing (AREA)
- Color Television Image Signal Generators (AREA)
Abstract
The invention provides a Bayer image united denoising interpolation method. Compared with an existing united denoising interpolation method, the Bayer image united denoising interpolation method has the advantages that denoising and interpolation are conducted in a small local window, and the noise information of an overall image is not needed; the local correlation is considered, transverse filtering and longitudinal filtering are conducted simultaneously, and therefore the denoising performance is improved. According to the Bayer image united denoising interpolation method, an overall algorithm is simple, only simple addition, subtraction, multiplication and division operations are related, most of multiplication and division operations can be replaced by manual operations, achievement of hardware is facilitated, resources are saved, and the algorithm performance is good.
Description
Technical field
The present invention relates to a kind of method that digital camera still image is processed, relate in particular to a kind of associating denoising interpolation processing of carrying out on bayer images, the method for the RGB image after the generation denoising.
Background technology
For realizing the monolithic colour imageing sensor, people introduce chromatic filter array CFA (Color Filter Array), on the basis of black and white cmos image sensor, increase color filter structure and colour information processing module and just can obtain coloured image.Cover such color filter array on the pixel of cmos image sensor, just can obtain the colour information of image, then process through colour information, just can obtain color coloured image true to nature.Fig. 1 has meaned a 8*8 color filters array, and this is that a typical checkerboard type colour filter is the Bayer colour filter.X, Y-axis has meaned the coordinate of each sensor devices.The B(blueness), G(green) former color device is staggered in odd-numbered line, G, R(redness) former color device is staggered in even number line.Only have two kinds of color-filter units on every a line of this colour filter: or G, R, or G, B.Therefore, on whole colour filter, the sampling unit number of G light is the twice of R light or B light.The mode of lining by line scan that can be advantageously used in this colour filter realizes coloured image effect true to nature.By top color filter structure, can be known, to some pixels, it has obtained the some values in two primary colours, and all the other two values will obtain from the neighborhood pixels interpolation.The process of interpolation is also referred to as demosaicing.
Bilinear interpolation, COk color ratio law interpolation method, Hamilton-Adams interpolation method, DLMMSE, LPA-ICI etc. that current interpolation method has bilinear interpolation, detects with direction.
General from transducer, bayer images (Bayer image) data out can contain noise, and wherein additive white Gaussian noise is a kind of typical representative, to the denoising of image, is therefore also a very important step handling process.General first denoising again interpolation than first interpolation, the performance of denoising is good again, but denoising and interpolation are joined together to do effect than separately carrying out as a step flow process.
Existing bayer images associating denoising interpolation algorithm performance is fine, but algorithm complex is very high, be unfavorable for the hardware realization, and the hardware consumption of realizing can be very large.The present invention is directed to above-mentioned deficiency, propose a kind of new bayer images associating denoising interpolation method, the method performance is suitable with existing algorithm, and complexity reduces greatly, is beneficial to hardware and realizes, hardware spending is little.
Summary of the invention
The invention is intended to address the above problem, i.e. bayer images associating denoising interpolation algorithm complexity is very high, is unfavorable for the hardware realization.The present invention is improved on existing interpolation, denoising method, proposes a kind of new bayer images associating denoising interpolation method, and the method performance is suitable with existing algorithm, and complexity reduces greatly, is beneficial to hardware and realizes, hardware spending is little.
In order to achieve the above object, the present invention has adopted following technical scheme:
1) the G passage is carried out to interpolation.Suppose that original bayer images size is M*N, and the current point coordinates that need to carry out interpolation be (m, n), this point is B or R.Figure 2 shows that the situation that current point is B or R (BR means a kind of color, or is full B, or is full R);
G_v(m,n)=(I(m-1,n)+I(m+1,n))/2+(2*I(m,n)-I(m-2,n)-I(m+2,n))/4；
G_h(m,n)=(I(m,n-1)+I(m,n+1))/2+(2*I(m,n)-I(m,n-2)-I(m,n+2))/4；
Wherein I (m, n) means original Bayer data, and G_v and G_h represent respectively vertical interpolation image and lateral interpolation image.
Formed like this complete vertical interpolation image of M*N by G_v and original G pixel.
Formed equally the complete lateral interpolation image of M*N by G_h and original G pixel.
2) for the interpolation of B and R, thus the complete plane of interpolation of complete B or R can not as G, be obtained because pixel is few, but backward forms the image that B and R interweave.The current point coordinates that need to carry out interpolation is (m, n), and this point is G, as shown in Figure 2.
Interpolation Process and G are just the same:
BR_v(m,n)=(I(m-1,n)+I(m+1,n))/2+(2*G_v(m,n)-G_v(m-1,n)-G_v(m+1,n))/2;
BR_h(m,n)=(I(m,n-1)+I(m,n+1))/2+(2*G_h(m,n)-G_h(m,n-1)-G_h(m,n+1))/2;
Vertical interpolation image that interweaves of the M*N that is a row B and a row R that BR_v obtains like this.
Vertical interpolation image that interweaves of the M*N that is a line B and a line R that same BR_h obtains.
3) conversion G and BR territory are to the Summation(summation) and Delta territory (territory, delta).Summation be G+B or G+R and, Delta is the poor of G-B or G-V.
summation_h=G_h+BR_h;
summation_v=G_v+BR_v;
delta_h=G_h-BR_h;
delta_v=G_v-BR_v;
The variation of Delta numeric field data is more slowly with level and smooth, so interpolation noise obtained etc. is all relatively less.In addition, can be so that can use G and B or R in next denoising step simultaneously, like this can be so that more enough more accurate when denoising.
4) employing Epsilon-Filter(Epsilon filter) carry out denoising for the first time.The Epsilon-Filter principle is relatively simple, when the excessive threshold value that surpasses of pixel difference of pixel to be processed and neighborhood (laterally adjacent 5 pixels), is judged to be uncorrelated, otherwise relevant.In the later stage weighted sum, these incoherent some weights is 0, the inverse that relevant some weights are all reference point numbers.Wherein the threshold value in Summation territory is T1, and the weights in Delta territory are T2.Obtain corresponding summation_denoise_h, summation_denoise_v and delta_denoise_h, delta_denoise_v after denoising.
5) by interpolation, the image of BR is inserted into to the independently complete image of B and R of 2 width.For example, concerning R, because the BR image itself is exactly that two row R clip a line B, vice versa.So just can adopt upper and lower two row additions to be averaging for that row R value of centre obtains.Consider the original raw data(untreatment data that half is arranged in a line R), and remaining half is the interpolation number that the first step obtains, so adopt following method to obtain being worth more accurately.That is:
A. when being raw data, 2 of the left and right of a point adopt 4 nearest some weighted averages of its diagonal.Because consider 4 of diagonal be all first step interpolation out, so these four essence are subject to the impact of 6 adjacent with it points, and middle 2 weights are 2 times of 4, border.
If b. 2 of the left and right of a point be interpolation out directly use 2 of left and right average all right, its essence is equivalent to 4 some weighted averages of four jiaos after launching.
6) second denoising.In full accord with the denoising for the first time of the 4th step, different is obtaining result for lateral interpolation and carry out horizontal filtering and change vertical filtering (vertically adjacent 5 pixels) into originally, thereby the every row of passivation and every row between saltus step.
7) transverse pattern and vertically pattern merging, adopt the most simply point of both correspondences to average.
With the existing denoising interpolation method of combining, compare, the inventive method has following advantage: carry out denoising and interpolation in a local wicket, do not need to know the noise information of global image; Denoising method is considered local correlations, and carries out horizontal and vertical and filtering simultaneously, increases denoising performance.Total algorithm is simple, only relates to simple addition subtraction multiplication and division computing, and most of multiplication and division computing can, with thinking that operation replaces, be beneficial to hardware and realize, saving resource; Algorithm performance is good.
The accompanying drawing explanation
Fig. 1 is the bayer images color filters array of a 8*8.
Fig. 2 is current point coordinates for (m, n) and for B or R(B, R mean a kind of color, or is full B, or entirely is R) time, the distribution situation of surrounding pixel point.
Embodiment
For making content of the present invention more clear understandable, below in conjunction with Figure of description, content of the present invention is described further.Certainly the present invention is not limited to this specific embodiment, and the known general replacement of those skilled in the art also is encompassed in protection scope of the present invention.
Below embodiments of the invention are described in detail, the present embodiment is implemented take technical solution of the present invention under prerequisite, provided detailed execution mode and concrete operating process, but protection scope of the present invention is not limited to following embodiment.
The present embodiment is realized bayer images associating denoising and interpolation in the following manner.
(1) Bayer format-pattern is as shown in Figure 1 carried out to interpolation.The bayer images size is M*N, and the current point coordinates that need to carry out interpolation is (m, n), and this point is G, B or R.First the G Color Channel is carried out to interpolation, as shown in Figure 2, the situation that current point is B or R (BR means a kind of color, or is full B, or is full R).
G_v(m,n)=(I(m-1,n)+I(m+1,n))/2+(2*I(m,n)-I(m-2,n)-I(m+2,n))/4;
G_h(m,n)=(I(m,n-1)+I(m,n+1))/2+(2*I(m,n)-I(m,n-2)-I(m,n+2))/4;
Wherein I (m, n) means original Bayer data, and G_v and G_h represent respectively vertical interpolation image and lateral interpolation image.
Formed like this complete vertical interpolation image of M*N by G_v and original G pixel.
Formed equally the complete lateral interpolation image of M*N by G_h and original G pixel.
Again B and R Color Channel are carried out to interpolation, the current point coordinates that now needs to carry out interpolation is (m, n), and this point is G, similar with Fig. 2.
Interpolation Process and G are just the same:
BR_v(m,n)=(I(m-1,n)+I(m+1,n))/2+(2*G_v(m,n)-G_v(m-1,n)-G_v(m+1,n))/2;
BR_h(m,n)=(I(m,n-1)+I(m,n+1))/2+(2*G_h(m,n)-G_h(m,n-1)-G_h(m,n+1))/2;
Vertical interpolation image that interweaves of the M*N that is a row B and a row R that BR_v obtains like this.
Vertical interpolation image that interweaves of the M*N that is a line B and a line R that same BR_h obtains.
(2) conversion G and BR territory are to Summation and Delta territory.Summation be G+B or G+R and, Delta is the poor of G-B or G-V.Conversion method is as follows.
summation_h=G_h+BR_h;
summation_v=G_v+BR_v;
delta_h=G_h-BR_h;
delta_v=G_v-BR_v;
(3) adopt Epsilon-Filter to carry out denoising for the first time.When the excessive threshold value that surpasses of pixel difference of pixel to be processed and neighborhood (laterally adjacent 5 pixels), be judged to be uncorrelated, otherwise relevant.In the later stage weighted sum, these incoherent some weights is 0, the inverse that relevant some weights are all reference point numbers.Wherein the threshold value in Summation territory is T1, and the weights in Delta territory are T2.Obtain corresponding summation_denoise_h, summation_denoise_v and delta_denoise_h, delta_denoise_v after denoising.Corresponding algorithm is as follows.
relevant=abs(summation_h(m,n-2:n+2)-summation_h(m,n))<T1;
summation_denoise_h(m,n)=sum(summation_h(m,n-2:n+2).*relevant)/sum(relevant);
relevant=abs(delta_h(m,n-2:n+2)-delta_h(m,n))<T2;
delta_denoise_h(m,n)=sum(delta_h(m,n-2:n+2).*relevant)/sum(relevant);
Wherein, abs means signed magnitude arithmetic(al), and sum means summation operation.
(4) by interpolation, the image of BR is inserted into to the independently complete image of B and R of 2 width.For example, concerning R, because the BR image itself is exactly that two row R clip a line B, vice versa.So just can adopt upper and lower two row additions to be averaging for that row R value of centre obtains.Consider the original raw data that half is arranged in a line R, and remaining half is the interpolation number that the first step obtains, so adopt following method to obtain being worth more accurately.That is:
A. when being rawdata, 2 of the left and right of a point adopt 4 nearest some weighted averages of its diagonal.Because consider 4 of diagonal be all first step interpolation out, so these four essence are subject to the impact of 6 adjacent with it points, and middle 2 weights are 2 times of 4, border.
If b. 2 of the left and right of a point be interpolation out directly use 2 of left and right average all right, its essence is equivalent to 4 some weighted averages of four jiaos after launching.
Corresponding is as follows for algorithm.
Current point for the odd column odd-numbered line:
delta_R_v(m,n)=(delta_denoise_v(m,n-1)+delta_denoise_v(m,n+1))/2;
Current point for the odd column even number line:
delta_R_v(m,n)=(delta_denoise_v(m-1,n-1)+delta_denoise_v(m+1,n+1)+delta_denoise_v(m+1,n-1)+delta_denoise_v(m-1,n+1))/4;
Other delta_R_h, delta_B_v, delta_B_h can similarly obtain.
(5) second denoising.In full accord with denoising for the first time, different is obtaining result for lateral interpolation and carry out horizontal filtering and change vertical filtering (vertically adjacent 5 pixels) into originally, thereby the every row of passivation and every row between saltus step.
(6) transverse pattern and vertically pattern merge, addition is averaging the image obtained after final denoising and interpolation.
Although the present invention discloses as above with preferred embodiment; so it is not in order to limit the present invention; have and usually know the knowledgeable in technical field under any; without departing from the spirit and scope of the present invention; when doing a little change and retouching, so protection scope of the present invention is as the criterion when looking claims person of defining.
Claims (2)
1. a bayer images associating denoising interpolation method comprises:
The G passage is carried out to interpolation:
If original bayer images size is M*N, and the current point coordinates that need to carry out interpolation is (m, n),
G_v(m,n)=(I(m-1,n)+I(m+1,n))/2+(2*I(m,n)-I(m-2,n)-I(m+2,n))/4，
G_h(m,n)=(I(m,n-1)+I(m,n+1))/2+(2*I(m,n)-I(m,n-2)-I(m,n+2))/4，
Wherein I (m, n) means the data of original bayer images, and G_v and G_h represent respectively and need to carry out the vertical interpolation image of interpolation and lateral interpolation image;
B passage and R passage are carried out to interpolation:
BR_v(m,n)=(I(m-1,n)+I(m+1,n))/2+(2*G_v(m,n)-G_v(m-1,n)-G_v(m+1,n))/2，
BR_h(m,n)=(I(m,n-1)+I(m,n+1))/2+(2*G_h(m,n)-G_h(m,n-1)-G_h(m,n+1))/2；
Conversion G passage and B passage, R passage be to summation and territory, delta, described summation be G passage and B passage with or G passage and R passage with, described delta be the poor of the difference of G passage and B passage or G passage and V passage:
summation_h=G_h+BR_h，
summation_v=G_v+BR_v，
delta_h=G_h-BR_h，
delta_v=G_v-BR_v，
Described summation_h is vertical summation, and summation_v is horizontal summation, and delta_h is territory, vertical delta, and delta_v is territory, horizontal delta;
Adopt the Epsilon filter to carry out denoising for the first time, the excessive threshold value that surpasses of pixel difference when pixel to be processed and neighborhood (laterally adjacent 5 pixels), be judged to be uncorrelated, otherwise relevant, in the later stage weighted sum, these incoherent some weights is 0, the inverse that relevant some weights are all reference point numbers;
By interpolation, the image of B passage, R passage is inserted into to the independently complete image of B and R of two width, to the R passage, adopt its upper and lower two row additions to be averaging acquisition the value of centre row R passage, for the B passage, adopt its upper and lower two row additions to be averaging acquisition the value of centre row B passage;
Adopt the Epsilon filter to carry out denoising for the second time;
By transverse pattern and vertically pattern merging, described transverse pattern and vertical point corresponding to pattern are averaged.
2. bayer images associating denoising interpolation method as claimed in claim 1, is characterized in that, by interpolation the image of B passage, R passage be inserted into two width independently the step of the complete image of B and R comprise:
When being untreatment data, 2 of the left and right of a point adopt 4 nearest some weighted averages of its diagonal, because consider 4 of diagonal be all first step interpolation out, so these four essence are subject to the impact of 6 adjacent with it points, and middle 2 weights are 2 times of 4, border;
If 2 of the left and right of a point be interpolation out directly use 2 of left and right average all right, its essence is equivalent to 4 some weighted averages of four jiaos after launching.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310463616.8A CN103491280B (en) | 2013-09-30 | 2013-09-30 | A kind of bayer images associating denoising interpolation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310463616.8A CN103491280B (en) | 2013-09-30 | 2013-09-30 | A kind of bayer images associating denoising interpolation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103491280A true CN103491280A (en) | 2014-01-01 |
| CN103491280B CN103491280B (en) | 2016-01-20 |
Family
ID=49831220
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310463616.8A Expired - Fee Related CN103491280B (en) | 2013-09-30 | 2013-09-30 | A kind of bayer images associating denoising interpolation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103491280B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9727947B2 (en) | 2015-03-23 | 2017-08-08 | Microsoft Technology Licensing, Llc | Downscaling a digital raw image frame |
| US9773297B2 (en) | 2015-05-15 | 2017-09-26 | SZ DJI Technology Co., Ltd. | System and method for supporting image denoising based on neighborhood block dimensionality reduction |
| CN107392856A (en) * | 2016-05-16 | 2017-11-24 | 合肥君正科技有限公司 | A kind of image filtering method and its device |
| CN108053383A (en) * | 2017-12-28 | 2018-05-18 | 努比亚技术有限公司 | A kind of noise-reduction method, equipment and computer readable storage medium |
| CN110324541A (en) * | 2019-07-12 | 2019-10-11 | 上海集成电路研发中心有限公司 | The filtration combined denoising interpolation method of one kind and device |
| CN114004769A (en) * | 2021-12-30 | 2022-02-01 | 江苏游隼微电子有限公司 | Discrete weight-based Bayer denoising color interpolation method |
| WO2022126516A1 (en) * | 2020-12-17 | 2022-06-23 | Covidien Lp | Adaptive image noise reduction system and method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1933550A (en) * | 2005-05-26 | 2007-03-21 | 智多微电子（上海）有限公司 | Bell field signal image processing method and module |
| CN101821772A (en) * | 2007-08-07 | 2010-09-01 | 德克索实验室 | Be used to handle the method and the related system thereof of digital object |
| WO2011049781A1 (en) * | 2009-10-20 | 2011-04-28 | Apple Inc. | System and method for demosaicing image data using weighted gradients |
| JP5268321B2 (en) * | 2007-10-16 | 2013-08-21 | シリコン ヒフェ ベー．フェー． | Image processing apparatus, image processing method, and image processing program |
-
2013
- 2013-09-30 CN CN201310463616.8A patent/CN103491280B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1933550A (en) * | 2005-05-26 | 2007-03-21 | 智多微电子（上海）有限公司 | Bell field signal image processing method and module |
| CN101821772A (en) * | 2007-08-07 | 2010-09-01 | 德克索实验室 | Be used to handle the method and the related system thereof of digital object |
| JP5268321B2 (en) * | 2007-10-16 | 2013-08-21 | シリコン ヒフェ ベー．フェー． | Image processing apparatus, image processing method, and image processing program |
| WO2011049781A1 (en) * | 2009-10-20 | 2011-04-28 | Apple Inc. | System and method for demosaicing image data using weighted gradients |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9727947B2 (en) | 2015-03-23 | 2017-08-08 | Microsoft Technology Licensing, Llc | Downscaling a digital raw image frame |
| US9773297B2 (en) | 2015-05-15 | 2017-09-26 | SZ DJI Technology Co., Ltd. | System and method for supporting image denoising based on neighborhood block dimensionality reduction |
| US10026154B2 (en) | 2015-05-15 | 2018-07-17 | SZ DJI Technology Co., Ltd. | System and method for supporting image denoising based on neighborhood block dimensionality reduction |
| US10515438B2 (en) | 2015-05-15 | 2019-12-24 | SZ DJI Technology Co., Ltd. | System and method for supporting image denoising based on neighborhood block dimensionality reduction |
| CN107392856A (en) * | 2016-05-16 | 2017-11-24 | 合肥君正科技有限公司 | A kind of image filtering method and its device |
| CN108053383A (en) * | 2017-12-28 | 2018-05-18 | 努比亚技术有限公司 | A kind of noise-reduction method, equipment and computer readable storage medium |
| CN110324541A (en) * | 2019-07-12 | 2019-10-11 | 上海集成电路研发中心有限公司 | The filtration combined denoising interpolation method of one kind and device |
| CN110324541B (en) * | 2019-07-12 | 2021-06-15 | 上海集成电路研发中心有限公司 | Filtering joint denoising interpolation method and device |
| WO2022126516A1 (en) * | 2020-12-17 | 2022-06-23 | Covidien Lp | Adaptive image noise reduction system and method |
| CN114004769A (en) * | 2021-12-30 | 2022-02-01 | 江苏游隼微电子有限公司 | Discrete weight-based Bayer denoising color interpolation method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103491280B (en) | 2016-01-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103491280B (en) | A kind of bayer images associating denoising interpolation method | |
| CN101006731B (en) | Image processing device and method, and imaging device | |
| CN103327220B (en) | With green channel for the denoising method guided on low-light (level) Bayer image | |
| CN105247568B (en) | The method and apparatus for generating improved color image with the sensor with coloured filter | |
| CN110730336B (en) | Demosaicing method and device | |
| JP2005136766A (en) | Image processor and image processing method | |
| CN102884796A (en) | Color imaging device | |
| CN102292975A (en) | Solid state imaging element, camera system and method for driving solid state imaging element | |
| TWI547169B (en) | Image processing method and module | |
| CN101399996A (en) | Color filter lens array interpolation method | |
| EP2680592B1 (en) | Color imaging device | |
| CN103905802A (en) | Method and device for mosaic removal based on P-mode color filter array | |
| CN110430403B (en) | Image processing method and device | |
| CN102106150A (en) | Imaging processor | |
| US8743246B2 (en) | Color imaging device | |
| CN103841388A (en) | Mosaic removing method and device | |
| CN113068011B (en) | Image sensor, image processing method and system | |
| US20150055861A1 (en) | Methods and Systems for Image Demosaicing | |
| CN104680484A (en) | Image enhancement method and device | |
| CN116503259B (en) | Mosaic interpolation method and system | |
| US20120127347A1 (en) | Image processing apparatus and system | |
| Fan et al. | Constant-hue-based color filter array demosaicking sensor for digital still camera implementation | |
| CN103259960B (en) | The interpolation method of data and device, image output method and device | |
| US7978908B2 (en) | Color signal interpolation system and method | |
| CN113242413B (en) | Interpolation calculation method and system for anti-sawtooth RCCB filter array |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160120 Termination date: 20180930 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |