CN104954772B - Image adjacent-grey pixel selection algorithm applied to automatic white balance algorithm - Google Patents

Abstract

The invention provides an image adjacent-grey pixel selection algorithm applied to an automatic white balance algorithm. The image adjacent-grey pixel selection algorithm comprises steps as follows: S1, pixels in a target image are read, and YCbCr values of the pixels are acquired; S2, the YCbCr values of the pixels in the image are substituted into eight inequalities; S3, adjacent-grey points are selected from the image: if the YCbCr values of the pixels of the image meet the eight inequalities, the pixels are thought to be adjacent-grey pixels and sent to a follow-up algorithm module for color temperature estimation; if the YCbCr values of the pixels of the image do not meet any of the eight inequalities, the pixels are not adjacent-grey pixels and are not processed any longer. According to the image adjacent-grey pixel selection algorithm, a group of new inequalities are put forward, so that most pixels in three grey color lumps with highest brightness can be selected from a standard color card image at various color temperatures and various illuminance levels, besides, the phenomenon that a large quantity of pixels in a colored color lump are mistakenly selected is effectively avoided, and more comprehensive and more accurate original color difference information is provided for the follow-up algorithm.

Description

A kind of nearly gray pixels Algorithms of Selecting of the image for being applied to AWB algorithm

Technical field

The present invention relates to a kind of pixel Algorithms of Selecting, specifically a kind of image for being applied to AWB algorithm is near Gray pixels Algorithms of Selecting, belongs to technical field of image processing.

Background technology

Different light sources has different spectral component and distribution, and this is referred to as colour temperature in colorimetry.The wavelength of light is got over Short, colour temperature is higher；Wavelength is longer, and colour temperature is lower.In the yellowish green ultramarine purple seven-colour-light of common blood orange, from left to right wavelength drops successively Low, colour temperature increases successively.The object of one white, can be partially red under the light irradiation of low colour temperature, and the light in high color temperature shines Can be partially blue under penetrating, the aberration that referred to as colour temperature causes.

Human eye evolution has gone out adaptability, makes us to find the change of colour temperature under normal circumstances.For example, people is in tengsten lamp Stop for a long time under (luminous colour temperature is low), can't think that the blank sheet of paper under tengsten lamp is partially red, if fluorescent lamp is changed suddenly For tungsten lighting. will feel that the color for finding blank sheet of paper is partially red, but this sensation is also merely able to continue a little while, human eye (bag Include associated color perception system) chromatic aberration correction that this colour temperature causes can be come during this period.

Video camera (note：" video camera " described in this patent refers to DV, provides the number of automatic white balance function Camera) " IMAQ -- processing system " do not possess this adaptability of human eye.If not according to the colour temperature of scenery illumination Targetedly adjusted, the image that video camera is obtained will occur colour cast.Therefore the concept of white balance, Bai Ping have been occurred as soon as The purpose of weighing apparatus is exactly：For various colour temperature conditions, the aberration that colour temperature causes is offset by the color computing inside video camera, make bat Next image is taken out closer to the visual custom of human eye.

Simply white balance can be interpreted as：Under the conditions of any colour temperature, the reference white Jing of video camera photographic subjects The adjustment of oversampling circuit, remains as white after imaging.

AWB is the side by video camera to obtain color temperature information automatically, processed image automatically according to colour temperature The white balance that formula is realized.

Automatically the approach for obtaining color temperature information is divided into two classes：The first kind is the equipment colour temperature measurement part in video camera, real When obtain color temperature information, this scheme is due to relatively costly and be of limited application；Equations of The Second Kind is by the computing core in video camera Piece (CPU, DSP, FPGA, ASIC etc.) performs algorithm, and the image that shooting is obtained is analyzed, and therefrom obtains color temperature information.This Patent AWB algorithm indication referenced below is the algorithm for obtaining information approach automatically based on Equations of The Second Kind colour temperature.

AWB algorithm generally includes three below basic step：

1) color temperature estimation

Most classical color temperature estimation algorithm is Grey-world algo-rithms.This algorithm based on the assumption that：For a web has Enough for the image of color change, the average reflection of whole scene can offset aberration.Most basic method is exactly to calculate whole The average color difference of image.But, if color of image is more single, the colour temperature that algorithm above is tried to achieve will be very inaccurate.For This, it is necessary to according to certain constraints, select suitable pixel to calculate aberration, improve the accuracy of color temperature estimation.

" suitable pixel " mentioned here typically refers to " nearly gray pixels ", and this is current nearly all AWB The common choice of algorithm.Nearly gray pixels refer to the pixel that R, G, B tristimulus values are more or less the same, and color temperature estimation algorithm assumes nearly grey Pixel is obtained by gray pixels under colour cast light irradiation.Deviateed by statistics, the nearly gray pixels rgb value calculated in image Pure gray pixels rgb value (R=G=B), the degree of CbCr values (Cb=0, Cr=0) estimating colour temperature, in actual operation process In, generally R, G, B value is converted to into Y, Cb, Cr value, to facilitate analysis.

2) gain is calculated

It is on the basis of color temp is estimated, by certain method channel gain to be obtained that gain is calculated.Channel gain Namely the color temperature correction factor, typically there is blue and red two components, and image blueness and red channel are adjusted Amplitude.

Cb, Cr of gray pixels is 0, therefore, Cb, Cr can be exactly adjusted to 0 by channel gain (or two of close 0) Coefficient μ and ν.The method that gain is calculated has various, such as look-up table, iterative method etc..Look-up table is to count a table in advance, The channel gain corresponding to different colour temperatures is recorded, with fireballing advantage, but due to the finite capacity of table, it is impossible to realize each Plant continuously adjusting for colour temperature situation.Iterative algorithm is then the relation according to Cb, Cr, constantly regulate μ, ν, according to after adjusting every time To Cb, Cr value determine the amount that adjusts next time, until Cb, Cr are adjusted to into close 0.

3) color temperature correction

Color temperature correction is exactly to be multiplied by respective gain in the blueness and red channel of image, so as to adjust the color of R, G, B tri- Ratio (or value of Cb, Cr).It is generally directed in video camera be used for the CCD, the CMOS chip that gather image raw information (i.e. Sensor) rgb signal of sampling output is carried out.

In the color temperature estimation step of AWB algorithm, the algorithm that nearly gray pixels are chosen from image is crucial ring One of section.A kind of classical nearly gray pixels Algorithms of Selecting is a three-dimensional structure defined in YCbCr space, referred to as near ash Color region, it is all to fall into pixel therein and be regarded as nearly grey, and the pixel outside it then thinks non-near grey.Calculating During aberration, the average color difference of nearly gray pixels is only calculated, to substitute the aberration of whole image, so as to improve the accurate of color temperature estimation Degree.

The determination that it is critical only that nearly gray area of this kind of algorithm.The Algorithms of Selecting foundation of classical nearly gray pixels Formula is：

Y<φ1

Y-|Cb|-|Cr|>φ2

Wherein, φ 1, φ 2 are default thresholding.Meet Y<φ 1 and Y- | Cb |-| Cr |>The region of the inequality of φ 2 exists The big truncated rectangular pyramids in the little top in bottom are shown as in YCbCr space, as shown in figure 1, position pixel in the inner is considered as near Grey, for calculating the aberration that colour temperature causes.

In order to verify that nearly gray area delimit the validity of algorithm, need to use the colour standard colour atla (Ai Seli of Ai Seli 24 Color Checker Classic, this is a kind of test standard color card, hereinafter referred to as " standard color card "), as shown in Figure 2.

In standard color card, the Position Number of most next line be 11～16 totally 6 color lumps be by the bright grey to dark change Color lump, above 3 row Position Numbers be that 21～26,31～36,41～46 to amount to 18 color lumps be the colours such as red, green, blue, Huang, palm fibre Color lump.Standard color card is used as shooting, the photographic subjects of camera installation, to test matter of the gained image in terms of color rendition Amount.

Verification method to the delimitation algorithm effect of nearly gray area：Standard color card is placed in the lucifuge lamp box of specialty, Under 4 kinds of colour temperatures (A, U35, D50, D75), irradiations of the light source of various illumination, its photo is shot.Chosen with nearly gray pixels and calculated Method is processed these photos, and the nearly gray pixels elected keep constant, will be in addition to nearly gray pixels with additional algorithm Pixel be changed to specific color (such as blue, in order to intuitively distinguish), be output as new image.Then new figure is observed Picture, sees in the nearly gray pixels wherein chosen whether include nearly gray pixels (reference colour as much as possible in raw image In card positioned at 11～16 positions color lump in pixel), whether " falsely dropped " non-near gray pixels (standard color card as few as possible In be located at 21～46 positions color lump in pixel).

The corresponding Y of Fig. 1 are used under above-mentioned Validation Mode<φ 1 and Y- | Cb |-| Cr |>2 two inequality of φ carry out nearly grey The selection of point, sign, it is found that there are the following problems：When the value of φ 1, φ 2 is chosen to be compromise value (95,245 or so), in low color Under warm (A light sources), especially low-light (level) (240lx or so) when, from each color lump of standard color card image, be only capable of selecting brightness most Partial pixel in high gray patches (position is 11 color lump), the gray patches of other relatively low GTGs are left out；With This simultaneously, higher color temperature (U35, D50, D75 light source) although when can select 3 higher gray scales color lump (position be 11,12,13 Color lump), but falsely dropped 2～3 colored color lumps (position is 21～46 color lump).On this basis, increasing the values of φ 2 can enter One step reduces the quantity of the gray pixels selected under low colour temperature, reduces φ 2 and is worth, and can further increase and be falsely dropped under higher color temperature Colour element quantity (threshold phi 1 is only used for excluding the excessive point of Y value, judges colour temperature to affect little).Because follow-up colour temperature is estimated The effect of calculating method will be based on the coverage of selected nearly gray pixels and (most choosing should be selected, to ensure to obtain more complete color Warm information), purity (should not select, to reduce interference of the non-near gray pixels to colour temperature estimation procedure), so, it is whether few Choosing is still falsely dropped, and is all unfavorable for aberration, estimated color temperature is accurately calculated.

The content of the invention

For above-mentioned deficiency, the invention provides a kind of nearly gray pixels of image for being applied to AWB algorithm are chosen Algorithm, it can accurately calculate aberration and estimated color temperature, and classical selection image is near in effectively solving AWB algorithm Gray pixels algorithm does not adapt to wide reference color temperature, the problem of various illumination environments, still further provides a kind of AWB Algorithm, it can adapt to wide reference color temperature and various illumination environments, and reach the mesh of white balance by automatic channel Gain tuning 's.

The present invention solves its technical problem and adopts the technical scheme that：A kind of image for being applied to AWB algorithm is near Gray pixels Algorithms of Selecting, is characterized in that, comprise the following steps：

S1：The pixel in target image is read, the YCbCr values of each pixel are obtained；

S2：The YCbCr values for obtaining image pixel are substituted into into following inequality：

Cr>φ1 (1)

Cr<φ2 (2)

Cb>φ3 (3)

Cb<φ4 (4)

|Cr|-|Cb|<φ5 (5)

|Cb+Cr|<φ6 (6)

Y<φ7 (7)

Y>φ8 (8)

In formula, Y, Cb, Cr are Y, Cb, Cr value of image pixel, and φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7, φ 8 are door Limit parameter, φ 1<φ 2, φ 3<φ 4, φ 8<φ7；

S3：Nearly Grey Point is selected from image：If the YCbCr values of image pixel meet all of in step S2 Formula, then it is assumed that it is nearly gray pixels, and issued subsequent algorithm module and carry out color temperature estimation；If image pixel YCbCr values do not meet in step S2 any one inequality in all inequality, then it is assumed that this pixel is not nearly gray pixels, Any process is no longer done to it.

By the one group of new inequality for proposing, so as to standard color card figure can be selected in various colour temperatures, various illumination Most of pixels as in 3 gray patches of brightness highest (position is 11,12,13 color lump), while effectively prevent The pixel in colored color lump (position is 21～46 color lump) is falsely dropped in a large number, is provided more comprehensively, more accurately for subsequent algorithm Original colour difference information.

Preferably, if the pixel value obtained from target image is rgb value, rgb value is converted to into YCbCr values, its Conversion formula is as follows：

Y=0.299*R+0.587*G+0.114*B

Cb=-0.1687*R-0.3313*G+0.5*B

Cr=0.5*R-0.4187*G-0.0813*B

YCbCr values are converted to by carrying out rgb value, the color character for making pixel is easier to be closed by analyzing each component values The mode of system is extracted.

Preferably, the process that the whether near gray pixels of image pixel are judged in step s3 specifically includes following step Suddenly：

S301：Whether the YCbCr values for judging the image pixel meet inequality (1), inequality (2), inequality (3) and not Equation (4), if φ 1<Cr<φ 2 and φ 3<Cb<φ 4 then enters next step, otherwise exits；

S302：Whether the YCbCr values for judging the image pixel meet inequality (5), enter if | Cr |-| Cb | >=φ 5 Enter next step, otherwise exit；

S303：Whether the YCbCr values for judging the image pixel meet inequality (6), if-φ 6<Cb+Cr<φ 6 then enters Enter next step, otherwise exit；

S304：Whether the YCbCr values for judging the image pixel meet inequality (7) and inequality (8), if φ 8<Y<φ 7 judge that the image pixel is nearly gray pixels, otherwise exit.

The method progressively excluded by the YCbCr values to image pixel determined whether for nearly gray pixels, by Formula (1)～(8) have together decided on the enclosed region in YCbCr space, and the point within this region is judged as nearly ash Color pixel, outside point be judged as non-near gray pixels, it is ensured that 3 grey colors of brightness highest in standard color card can be selected Most of pixels in block, and the pixel falsely dropped in a large number in colored color lump is avoided, realizing should select most choosing, should not select not Choosing, for subsequent algorithm provide more comprehensively, more accurately original colour difference information.

Preferably, the determination process of the threshold parameter is：

1) standard color card is shot under various colour temperatures, illumination, it is each at autobalance module inlet in video camera to intercept one Frame original image；

2) for every two field picture, Y arithmetic mean of instantaneous values, the Cb arithmetic averages of some pixels in the middle part of each color lump is calculated Value, Cr arithmetic mean of instantaneous values, obtain 24 cell means；

3) each mean value is stored in statistical form, the statistics tableau format is referring to subordinate list 1 and subordinate list 2；The He of subordinate list 1 Subordinate list 2 is exemplified with Y, Cb, Cr numerical value and colour temperature, illumination that each color lump of image is intercepted at autobalance module inlet in video camera Relation, correspond respectively to A light sources and D75 light sources, but the corresponding light source of the data of actual storage is not limited to A light in statistical form Source and D75 light sources both light sources,

Subordinate list 1：

Subordinate list 2：

In subordinate list 1 and subordinate list 2, each " colour temperature -- illumination " is combined and arranges YCrCb mean values and corresponding to this for mark 1 The piece image shot under illumination condition, the Position Number of line number, row number corresponding to each color lump in image；

4) travel through line label in statistical form be 1, row be numbered Y under various colour temperatures, illumination of 1,2,3 each color lump, Cr, Cb values；

5) to min (Cr), max (Cr), min (Cb), max (Cb), max (| Cr |-| Cb |), max (| Cb+Cr |), max (Y) counted with min (Y)；

6) according to the discrete case of Y, Cr, Cb value in each color lump of original image, by max (Cr), max (Cb), max (| Cr |- | Cb |), max (| Cb+Cr |) and max (Y) respectively add surplus ω, min (Cr), min (Cb) and min (Y) are subtracted respectively Go surplus ω, by the min (Cr) after adjustment, max (Cr), min (Cb), max (Cb), max (| Cr |-| Cb |), max (| Cb + Cr |), max (Y) and min (Y) respectively as φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and φ 8 initial value, wherein surplus Typically between 2～5, the ω values that each extreme value is adopted need not be identical for the value of ω；

7) the above-mentioned threshold parameter to obtaining is optimized process, determines φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, the and of φ 7 The end value of the threshold parameters of φ 8.

Preferably, the optimization process of the threshold parameter is：

(1) initial value of φ 1 to φ 8 is read in video camera as the value of φ 1 to φ 8 in inequality (1) to (8) The the first frame original image intercepted at autobalance module inlet；

(2) execution step S1 to S3 chooses the nearly gray pixels in present image；

(3) to each nearly gray pixels, its position coordinates in the picture is carried out with the border line coordinates of 24 color lumps Contrast, determines it in which color lump；

(4) the nearly gray pixels sum N of present image is counted, number Nij of nearly gray pixels in each color lump, i, j is counted Number for color lump coordinate in the colour standard colour atlas of Ai Seli 24, i=1,2,3,4, j=1,2,3,4,5,6；

(5) judge Σ (N21～N46)/N whether more than threshold deltaIf yes then enter next step, step is otherwise proceeded to 11, wherein, Σ (N21～N46)=N21+N22+...+N26+N31+N32+...+N46；

(6) Nij is traveled through, finds out a maximum color lump of Nij values, i=2,3,4；

(7) its position in statistical form is determined according to this color lump position in the picture, the colour temperature of its affiliated image, illumination Put, obtain its Y, Cr, Cb value；

(8) this Y, Cr, Cb value is substituted into respectively in inequality (1)～(8), finds out the inequality closest to current φ i values, I=1,2 ..., 8；

(9) change the value of the φ i, inequality is no longer set up, and leave surplus, i=1,2 ..., 8；

(10) value of φ after modification 1 to φ 8 and is proceeded to into step as the value of φ 1 to φ 8 in inequality (1) to (8) (2)；

(11) all original images intercepted at autobalance module inlet in video camera whether are had stepped throughIf not yet Having then enter step (12), if yes then enter step (13),

(12) the next frame original image intercepted at autobalance module inlet in video camera is read, and proceeds to step (2)；

(13) value for determining current φ 1 to φ 8 is φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and the threshold parameters of φ 8 End value.

Threshold parameter is determined and optimization to threshold parameter by above-mentioned, further increases the nearly grey of image The accuracy of pixel Algorithms of Selecting, guarantees that in various colour temperatures, various illumination standard color card image can be selected to a greater degree Most of pixels in 3 gray patches of middle brightness highest, and the pixel falsely dropped in a large number in colored color lump is effectively prevent, For subsequent algorithm provide more comprehensively, more accurately original colour difference information.

Present invention also offers a kind of AWB algorithm using the nearly gray pixels Algorithms of Selecting of image described above, It is characterized in that, comprise the following steps：

(1) the initial channel gain of video camera is set：Blue gain μ and red gain v；

(2) pixel in collection image is read, and nearly grey picture is selected using the nearly gray pixels Algorithms of Selecting of described image Element；

(3) the Cb values and Cr values of selecting nearly gray pixels are added to respectively in Σ Cb and Σ Cr, while recording nearly grey Number N of pixel, Cb '=Σ Cb/N, Cr '=Σ Cr/N,

(4) judge whether | Cb ' | and | Cr ' | is respectively less than σ, video camera is using current if | Cb ' | ＜ σ ＆ | Cr ' | ＜ σ Channel gain, otherwise carries out channel gain and adjusts and repeat execution step (2) to step (4) until | Cb ' | ＜ σ | Cr ' | ＜ σ Till, wherein, σ is default threshold value.

By the AWB algorithm using the nearly gray pixels Algorithms of Selecting of image of the present invention, can be in wide color The purpose of white balance is realized in warm scope and various illumination environments.

Preferably, the channel gain adjustment process is comprised the following steps：

(1) judge whether | Cb ' | is more than | Cr ' |, into step (2) if | Cb ' | ＞ | Cr ' |, otherwise proceed to step (3)；

(2) whether Cb ' is judged more than zero, μ=μ-λ if Cb ' ＞ 0, otherwise μ=μ+λ；

(3) whether Cr ' is judged more than zero, the v=v- λ if Cr ' ＞ 0, otherwise v=v+ λ；

Wherein, λ is regulation step-length.

By automatic channel Gain tuning come adjust automatically white balance, preferably to reach the purpose of white balance.

The invention has the beneficial effects as follows：

(1) one group of new inequality that the nearly gray pixels Algorithms of Selecting of image of the invention passes through proposition, so that various 3 gray patches of brightness highest in standard color card image can be selected when colour temperature, various illumination, and (position is 11,12,13 color Block) in most of pixels, while effectively prevent the pixel falsely dropped in a large number in colored color lump (position is 21～46 color lump), For subsequent algorithm provide more comprehensively, more accurately original colour difference information.

(2) in the nearly gray pixels Algorithms of Selecting of image, YCbCr values are converted to by carrying out rgb value, make the color of pixel Feature is easier to be extracted by way of analyzing each component values relation.

(3) in the nearly gray pixels Algorithms of Selecting of image, the side progressively excluded by the YCbCr values to image pixel Method determines whether for nearly gray pixels, by inequality (1)～(8) enclosed region in YCbCr space, position to have been together decided on Point within this region is judged as nearly gray pixels, outside point be judged as non-near gray pixels, it is ensured that can select Most of pixels in standard color card in 3 gray patches of brightness highest, and avoid and falsely drop in a large number in colored color lump Pixel, realizing should select most choosing, should not select, for subsequent algorithm provide more comprehensively, more accurately original colour difference information.

(4) in the nearly gray pixels Algorithms of Selecting of image, by being determined to threshold parameter and to threshold parameter Optimization, further increases the accuracy of the nearly gray pixels Algorithms of Selecting of image, guarantees to a greater degree in various colour temperatures, various The most of pixels in 3 gray patches of brightness highest in standard color card image can be selected during illumination, and is prevented effectively from Falsely drop the pixel in colored color lump in a large number, for subsequent algorithm provide more comprehensively, more accurately original colour difference information.

(5) AWB algorithm of the invention, can be effective by adopting the new nearly gray pixels Algorithms of Selecting of image Adapt to wide reference color temperature, the environment of various illumination.

Description of the drawings

With reference to Figure of description, the present invention will be described.

Fig. 1 is the schematic diagram of the nearly gray-scale pixels choosing method correspondence image in YCbCr space of prior art；

Fig. 2 is the schematic diagram of the colour standard colour atlas of Ai Seli 24；

Fig. 3 is the method flow diagram of the nearly gray pixels Algorithms of Selecting of image of the present invention；

Fig. 4 (a) to Fig. 4 (d) is that the present invention judges the process of the whether near gray pixels of image pixel in YCbCr space The schematic diagram of middle correspondence image；

Fig. 5 is the method flow diagram of the determination process of threshold parameter of the present invention；

Fig. 6 is the method flow diagram of the optimization process of threshold parameter of the present invention；

Fig. 7 is the method flow diagram of AWB algorithm of the present invention.

Specific embodiment

Clearly to illustrate the technical characterstic of this programme, below by specific embodiment, and with reference to its accompanying drawing, to this It is bright to be described in detail.Following disclosure provides many different embodiments or example is used for realizing the different knots of the present invention Structure.In order to simplify disclosure of the invention, hereinafter the part and setting of specific examples are described.Additionally, the present invention can be with Repeat reference numerals and/or letter in different examples.This repetition is that for purposes of simplicity and clarity, itself is not indicated Relation between various embodiments being discussed and/or being arranged.It should be noted that part illustrated in the accompanying drawings is not necessarily to scale Draw.Present invention omits to the description of known assemblies and treatment technology and process avoiding being unnecessarily limiting the present invention.

As shown in figure 3, the present invention is a kind of nearly gray pixels Algorithms of Selecting of image for being applied to AWB algorithm, it Comprise the following steps：

S1：The pixel in target image is read, the YCbCr values of each pixel are obtained；

S2：The YCbCr values for obtaining image pixel are substituted into into following inequality：

Cr>φ1 (1)

Cr<φ2 (2)

Cb>φ3 (3)

Cb<φ4 (4)

|Cr|-|Cb|<φ5 (5)

|Cb+Cr|<φ6 (6)

Y<φ7 (7)

Y>φ8 (8)

In formula, Y, Cb, Cr be image color pixel, φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7, φ 8 be threshold parameter, φ 1<φ 2, φ 3<φ 4, φ 8<φ7.Used as reference, in typical applications, the grade threshold parameters of φ 1 to φ 8 are usually set to following Representative value：φ 1=-12, φ 2=50, φ 3=-50, φ 4=15, φ 5=4, φ 6=25, φ 7=250, φ 8=110.

S3：Nearly Grey Point is selected from image：If the YCbCr values of image pixel meet all of in step S2 Formula, then it is assumed that it is nearly gray pixels, and issued subsequent algorithm module and carry out color temperature estimation；If image pixel YCbCr values do not meet in step S2 any one inequality in all inequality, then it is assumed that this pixel is not nearly gray pixels, Any process is no longer done to it.

By the one group of new inequality for proposing, with above-mentioned 8 inequality nearly Grey Point is selected from image, need by Pixel in individual reading image, obtains the YCbCr values of each pixel, then calculate this pixel YCbCr values whether meet this 8 Individual inequality, if all meeting, then it is assumed that it is nearly gray pixels, is issued the algorithm mould of follow-up execution color temperature estimation algorithm Block；If not meeting one inequality of any of which, then it is assumed that this pixel is not nearly gray pixels, any place is no longer done to it Reason.So so as to 3 gray patches of brightness highest in standard color card image can be selected in various colour temperatures, various illumination Most of pixels in (position is 11,12,13 color lump), with effectively prevent falsely drop in a large number colored color lump (position be 21～ 46 color lump) in pixel, for subsequent algorithm provide more comprehensively, more accurately original colour difference information.

Preferably, if the pixel value obtained from target image is rgb value, rgb value is converted to into YCbCr values, its Conversion formula is as follows：

Y=0.299*R+0.587*G+0.114*B

Cb=-0.1687*R-0.3313*G+0.5*B

Cr=0.5*R-0.4187*G-0.0813*B

YCbCr values are converted to by carrying out rgb value, the color character for making pixel is easier to be closed by analyzing each component values The mode of system is extracted.

Preferably, the process that the whether near gray pixels of image pixel are judged in step s3 specifically includes following step Suddenly：

S301：As shown in Fig. 4 (a), inequality (1), (2), (3), (4) depict a φ 1 in CbCr planes<Cr< φ2、φ3<Cb<The rectangle of φ 4.In actual operation, for current pixel Cr, Cb value calculate inequality (1), (2), (3), (4) whether all set up, meet if all setting up, next step computing is performed to this pixel；Otherwise do not meet, skip this pixel, Read next pixel and start to judge from inequality (1).

S302：As shown in Fig. 4 (b), inequality (5) dug up on rectangle from Fig. 4 (a) left and right two parts, i.e. | Cr |-| Cb | the part of >=φ 5 so as to become the nonagon (part of hatching is stamped in Fig. 4 (b)) for caving in a both sides.Actual fortune In calculation, for Cr, Cb value of current pixel following inequality Cr-Cb is calculated<φ5(Cb>0,Cr>0)、-Cr-Cb<φ5(Cb>0, Cr<0)、Cr+Cb<φ5(Cb<0,Cr>0) with-Cr+Cb<φ5(Cb<0,Cr<0) whether set up simultaneously, if while setting up Meet, next step computing is performed to this pixel；Otherwise do not meet, skip this pixel, read next pixel and from inequality (1) Start to judge.

S303：As shown in Fig. 4 (c), what inequality (6) was delimited be that-the Cr of straight line Cb=φ 6, Cb=- are located in CbCr planes Region between the-Cr of φ 6, takes its common factor with above-mentioned nonagon, cuts away its two angles, and a phase is obtained in CbCr planes To narrower hendecagon (part see hatching is stamped in Fig. 4 (c)).In actual operation, for Cr, Cb value of current pixel Calculate inequality Cb+Cr<φ 6 and Cb+Cr>Whether-φ 6 sets up simultaneously, meets if while setting up, under performing to this pixel One step computing；Otherwise do not meet, skip this pixel, read next pixel and start to judge from inequality (1).

S304：As shown in Fig. 4 (d), inequality (7) and (8) have been separated out a Y value between φ 7, φ 8 in YCbCr space Between Layered-space, with reference to inequality (1) to (6), the final nearly Grey Point region delimited is with Fig. 4 (c) in YCbCr space Shown hendecagon is bottom surface, the distance of the plane of bottom surface to Y=0 is φ 8, is highly the 11 prisms of-φ 8 of φ 7.Actual fortune In calculation, for the Y value of current pixel inequality Y is calculated<φ 7 and Y>Whether φ 8 sets up simultaneously, meets if while setting up, Judge that this pixel is nearly gray pixels, be sent to follow-up algoritic module；Otherwise do not meet, skip this pixel, read next Individual pixel simultaneously starts to judge from inequality (1).

Repeat the above steps S301 select all nearly grey pictures to step S304 up to the whole pixels judged in image Element, be sent to follow-up algoritic module.The Algorithms of Selecting of nearly gray pixels, i.e., together decided on and be located at by inequality (1)～(8) Enclosed region in YCbCr space, the point within this region is judged as nearly gray pixels, outside point be judged as it is non- Nearly gray pixels.The φ 1 to φ 8 used in Fig. 4 (a) to Fig. 4 (d) takes representative value (φ 1=-12, φ 2=50, φ 3=- 50, φ 4=15, φ 5=4, φ 6=25, φ 7=250, φ 8=110).

The method progressively excluded by the YCbCr values to image pixel determined whether for nearly gray pixels, by Formula (1)～(8) have together decided on the enclosed region in YCbCr space, and the point within this region is judged as nearly ash Color pixel, outside point be judged as non-near gray pixels, it is ensured that 3 grey colors of brightness highest in standard color card can be selected Most of pixels in block, and the pixel falsely dropped in a large number in colored color lump is avoided, realizing should select most choosing, should not select not Choosing, for subsequent algorithm provide more comprehensively, more accurately original colour difference information.

In order to determine the value of parameter phi 1 to φ 8, need to test in video camera before AWB module " image is adopted This 8 parameters are calculated, adjusted by the characteristic of collection-processing system " based on this.Preferably, the threshold parameter is really Determining process is：

1) standard color card is shot under various colour temperatures, illumination, it is each at autobalance module inlet in video camera to intercept one Frame original image；

2) for every two field picture, Y arithmetic mean of instantaneous values, the Cb arithmetic averages of some pixels in the middle part of each color lump is calculated Value, Cr arithmetic mean of instantaneous values, obtain 24 cell means；

3) each mean value is stored in statistical form, the statistics tableau format is referring to subordinate list 1 and subordinate list 2, the He of subordinate list 1 Subordinate list 2 is exemplified with Y, Cb, Cr numerical value and colour temperature, illumination that each color lump of image is intercepted at autobalance module inlet in video camera Relation, correspond respectively to A light sources and D75 light sources, but in statistical form the corresponding light source of the data of actual storage be not limited to it is attached A light sources shown in table 1 and subordinate list 2 and D75 light sources both light sources,

Subordinate list 1：

Subordinate list 2：

In subordinate list 1 and subordinate list 2, each " colour temperature -- illumination " is combined and arranges YCrCb mean values and corresponding to this for mark 1 The piece image shot under illumination condition, the Position Number of line number, row number corresponding to each color lump in image；

4) travel through line label in statistical form be 1, row be numbered 1,2,3 each color lump (i.e. nearly grey color of brightness highest 3 Block) Y, Cr, Cb value under various colour temperatures, illumination；

5) to min (Cr), max (Cr), min (Cb), max (Cb), max (| Cr |-| Cb |), max (| Cb+Cr |), max (Y) counted with min (Y)；

6) according to the discrete case of Y, Cr, Cb value in each color lump of original image, by max (Cr), max (Cb), max (| Cr |- | Cb |), max (| Cb+Cr |) and max (Y) respectively add surplus ω, min (Cr), min (Cb) and min (Y) are subtracted respectively Go surplus ω, by the min (Cr) after adjustment, max (Cr), min (Cb), max (Cb), max (| Cr |-| Cb |), max (| Cb + Cr |), max (Y) and min (Y) respectively as φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and φ 8 initial value, wherein, it is remaining Typically between 2～5, the ω values that each extreme value is adopted need not be identical for the value of amount ω；

7) the above-mentioned threshold parameter to obtaining is optimized process, determines φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, the and of φ 7 The end value of the threshold parameters of φ 8.

Preferably, the optimization process of the threshold parameter is：

(1) initial value of φ 1 to φ 8 is read in video camera as the value of φ 1 to φ 8 in inequality (1) to (8) The the first frame original image intercepted at autobalance module inlet；

(2) execution step S1 to S3 chooses the nearly gray pixels in present image；

(3) to each nearly gray pixels, its position coordinates in the picture is carried out with the border line coordinates of 24 color lumps Contrast, determines it in which color lump；

(4) the nearly gray pixels sum N of present image is counted, number Nij (i, j of nearly gray pixels in each color lump is counted Number for color lump coordinate in the colour standard colour atlas of Ai Seli 24, i=1,2,3,4, j=1,2,3,4,5,6)；

(5) judge Σ (N21～N46)/N whether more than threshold deltaIf yes then enter next step, step is otherwise proceeded to 11, wherein, Σ (N21～N46)=N21+N22+...+N26+N31+N32+...+N46, threshold delta default value optional 0.1 is real Adjusted according to system requirements, the demand of subsequent algorithm module in the application of border；

(6) Nij (i=2,3,4) is traveled through, finds out a maximum color lump of Nij values；

(7) its position in statistical form is determined according to this color lump position in the picture, the colour temperature of its affiliated image, illumination Put, obtain its Y, Cr, Cb value；

(8) this Y, Cr, Cb value is substituted into respectively in inequality (1)～(8), find out closest to current φ i (i=1,2 ..., 8) inequality of value；

(9) change the φ i (i=1,2 ..., 8) value, inequality is no longer set up, and leave surplus；

(10) value of φ after modification 1 to φ 8 and is proceeded to into step as the value of φ 1 to φ 8 in inequality (1) to (8) (2)；

(11) all original images intercepted at autobalance module inlet in video camera whether are had stepped throughIf not yet Having then enter step (12), if yes then enter step (13),

(12) the next frame original image intercepted at autobalance module inlet in video camera is read, and proceeds to step (2)；

(13) value for determining current φ 1 to φ 8 is φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and the threshold parameters of φ 8 End value.

Threshold parameter is determined and optimization to threshold parameter by above-mentioned, further increases the nearly grey of image The accuracy of pixel Algorithms of Selecting, guarantees that in various colour temperatures, various illumination standard color card image can be selected to a greater degree Most of pixels in 3 gray patches of middle brightness highest, and the pixel falsely dropped in a large number in colored color lump is effectively prevent, For subsequent algorithm provide more comprehensively, more accurately original colour difference information.

From the angle being embodied as, here provides a kind of using the automatic of the nearly gray pixels Algorithms of Selecting of above-mentioned image White balance algorithm, is characterized in that, comprise the following steps：

(1) the initial channel gain of video camera is set：Blue gain μ and red gain v；

(2) pixel in collection image is read, and nearly grey picture is selected using the nearly gray pixels Algorithms of Selecting of described image Element；

(3) the Cb values and Cr values of selecting nearly gray pixels are added to respectively in Σ Cb and Σ Cr, while recording nearly grey Number N of pixel, Cb '=Σ Cb/N, Cr '=Σ Cr/N,

(4) judge whether | Cb ' | and | Cr ' | is respectively less than σ, video camera is using current if | Cb ' | ＜ σ ＆ | Cr ' | ＜ σ Channel gain, otherwise carries out channel gain and adjusts and repeat execution step (2) to step (4) until | Cb ' | ＜ σ | Cr ' | ＜ σ Till, wherein, σ is default threshold value, for judging the value whether close 0 of Cb ', Cr '.

The AWB algorithm of the nearly gray pixels Algorithms of Selecting of image of the present invention, Neng Gou are adopted by this The purpose of white balance is realized in wide reference color temperature and various illumination environments.

Preferably, the channel gain adjustment process is comprised the following steps：

(1) judge whether | Cb ' | is more than | Cr ' |, into step (2) if | Cb ' | ＞ | Cr ' |, otherwise proceed to step (3)；

(2) whether Cb ' is judged more than zero, μ=μ-λ if Cb ' ＞ 0, otherwise μ=μ+λ；

(3) whether Cr ' is judged more than zero, the v=v- λ if Cr ' ＞ 0, otherwise v=v+ λ；

Wherein, λ, to adjust step-length, can be fixed value, it is also possible to dynamic regulation, be determined by specific algorithm.

By automatic channel Gain tuning come adjust automatically white balance, preferably to reach the purpose of white balance.

The above is the preferred embodiment of the present invention, for those skilled in the art, Without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications are also regarded as this Bright protection domain.

Claims (6)

1. a kind of nearly gray pixels Algorithms of Selecting of image for being applied to AWB algorithm, is characterized in that, comprise the following steps：

S1：The pixel in target image is read, the YCbCr values of each pixel are obtained；

S2：The YCbCr values for obtaining image pixel are substituted into into following inequality：

Cr>φ1 (1)

Cr<φ2 (2)

Cb>φ3 (3)

Cb<φ4 (4)

|Cr|-|Cb|<φ5 (5)

|Cb+Cr|<φ6 (6)

Y<φ7 (7)

Y>φ8 (8)

In formula, Y, Cb, Cr be image color pixel, φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7, φ 8 be threshold parameter, φ 1<φ 2, φ 3<φ 4, φ 8<φ7；

S3：Nearly gray pixels are selected from image：If the YCbCr values of image pixel meet all of inequality in step S2, Then think that it is nearly gray pixels, and issued subsequent algorithm module to carry out color temperature estimation；If the YCbCr values of image pixel Any one inequality in all inequality is not met in step S2, then it is assumed that this pixel is not nearly gray pixels, no longer right It does any process；

The determination process of the threshold parameter is：

1) standard color card is shot under various colour temperatures, illumination, it is each at autobalance module inlet in video camera to intercept frame original Beginning image；

2) for every two field picture, Y arithmetic mean of instantaneous values, Cb arithmetic mean of instantaneous values, the Cr of some pixels in the middle part of each color lump is calculated Arithmetic mean of instantaneous value, obtains 24 cell means；

3) each mean value is stored in statistical form, the statistics tableau format referring to subordinate list 1 and subordinate list 2, subordinate list 1 and subordinate list 2 exemplified with Y, Cb, Cr numerical value and colour temperature, the pass of illumination that each color lump of image is intercepted at autobalance module inlet in video camera System, corresponds respectively to A light sources and D75 light sources, but in statistical form the corresponding light source of the data of actual storage be not limited to A light sources and D75 light sources both light sources,

Subordinate list 1：

Subordinate list 2：

In subordinate list 1 and subordinate list 2, each " colour temperature -- illumination " is combined and arranges YCrCb mean values and corresponding to this illumination for mark 1 Under the conditions of the piece image that shoots, the Position Number of line number, row number corresponding to each color lump in image；

4) travel through line label in statistical form be 1, row be numbered Y, Cr, the Cb of 1,2,3 each color lump under various colour temperatures, illumination Value；

5) to min (Cr), max (Cr), min (Cb), max (Cb), max (| Cr |-| Cb |), max (| Cb+Cr |), max (Y) and Min (Y) is counted；

6) according to the discrete case of Y, Cr, Cb value in each color lump of original image, by max (Cr), max (Cb), max (| Cr |-| Cb |), max (| Cb+Cr |) and max (Y) respectively add surplus ω, min (Cr), min (Cb) and min (Y) are individually subtracted into one Individual surplus ω, by the min (Cr) after adjustment, max (Cr), min (Cb), max (Cb), max (| Cr |-| Cb |), max (| Cb+Cr |), max (Y) and min (Y) respectively as φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and φ 8 initial value, wherein, surplus ω Value typically between 2～5, the ω values that each extreme value is adopted need not be identical；

7) the above-mentioned threshold parameter to obtaining is optimized process, determines φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and φ 8 The end value of limit parameter.

2. the nearly gray pixels Algorithms of Selecting of a kind of image for being applied to AWB algorithm according to claim 1, its It is characterized in that, if the pixel value obtained from target image is rgb value, rgb value is converted to into YCbCr values.

3. the nearly gray pixels Algorithms of Selecting of a kind of image for being applied to AWB algorithm according to claim 1, its It is characterized in that, the process that the whether near gray pixels of image pixel are judged in step s3 specifically includes following steps：

S301：Whether the YCbCr values for judging the image pixel meet inequality (1), inequality (2), inequality (3) and inequality (4), if φ 1<Cr<φ 2 and φ 3<Cb<φ 4 then enters next step, otherwise exits；

S302：Whether the YCbCr values for judging the image pixel meet inequality (5), if | Cr |-| Cb |<φ 5 then enters next Step, otherwise exits；

S303：Whether the YCbCr values for judging the image pixel meet inequality (6), if-φ 6<Cb+Cr<φ 6 is then entered down One step, otherwise exits；

S304：Whether the YCbCr values for judging the image pixel meet inequality (7) and inequality (8), if φ 8<Y<φ 7 is then Judge that the image pixel is nearly gray pixels, otherwise exit.

4. the nearly gray pixels Algorithms of Selecting of a kind of image for being applied to AWB algorithm according to claim 1, its It is characterized in that, the optimization process of the threshold parameter is：

(1) initial value of φ 1 to φ 8 is read in video camera automatically as the value of φ 1 to φ 8 in inequality (1) to (8) The first frame original image that balance module porch intercepts；

(2) execution step S1 to S3 chooses the nearly gray pixels in present image；

(3) to each nearly gray pixels, its position coordinates in the picture is contrasted with the border line coordinates of 24 color lumps, Determine it in which color lump；

(4) the nearly gray pixels sum N of present image is counted, number Nij of nearly gray pixels in each color lump is counted, i, j is love In the beautiful 24 colour standard colour atla of color color lump coordinate numbering, i=1,2,3,4, j=1,2,3,4,5,6；

(5) judge Σ (N21～N46)/N whether more than threshold deltaIf yes then enter next step, step 11 is otherwise proceeded to, its In, Σ (N21～N46)=N21+N22+...+N26+N31+N32+...+N46；

(6) Nij is traveled through, finds out a maximum color lump of Nij values, i=2,3,4；

(7) its position in statistical form is determined according to this color lump position in the picture, the colour temperature of its affiliated image, illumination, Obtain its Y, Cr, Cb value；

(8) this Y, Cr, Cb value is substituted into respectively in inequality (1)～(8), finds out the inequality closest to current φ i values；

(9) change the value of the φ i, inequality is no longer set up, and leave surplus, i=1,2 ..., 8；

(10) value of φ after modification 1 to φ 8 and is proceeded to into step (2) as the value of φ 1 to φ 8 in inequality (1) to (8)；

(11) all original images intercepted at autobalance module inlet in video camera whether are had stepped throughIf without if Into step (12), if yes then enter step (13),

(12) the next frame original image intercepted at autobalance module inlet in video camera is read, and proceeds to step (2)；

(13) value for determining current φ 1 to φ 8 is the final of φ 1, φ 2, φ 3, φ 4, φ 5, φ 6, φ 7 and the threshold parameters of φ 8 Value.

5. the AWB algorithm of the nearly gray pixels Algorithms of Selecting of a kind of employing the claims described image, its feature It is to comprise the following steps：

(1) the initial channel gain of video camera is set：Blue gain μ and red gain v；

(2) pixel in collection image is read, and nearly gray pixels is selected using the nearly gray pixels Algorithms of Selecting of described image；

(3) the Cb values and Cr values of selecting nearly gray pixels are added to respectively in Σ Cb and Σ Cr, while recording nearly gray pixels Number N, Cb '=Σ Cb/N, Cr '=Σ Cr/N,

(4) judge whether | Cb ' | and | Cr ' | is respectively less than σ, video camera adopts current channel if | Cb ' | ＜ σ ＆ | Cr ' | ＜ σ Gain, otherwise carries out channel gain and adjusts and repeat execution step (2) to step (4) until | Cb ' | ＜ σ | Cr ' | ＜ σ, Wherein, σ is default threshold value.

6. the AWB algorithm of the nearly gray pixels Algorithms of Selecting of a kind of employing image according to claim 5, it is special Levying is, the channel gain adjustment process is comprised the following steps：

(1) judge whether | Cb ' | is more than | Cr ' |, into step (2) if | Cb ' | ＞ | Cr ' |, otherwise proceed to step (3)；

(2) whether Cb ' is judged more than zero, μ=μ-λ if Cb ' ＞ 0, otherwise μ=μ+λ；

(3) whether Cr ' is judged more than zero, the v=v- λ if Cr ' ＞ 0, otherwise v=v+ λ；

Wherein, λ is regulation step-length.