CN104217403B - A kind of method that coloured image is converted to gray level image - Google Patents

Abstract

The present invention relates to computer realm, a kind of method that coloured image is converted to gray level image is specifically referred to.The invention discloses a kind of method by image from color conversion to gray scale, realized by two steps, image is carried out into preliminary gray processing with simplest method first, obtain a gray level image；Then the color contrast information of original image is combined with gray level image again, builds an error energy function, try to achieve final colour killing result.The gray level image so tried to achieve can not only well keep global contrast, also can well embody local detail information.

Description

A kind of method that coloured image is converted to gray level image

Technical field

The present invention relates to computer realm, a kind of method that coloured image is converted to gray level image is specifically referred to.

Background technology

Recently as continuing to develop for scientific and technological level, present Digital printing technology can not meet people To original image keep degree requirement.Although occurring in that colour print technology currently on the market, more details can be kept Information, but colour print technology many energy consumptions in need, price costly, temporarily cannot also in our daily life Popularization；The Demand Design many different styles, wherein image ash of many digital camera also according to client at present simultaneously Degree wind transmission lattice but have lost many important detailed information, it is impossible to meet wanting for the artistic study personnel strict to image request Ask.For this problem, if it is possible to propose new image gray processing algorithm, result images is kept as much as possible The contrast information of original image, can make Digital printing technology and image stylization technology more conform to the demand of Vehicles Collected from Market.

The content of the invention

To overcome problem above, it is an object of the invention to provide it is a kind of for keep original image contrast by colour The method that image is converted to gray level image.

A kind of method that coloured image is converted to gray level image, comprises the following steps：

1) preliminary gray processing

Tri- passages of R, G, B of coloured image are extracted first and by its vectorization, become a column vector；Then formula is used (1) preliminary gray processing is carried out.

H (i)=0.299*R (i)+0.587*G (i)+0.114*B (i) (1)

Wherein：H (i) represents pixel i by the gray value after preliminary gray processing, and R (i), G (i), B (i) represents artwork respectively Gray values of the pixel i in R, G, channel B as in.

Weights before three passages are that the sensitivity of color is determined according to human eye.Due in these three colors In, human eye is most sensitive to green, and to blue least sensitivity, therefore G passages account for the proportion of maximum in conversion, and B accounts for minimum Proportion.Through the research of theory and practice, we are using the weighted value shown in formula (1).

2) details enhancing

After preliminary gray processing result is obtained, it would be desirable to which the color contrast of original image is added into just step gray scale In the result of change, therefore construct the error energy function as shown in formula (2)：

Wherein：N represents the total number of pixels of entire image；λ₁, λ₂Two weighted values are represented, our acquiescences general take λ₁= 1, λ₂=1.g_iThe gray value of pixel i, H in the final colour killing image of expression_iMeaning it is consistent with formula (1).K represents each picture The number of the neighbor point existed around plain, such as the pixel on four edges circle of image only has 3 adjacent pixels, in image four 4 pixels at individual drift angle only have 2 adjacent pixels, and other pixels have 4 adjacent pixels, therefore K to different pixels Point has different values.And in formula The contrast between coloured image adjacent pixel is represented, its tool Body formula is as follows.

Due to colour contrast, we directly use three contrast sums of passage, grey-scale contrast and colour contrast Mutually near the scope of grey-scale contrast can be caused to become wide, local contrast information is more protruded.

The α of formula (2)_ijRepresentDirection, its value can only be 1 or -1, be if taking the direction of 1 expression color contrast Positive, taking -1 expression and bear, its judgment criterion is as follows：

If a) R_i>>R_j&G_i>>G_j&B_i>>B_j, then α is taken_ij=1；If conversely, R_i<<R_j&G_i<<G_j&B_i<<B_j, then α is taken_ij=- 1.If this method can not judge, using criterion b).

If b) thering are one or two to be close in two R, G, B of pixels, tri- gray values, another two or an ash Angle value differs larger, such as R_i≈R_j,G_i>>G_j,B_i>>B_j, then α_ij=1, otherwise α_ij=-1.If c) the equal nothing of both above method Method judgement, then take the grey-scale contrast direction of the gray level image that preceding step draws, i.e. H_i-H_jDirection.

Below by taking pixel i as an example, the method for solving of formula (2) is described in detail.Assuming that the size of image is M × N and pixel i With 4 adjacent pixel i+M, i-M, i+1, i-1, then：

By formula (4) to g_iDerivation, obtains

OrderThen

The left side of wherein equation is made up of coefficient with unknown number, and equation the right is a constant.

Sued for peace when by the energy function of all pixels, just can obtain total energy function, such as shown in formula (2), Wo Menye Derivation is carried out to each pixel by the way of shown in formula (5) and (6), and makes its derivative be equal to 0.So, total energy letter The formula obtained after number derivation can regard an equation of AX=B as, and wherein A is a sparse square for M*N × M*N sizes Battle array, its value per a line is just as the coefficient value on equation (6) left side.And X is the unknown matrix number of M × N sizes, i.e. X= [g₁,g₂,Lg_MN]^T, B is the constant matrices of M × N sizes.Finally, we can be drawn final by solution by iterative method equation Colour killing result.

In computer realm, gray level image is the image of each pixel only one of which sample color.This kind of image is usual It is shown as from most dark black to most bright white gray scale, the span of its gray value is [0,255].And coloured image bag It is the dimension for reducing image by the task that coloured image is transformed into gray level image containing tri- channel values of R, G, B, will necessarily so damages The information of a part is lost, thus such as to obtain a good result, it is necessary to keep the contrast of original image as much as possible.

And show currently for human visual perception systematic research, human visual system can not accurately perceive tone Also differed with the change of brightness, and everyone sensitivity to brightness change.The principle for being converted from colour killing simultaneously, has The gray level of limit is impossible to represent each color in color space correspondingly, therefore we can only be as many as possible Ground retains the most sensitive contrast changing unit of visually-perceptible.Again because human eye for image adjacent area color change the most Sensitivity, so it is considered that the information between adjacent pixel plays an important role in greyscale transformation.We have proposed colour The gradient field optimized algorithm of image gray processing.Compared with existing algorithm, the algorithm can not only preferably keep original image Structural information, it is also possible to greatest extent retain original image local contrast information.

The invention discloses a kind of method by image from color conversion to gray scale, realized by two steps, it is first Image is first carried out into preliminary gray processing with simplest method, a gray level image is obtained；Then again by the color of original image Contrast information is combined with gray level image, builds an error energy function, tries to achieve final colour killing result.So try to achieve Gray level image can not only well keep global contrast, also can well embody local detail information.

Brief description of the drawings

Fig. 1 is the algorithm of algorithm of the invention and Grundland and Dodgson^[1]It is right with the algorithm of Gooch et al. [2] Than figure.

Fig. 2 is the algorithm of algorithm of the invention and Rasche et al.^[3]And the algorithm of Smith et al.^[4]Contrast on effect Figure.

Fig. 3 is the algorithm of algorithm of the invention and Lu et al.^[5]And the algorithm of Smith et al.^[4]Effect contrast figure.

Original image in accompanying drawing is coloured image, due to that cromogram can not occur in patent application specification limitation article Picture, therefore, we are directly by the rgb2gray functions in matalab using the result after original image gray processing as original herein Image shows.

Specific embodiment

The present invention is expanded on further with reference to specific embodiment, it should be appreciated that following examples are merely to illustrate the present invention Rather than limit the scope of the invention.

Embodiment

The size of image is M × N, and by taking pixel i as an example, and pixel i has 4 adjacent pixel i+M, i-M, i+1, i-1, When implementing coloured image gray processing, our algorithms are comprised the following steps that：1. a width coloured image first read, and the number of image According to being integer, we will convert thereof into double precision and float

Point-type；The size of image is M × N, and pixel i has 4 adjacent pixels

i+M,i-M,i+1,i-1

2. the double precision coloured image for the first step being obtained carries out preliminary gray processing conversion with formula (1)；

H (i)=0.299*R (i)+0.587*G (i)+0.114*B (i)

(1)

Wherein,

H (i) represents pixel i by the gray value after preliminary gray processing；

R (i), G (i), B (i) represent gray values of the pixel i in R, G, channel B in original image respectively.

3. the color contrast that formula (3) calculates original image is pressed, and use a), b), c) three criterions judge face

The direction of color contrast；

α_ijRepresentDirection, its value be 1 or -1, if the direction for taking 1 expression color contrast is positive, take -1 table Show negative, its judgment criterion is as follows：

If a) R_i>>R_j&G_i>>G_j&B_i>>B_j, then α is taken_ij=1；If conversely,

R_i<<R_j&G_i<<G_j&B_i<<B_j, then α is taken_ij=-1；

If b) criterion a) cannot judge, use：If having one or two in two R, G, B of pixels, tri- gray values It is close, another two or a gray value differ larger, such as R_i≈R_j,G_i>>G_j,B_i>>B_j, then α_ij=1, otherwise α_ij=- 1；

If c) a) and b) two kinds of criterions cannot judge, step 1 is taken) the grey-scale contrast side of gray level image that draws To i.e. H_i-H_jDirection.

4. the corresponding error energy function of image is listed by formula (2) again；

Wherein,

N represents the total number of pixels of entire image；

λ₁, λ₂Represent two weighted values, λ₁=1, λ₂=1；

g_iThe gray value of pixel i in the final colour killing image of expression；

H_iRepresent pixel i by the gray value after preliminary gray processing；

K represents the number of the neighbor point existed around each pixel；

5. formula (4) is pressed again, and error energy function is converted into corresponding AX=B equations by (5), (6)；

By formula (4) to g_iDerivation, obtains

OrderThen

The left side of wherein equation is made up of coefficient with unknown number, and equation the right is a constant.

Sued for peace when by the energy function of all pixels, just can obtain total energy function, such as shown in formula (2), Wo Menye Derivation is carried out to each pixel by the way of shown in formula (5) and (6), and makes its derivative be equal to 0.So, total energy letter The formula obtained after number derivation can regard an equation of AX=B as, and wherein A is a sparse square for M*N × M*N sizes Battle array, its value per a line is just as the coefficient value on equation (6) left side.And X is the unknown matrix number of M × N sizes, i.e. X= [g₁,g₂,…g_MN]^T, B is the constant matrices of M × N sizes.

6. final gray value finally is obtained with the mode of iterative, obtain final gray level image.

In order to more directly highlight the superiority of this algorithm, we are by the knots of the result of inventive algorithm and five kinds of algorithms Fruit is compared, and is respectively the algorithm of classical Gooch et al.^[2]With the algorithm of Rasche et al.^[3], pass throughExamination Test the algorithm of the two kinds of better performances ground algorithm-Grunalana and Doagson for drawing^[1]With the algorithm of Smith et al.^[4], with And the algorithm of the preferable Lu for being done in coloured image gray processing field in recent years et al..

Show the algorithm of the algorithm of oneself and Grunalana and Doagson such as accompanying drawing 1^[1]With the calculation of Gooch et al. The comparison diagram of method.It will be seen that the algorithm of Gooch et al. from figure^[2]Result can keep the local contrast of image to believe Breath, but global contrast information can be lost, and it can make whole image seem that comparing is obscured.And Grunalana and Doagson Algorithm^[1]Although in global information holding, the algorithm than Gooch et al.^[2]It is substantially good a lot, but can have lost a lot Local contrast information, the contrast information of petal such as in the details and the second width image of the clothes in piece image.And Our algorithm, will be well upper many than both algorithms in contrast holding, in piece image, our result energy Enough it is clear that the texture information on clothes and on floor；In the second width image, our result can be well Keep the grain shape of the texture and soil on petal.

Fig. 2 shows the algorithm of this algorithm and Rasche et al.^[3]And the algorithm of Smith et al.^[4]Contrast on effect Figure.We can become apparent from coming from figure, Rasche et al.^[3]Although the result global structure that can keep substantially, But many local contrast informations can be lost.And the algorithm of Smith et al.^[4]Although having been able to preferably keep local right Than degree information, but still a part of obvious contrast information can be lost.Such as the lower left part of piece image, its phase The contrast in neighbouring region is just less obvious, not clear enough although the face color block of personage can be recognized.From figure we It can be found that our algorithm can preferably keep the contrast information of global and local.

The algorithm of Lu et al.^[5]It is the outstanding algorithm of the comparing done in image gray processing field in recent years, the algorithm is not only examined Consider global structural information and also maintain the local contrast information of a part, and relax the pact to color contrast direction Beam, suitable symbol is automatically selected with bimodal distribution function.But this method can still lose the important local contrast in part Degree information, and requirement of the different images to parameter is also different, and user is difficult to debug out optimal result.As shown in figure 3, second Row is the result of this chapter algorithms, the 3rd, fourth line be Lu et al. algorithm^[5]As a result, the result of fourth line is using Lu etc. People^[5]In the default value that article is proposed, and the result of the third line be it is adapted after the optimum that obtains.We can be with from figure Find out, if we are directly using Lu et al.^[5]Propose that default value carrys out gray processing coloured image in article, can become result images It is smudgy, and distribution to gray level has defect；By our multiple adjusting parameters, it is found that its parameter value is When 0.9, its image is optimal, and the result is come clearly a lot than default value, but still have lost many contrast informations.And I Result can not only well keep global contrast, also can well keep local contrast information.It is right from figure Than in figure it can be found that the algorithm of our algorithms in details holding than Lu et al.^[5]Go out many well, such as it is old in first figure The information such as the text color in wrinkle on face and on hand and second image on child's clothes.

In addition, the total algorithm framework and document of this algorithm：The method that coloured image based on gradient field turns gray level image (Zhang Weixiang, Zhou Bingfeng)^[7]It is different, their algorithm is spatially to calculate Grad in LCrCb, they are according to bright Spend the size of Grad and color gradient value to change brightness value, and build Poisson's equation.And algorithm of the invention is first Carry out preliminary gray processing and obtain preliminary gray level image, Grad, last preliminary gray level image are then directly calculated on rgb space It is combined with Grad, constructs energy function.Although our algorithms are similar with the name of their algorithms, content is completely not Equally.

Bibliography：

[1] the gloomy Neils Anthonies colour killings of gram Lu Lan mark and Dodge：Quickly, the enhanced colour of contrast arrives gray scale Conversion [J] pattern-recognitions, 2007,40 (11):2891–2896.M.Grundland and N.A.Dodgson.Decolorize:Fast,contrast enhancing,color to grayscale conversion [J].Pattern Recognition,2007,40(11):2891–2896.

[2] ancient strange Amies, Ao Er are gloomy polite, the rich woods Jacks of child, and ancient strange Bruces is colored to become gray scale：Conspicuousness is protected Colour fading algorithm [J] the Associations for Computing Machinery computer graphics proceedings held, 2005,24 (3):634–639.

Amy A.Gooch,Sven C.Olsen,Jack Tumblin,Bruce Gooch.Color2gray: salience-preserving color removal[J].ACM Transactions on Graphics,2005,24(3): 634–639.

[3] the thorough karrs in Lars, cover this special Robert, and dimension Manfred Stohl James regards person's for monochromasia and two primary colors Details keeps color image reproduction algorithm [J] IEEE-USA's computer graphics and application periodical, 2005,25(3):22–30.

K.Rasche,R.Geist,J.Westall.Detail preserving reproduction of color images for monochromats and dichromats[J].IEEE Computer Graphics and Applications,2005,25(3):22–30.

[4] Smith Kai Leju, Lan Desi Pi Aier, Te Longte Ju Erle, is notable for Mace Ke Weisiji Karols Gray processing：One easily and rapidly perceives sharp picture video conversion method computer graphics forum, 2008,27 (2): 193–200.

K.Smith,P.Landes,J.Thollot,K.Myszkowski.Apparent greyscale:A simple and fast conversion to perceptually accurate images and video[J].Computer Graphics Forum,2008,27(2):193–200.

[5] colour killing algorithm [C] U.S. electrics and Electronics Engineer association that Lu Cewu, Xu Li, Jia Jiaya contrasts keep Photography international conference collection of thesis, 2012,1-7.Cewu Lu, Li Xu and Jiaya Jia.Contrast can be calculated Preserving Decolorization[C].Proceedings of IEEE International Conference on Computational Photography,2012,1–7.

[6] colored perception evaluation [J] the computer graphics forums to greyscale image transitions of Ka Dike Martin, 2008, 27(7):1745–1754.

M.Perceptual Evaluation of Color-to-Grayscale Image Conversions [J].Computer Graphics Forum,2008,27(7):1745-1754. [7] Zhang Weixiang, Zhou Bingfeng are based on gradient field Coloured image turn method [J] camera works of gray level image, 2007,7:20–22.

Claims (1)

1. a kind of method that coloured image is converted to gray level image, it is characterised in that：Comprise the following steps：

1) preliminary gray processing

Tri- passages of R, G, B of coloured image are extracted first and by its vectorization, become a column vector；Then formula (1) is used Carry out preliminary gray processing：

H (i)=0.299*R (i)+0.587*G (i)+0.114*B (i) (1)

Wherein,

H (i) represents pixel i by the gray value after preliminary gray processing；

R (i), G (i), B (i) represent gray values of the pixel i in R, G, channel B in original image respectively；

2) details enhancing

After preliminary gray processing result is obtained, the color contrast of original image is added in the result of preliminary gray processing, structure Build the error energy function shown in formula (2)：

$E=\underset{i=1}{\overset{N}{\&Sigma;}}\&lsqb;{\mathrm{\&lambda;}}_1{(g_i-H_i)}^2+{\mathrm{\&lambda;}}_2\underset{j=1}{\overset{K}{\&Sigma;}}{(\&dtri;g_{ij}-{\mathrm{\&alpha;}}_{ij}\&dtri;C_{ij})}^2\&rsqb;---\left(2\right)$

Wherein,

N represents the total number of pixels of entire image；

λ₁, λ₂Represent two weighted values, λ₁=1, λ₂=1；

g_iThe gray value of pixel i in the final colour killing image of expression；

H_iRepresent pixel i by the gray value after preliminary gray processing；

K represents the number of the neighbor point existed around each pixel；

$\&dtri;g_{ij}=g_i-g_j;$

The contrast between coloured image adjacent pixel is represented,Calculate such as formula (3)：

$\&dtri;C_{ij}=\left|\right|\&dtri;R_{ij}\left|\right|+\left|\right|\&dtri;G_{ij}\left|\right|+\left|\right|\&dtri;B_{ij}\left|\right|=\left|\right|R_i-R_j\left|\right|+\left|\right|G_i-G_j\left|\right|+\left|\right|B_i-B_j\left|\right|---\left(3\right);$

α_ijRepresentDirection, its value is 1 or -1, if the direction for taking 1 expression color contrast is positive, it is negative to take -1 expression, Its judgment criterion is as follows：

If a) R_i＞ ＞ R_j&G_i＞ ＞ G_j&B_i＞ ＞ B_j, then α is taken_ij=1；If conversely, R_i＜ ＜ R_j&G_i＜ ＜ G_j&B_i＜ ＜ B_j, then Take α_ij=-1；

If b) criterion a) is unsatisfactory for or cannot judge, use：If have in two R, G, B of pixels, tri- gray values one or Two are close, and another two or a gray value difference are larger, if R_i≈R_j,G_i＞ ＞ G_j,B_i＞ ＞ B_j, then α_ij=1；Instead It, if R_i≈R_j,G_i＜ ＜ G_j,B_i＜ ＜ B_j, then α_ij=-1；

If c) a) and b) two kinds of criterions cannot judge, α_ijTake step 1) the grey-scale contrast direction of gray level image that obtains, Even H_i-H_jBe positive number, then α_ij=1；If conversely, H_i-H_jBe negative, then α_ij=-1；

By formula (2) to the error energy function pair g obtained by each pixel_iDerivation, and makeAnd by all pixels Energy function is sued for peace, and just can obtain total energy function, then, the formula obtained after total energy function derivation can be regarded as One equation of AX=B；

Finally, by solution by iterative method equation, final result is drawn.