CN103793897B - A kind of digital picture halftoning method based on wavelet multi-scale information fusion - Google Patents

Abstract

The present invention relates to a kind of digital picture halftoning method based on wavelet multi-scale information fusion, belong to digital picture prepress treatment technical field.Continuous toned image is converted into gray level image and judges whether gray level image is the 2 of standard by the present inventionⁿ×2ⁿGray level image；Employing two-dimensional discrete wavelet conversion obtains four dimensional information of the wavelet field of normal grayscale image, and between recycling yardstick, the wavelet details Relationship of Coefficients of the autocorrelation fusion interlayer of wavelet coefficient sets up marginal error measure function；K-means clustering procedure is used to be divided into respectively by normal grayscale imagekIndividual region, recycles each Local Deviation reciprocal as weight, sets up domain error measure function；The additivity of marginal error measure function and domain error measure function is set up combined error measure function, then uses direct two-value searching method to minimize normal grayscale image and the error of coarse half tone image.The present invention solves existing method cannot take into account image smoothing and sharpness problems.

Description

A kind of digital picture halftoning method based on wavelet multi-scale information fusion

Technical field

The present invention relates to a kind of digital picture halftoning method based on wavelet multi-scale information fusion, before belonging to digital picture print Processing technology field.

Background technology

The engraving mode of laser gravure plate-making is a kind of scan-type engraving method imitating print out equipment.Laser engraved gravure defeated Going out mode is the on off state by controlling laser, makes high energy laser beam be radiated on the specified point of surface of the work, makes this point rapid Produce evaporation, form out a depression points at surface of the work, thus show two kinds of tones of black and white.So for multi-grey image, Must convert it to be suitable for the black and white binary image of laser output by digital halftone technology before output.Removing optical maser wavelength, Outside the factors such as pulse frequency, the beam quality of laser and rapidoprint characteristic, the halftoning algorithm of multi-grey image is to laser Image output effect plays particularly critical effect.

Digital picture halftoning is to be set in the such as two-value such as laser platemaker, digital printer, laser printer by continuous toned image Standby upper development also produces the key technology of continuous toned image illusion in human visual system.It obtains in people's production, life Increasingly it is widely applied.At present from family, the small desk ink-jet of office, laser printer, laser platemaker to greatly The publication and printing system of type, digital halftone technology can be described as ubiquitous.It is hidden that digital halftone technology applies also for data Hiding and digital watermark technology, 3-dimensional digital halftone technique is also used widely in Layered Manufacturing Technology.Existing halftoning Method carries out image modeling from single yardstick, although obtain a certain degree of successful Application under static environment, but in dynamic environment Lower half tone image quality is the most unsatisfactory.Reason is owing under dynamic environment, the information of fine dimension is the most unreliable so that In single yardstick, space, frequency adaptive error are estimated information and are had stronger dynamic random.And due to human visual system's Attention mechanism, makes people can observe image under different resolution, and then picture quality is produced evaluation, but is subject to To the impact of the such as extraneous factor such as illumination variation, shade, background clutter, it is in the image information of fine dimension in time and space On there is stronger dynamic random.Multiresolution O&A mechanism under the most this dynamic environment, it is desirable in digital picture Halftoning process considers the multi-scale information of error metric, relates to how to merge and comprise thick scale error and estimate many Scale error estimates information to realize this technical barrier of digital halftone.

Summary of the invention

The invention provides a kind of digital picture halftoning method based on wavelet multi-scale information fusion, for solving dynamic Under state environment multiresolution O&A mechanism, merge and comprise the multiple dimensioned error metric information that thick scale error is estimated Realize this technical problem of digital halftone.

The technical scheme is that a kind of digital picture halftoning method based on wavelet multi-scale information fusion, described side Specifically comprising the following steps that of method

A, continuous toned image is converted into gray level image and judges whether gray level image is the 2 of standardⁿ×2ⁿGray level image；

B, employing two-dimensional discrete wavelet conversion obtain four dimensional information of the wavelet field of normal grayscale image, little between recycling yardstick The autocorrelation of wave system number merges the wavelet details Relationship of Coefficients of interlayer sets up marginal error measure function；

Normal grayscale image is divided into k region by C, employing K-means clustering procedure respectively, recycles each Local Deviation reciprocal As weight, set up domain error measure function；

D, the additivity of marginal error measure function and domain error measure function is set up combined error measure function, then use Directly two-value searching method minimizes normal grayscale image and the error of coarse half tone image, obtains the digital halftone figure of optimum Picture.

In described step B, what marginal error measure function was set up specifically comprises the following steps that

B1, use and there is the Harr small echo of orthogonality carry out two-dimensional discrete wavelet conversion, it is achieved four multi-scale wavelet of gray level image become Change and obtain four scale wavelet transform images；

B2, obtain level according to four scale wavelet transform images, be vertically distributed with diagonal angle subband wavelet coefficient；

It is B3, independent it is assumed that to level, be vertically normalized with three, diagonal angle direction wavelet coefficient according to wavelet sub-band, Obtain the distribution of the wavelet coefficient after normalized；

B4, the autocorrelation of wavelet coefficient between yardstick is utilized to merge each multi-scale wavelet coefficient after normalization, little after being merged Wave system number；

B5, will merge after wavelet coefficient as weight, set up marginal error measure functionWherein, l represents wavelet transformation progression；LH, HL, HH table respectively Show multi-scale wavelet territory vertical direction, horizontal direction and diagonally opposed wavelet coefficient fuse information；I, j represent image pixel；B_i,j Represent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent model printer error.

In described step C, what domain error measure function was set up specifically comprises the following steps that

C1, use K-means clustering procedure that normal grayscale image is divided into k area image, wherein k=1,2,3,4；

C2, the average calculating each region and variance；Using the inverse of variance as weight, weighted least squares is utilized to set up district Territory error metric functionWherein, k=1,2,3,4 represent cut zone；λ_kRepresent Weight coefficient；I, j represent image pixel；B_i,jRepresent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent that model printer is by mistake Difference.

In described step D, direct two-value searching algorithm minimizes specifically comprising the following steps that of combined error measure function

D1, the additivity of marginal error measure function and domain error measure function is set up combined error measure functionWherein, λ_kRepresent weight coefficient；K=1,2,3,4 Represent cut zone；z_i,jRepresent the collimation error；w_i,jRepresent model printer error；

D2, the eight neighborhood region of each pixel in normal grayscale image and coarse half tone image is utilized direct two-value search calculate Convert in method and jump exchange, makes combined error measure function minimum；

D3, when entire image combined error measure function minimum, algorithm terminate, obtain optimum half tone image.

Described coarse half tone image uses error-diffusion method process to obtain.

The operation principle of the present invention is:

Combined error measure function formulation process:

One, set up marginal error measure function:

If image wavelet coefficient on j yardstick after level Four wavelet decomposition is C_j,k, scale correlations coefficient is W_j,k=C_j,k×C_j+1,k。 Between the brotgher of node of wavelet coefficient interband and same father node, in band, wavelet coefficient all has stronger dependency.Utilize amplitude Wavelet coefficient combines dependency in interband band, defines relative coefficient between new wavelet coefficient.P_j,kRepresent that under j yardstick, kth is little Wave system number all child nodes wavelet coefficient maximum:λ_j,k, λ '_j,kRepresent that small echo two enters segmentation, P_j,kContain (j, k) all child node information during position, wavelet coefficient decays along with yardstick, and different yardsticks is multiplied by difference Scale factor 2^-js, (s value is determined by experiment), reach the mesh ground at its edge prominent.Define new maximum child node to be correlated with Coefficient: R_j,k=P_j,k×C_j,k.Coefficient R_j,kIt is equal to the product of all wavelet coefficients.Amendment R_j,kFor taking advantage of of adjacent yardstick Long-pending, define the correlation coefficient of new correlation coefficient-adjacent maximum child nodeAfter determining correlation coefficient, for making correlation coefficient and small echo Coefficient has comparability, is normalized correlation coefficient, obtains new wavelet coefficient d_j,k.At wavelet field, wavelet coefficient It is distributed on different yardsticks and subband, independent it is assumed that i.e. wavelet coefficient statistical iteration on yardstick and subband according to wavelet sub-band, Utilize the method that between yardstick, wavelet coefficient merges to obtain fine dimension Wavelet Fusion coefficient, consider the approximate information of image simultaneously. Wavelet coefficient after fusion can set up boundary error measure function as weight coefficient.

$E_l=\underset{l=1}{\overset{4}{\Σ}}(LH,HL,HH)\underset{(i,j)\∈B_{i,j}}{\Σ}{(z_{i,j}-w_{i,j})}^2$

Two, set up domain error measure function:

In uniform cartesian grid Ω, u is defined as a secondary continuous toned image, in normal range scale, it is assumed that Ω=[1:n] × [1:m].In Ω, any image v application linear factor K:v → K [v] is represented human visual system (HVS) The perception factor, grey level range is [0,1] (regardless of whether continuous toned image or half tone image), and K is assumed to be it is weak low pass, Therefore K [1] ≡ 1.

Can well the quality of predictive-coded picture by the HVS model of Mannos and Sakrison:

H_r(f_r)=2.6 (0.0192+0.114f_r)exp{-(0.114f_r)^1.1}

FrequencyIt is horizontal frequency f_xWith vertical frequency f_yVirtual value, unit is cycles/deg.

Assume that pixel is square, T_x=T_y=T the length of a cell (T be).Adjacent stain may overlapping and meeting The white point that covering part is adjacent.The radius of point is not less thanOnly in this way, the complete blackening of print image zone can just be made. This means the region always having some overlaps between stain and adjacent white point, cause the gray scale etc. creating pixel on white point Level so that distortion occurs in printing image.A kind of simple model printer is referred to as " circular shaped lattice point double exposure " model and is evaluating output This deformation is considered during the tonal gradation of each pixel of image.In cell, (i, central point j) is (x to pixel_i,y_j), wherein x_I=iT_x+T_x/ 2, y_J=jT_y+T_y/ 2, the bianry image of output is [b_i,j], work as b_i,jPixel center (x is represented when=1_i,y_j) it is stain, b_i,j=0 Time represent pixel center (x_i,y_j) it is white point.

$P_{i,j}=P\left(W_{i,j}\right)=\left\{\begin{array}{cc}1& if,b_{i,j}=1\\ f_1\mathrm{\α}+f_2\mathrm{\β}-f_3\mathrm{\γ}& if,b_{i,j}=0\end{array}\right.$

Window function W in formula_i,jIncluding b_i,jWith its eight neighborhood systems, f₁Represent both horizontally and vertically black site, f₂Represent diagonal angle side To black site and non-conterminous black site, f₃Representing one to be horizontally oriented, one is vertically oriented the most adjacent black Site.The ratio of dash area and grate area during α, β and γ are figure in above formula, if actual print point radius and ideally The ratio of radius is that ρ represents, for preferable site, parameter ρ=1, but output gradation of image grade can occur " dot gains " existing As, therefore in view of this deformation, ρ=1.25, α=0.33, β=0.03, γ=0.09.

When optimizing, the weight of unknown parameter is inversely proportional to the variance of each predictor variable value:

Selection that the visual quality of halftoning depends on initial point and the optimisation strategy taked.A kind of simple iterative strategy is as follows, For any image point, (i j), gives an initial estimation [b_i,j], find binarized pixel b_i,jSo that weighted quadratic value is minimum.

$E_{\mathrm{\λ}}=\underset{k=1}{\overset{4}{\Σ}}{\mathrm{\λ}}_k\underset{(i,j)\∈B_{i,j}}{\Σ}{(z_{i,j}-w_{i,j})}^2$

In formula, B_i,jIt is that (i, eight neighborhood j), if vision wave filter and neighborhood system select sufficiently large, minimize E to point_i,jIt is equal to Minimize global error ε.Attempt to produce optimal halftoning duplicating image, minimize two-value half tone image and original gray level Square error between image.Given one secondary gray level image [x_i,j], i=1 ..., N_W, j=1 ..., N_H。N_WFor piece image x direction On the number of pixel, N_HFor the number of the pixel on piece image y direction, x_i,jRepresent that pixel is positioned at grid i row and j row, Assume a secondary grayscale image [x_i,j], the tonal gradation of each pixel from 0 (white) to 1 (black), it is assumed that a site can To produce a pixel, therefore, grayscale image [x_i,j] and bianry image [b_i,j] there is identical yardstick.First one secondary initial half is obtained Tone images [b_i,j], for any image point, (i j) can find binarized pixel b_i,jMinimize the difference of two squares.

$\mathrm{\ϵ}=\underset{i=1}{\overset{N_w}{\Σ}}\underset{j=1}{\overset{N_H}{\Σ}}{(z_{i,j}-w_{i,j})}^2$

z_i,j=n (x_i,j)*h′_i,j

w_i,j=n (p_i,j)*h_i,j

p_i,j=P (W_i,j)

w_i,jIt is by b_i,jShown neighborhood system, * represents convolution.h_i,jWith h '_i,jRepresent half tone image and the vision of continuous toned image The impulse response that wave filter is different.Boundary condition assumes that does not has ink to be in outside image border.

The invention has the beneficial effects as follows:

1, solve existing method and cannot take into account image smoothing and sharpness problems, half tone image can be realized there is good limit Edge definition and the flatness in region, obtain high-quality half tone image.

2, by application weighted least squares, minimize combined error function, reach local optimum.

3, the digital halftoning method that the multi-scale information of the present invention merges is utilized, the laser plate-making image of available high-quality.

Accompanying drawing explanation

Fig. 1 is that the multi-scale information of the present invention merges halftone technique flow chart；

Fig. 2 is the image hierarchical relationship figure through level Four wavelet decomposition of the present invention；

Fig. 3 be the present invention yardstick between the quad-tree structure figure of wavelet coefficient；

Fig. 4 is the level Four wavelet transformation figure of the image of the present invention；

Fig. 5 is present invention cluster segmentation image as K=2；

Fig. 6 is present invention cluster segmentation image as K=3；

Fig. 7 is present invention cluster segmentation image as K=4；

Fig. 8 is quality evaluation of halftone image parameter mean square error (MSEv) comparison diagram of the present invention；

Fig. 9 is quality evaluation of halftone image parameter peak signal to noise ratio (PSNR) comparison diagram of the present invention.

Detailed description of the invention

Embodiment 1: as shown in figs 1-9, a kind of digital picture halftoning method based on wavelet multi-scale information fusion, described Specifically comprising the following steps that of method

A, continuous toned image is converted into gray level image and judges whether gray level image is the 2 of standardⁿ×2ⁿGray level image；

B, employing two-dimensional discrete wavelet conversion obtain four dimensional information of the wavelet field of normal grayscale image, little between recycling yardstick The autocorrelation of wave system number merges the wavelet details Relationship of Coefficients of interlayer sets up marginal error measure function；

Normal grayscale image is divided into k region by C, employing K-means clustering procedure respectively, recycles each Local Deviation reciprocal As weight, set up domain error measure function；

D, the additivity of marginal error measure function and domain error measure function is set up combined error measure function, then use Directly two-value searching method minimizes normal grayscale image and the error of coarse half tone image, obtains the digital halftone figure of optimum Picture.

In described step B, what marginal error measure function was set up specifically comprises the following steps that

B1, use and there is the Harr small echo of orthogonality carry out two-dimensional discrete wavelet conversion, it is achieved four multi-scale wavelet of gray level image become Change and obtain four scale wavelet transform images；

B2, obtain level according to four scale wavelet transform images, be vertically distributed with diagonal angle subband wavelet coefficient；

It is B3, independent it is assumed that to level, be vertically normalized with three, diagonal angle direction wavelet coefficient according to wavelet sub-band, Obtain the distribution of the wavelet coefficient after normalized；

B4, the autocorrelation of wavelet coefficient between yardstick is utilized to merge each multi-scale wavelet coefficient after normalization, little after being merged Wave system number；

B5, will merge after wavelet coefficient as weight, set up marginal error measure functionWherein, l represents wavelet transformation progression；LH, HL, HH table respectively Show multi-scale wavelet territory vertical direction, horizontal direction and diagonally opposed wavelet coefficient fuse information；I, j represent image pixel；B_i,j Represent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent model printer error.

In described step C, what domain error measure function was set up specifically comprises the following steps that

C1, use K-means clustering procedure that normal grayscale image is divided into k area image, wherein k=1,2,3,4；

C2, the average calculating each region and variance；Using the inverse of variance as weight, weighted least squares is utilized to set up district Territory error metric functionWherein, k=1,2,3,4 represent cut zone；λ_kRepresent Weight coefficient；I, j represent image pixel；B_i,jRepresent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent that model printer is by mistake Difference.

In described step D, direct two-value searching algorithm minimizes specifically comprising the following steps that of combined error measure function

D1, the additivity of marginal error measure function and domain error measure function is set up combined error measure functionWherein, λ_kRepresent weight coefficient；K=1,2,3,4 Represent cut zone；z_i,jRepresent the collimation error；w_i,jRepresent model printer error；

D2, the eight neighborhood region of each pixel in normal grayscale image and coarse half tone image is utilized direct two-value search calculate Convert in method and jump exchange, makes combined error measure function minimum；

D3, when entire image combined error measure function minimum, algorithm terminate, obtain optimum half tone image.

Described coarse half tone image uses error-diffusion method process to obtain.

Embodiment 2: as shown in figs 1-9, a kind of digital picture halftoning method based on wavelet multi-scale information fusion, described Specifically comprising the following steps that of method

A, continuous toned image is converted into gray level image and judges whether gray level image is the 2 of standardⁿ×2ⁿGray level image；

B, employing two-dimensional discrete wavelet conversion obtain four dimensional information of the wavelet field of normal grayscale image, little between recycling yardstick The autocorrelation of wave system number merges the wavelet details Relationship of Coefficients of interlayer sets up marginal error measure function:

B1 is as it is shown on figure 3, use the Harr small echo with orthogonality to carry out two-dimensional discrete wavelet conversion, it is achieved gray level image Four scale wavelet transform also obtain four scale wavelet transform images.

B2, obtain level according to four scale wavelet transform images, be vertically distributed with diagonal angle subband wavelet coefficient；

It is B3, independent it is assumed that to level, be vertically normalized with three, diagonal angle direction wavelet coefficient according to wavelet sub-band, Obtain the distribution of the wavelet coefficient after normalized；

B4, the autocorrelation of wavelet coefficient between yardstick is utilized to merge each multi-scale wavelet coefficient after normalization, little after being merged Wave system number；

B5, will merge after wavelet coefficient as weight, set up marginal error measure functionWherein, l represents wavelet transformation progression；LH, HL, HH table respectively Show multi-scale wavelet territory vertical direction, horizontal direction and diagonally opposed wavelet coefficient fuse information；I, j represent image pixel；B_i,j Represent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent model printer error；

Normal grayscale image is divided into k region by C, employing K-means clustering procedure respectively, recycles each Local Deviation reciprocal As weight, set up domain error measure function:

C1, use K-means clustering procedure that normal grayscale image is divided into k area image, wherein k=1,2,3,4；(will Normal grayscale image of adjusting continuously shown in Fig. 4 is divided into k area image, as k=2, obtains two shown in Fig. 5 cluster Area image, as k=3, obtains three shown in Fig. 6 cluster areas image, as k=4, obtains four shown in Fig. 7 Cluster areas image).

C2, the average calculating each region and variance；Using the inverse of variance as weight, weighted least squares is utilized to set up district Territory error metric functionWherein, k=1,2,3,4 represent cut zone；λ k represents Weight coefficient is as shown in table 1；I, j represent image pixel；B_i,jRepresent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent and beat Print machine model error；

Table 1 k is weight coefficient during different value

D, the additivity of marginal error measure function and domain error measure function is set up combined error measure function, then use Directly two-value searching method minimizes normal grayscale image and the error of coarse half tone image, obtains the digital halftone figure of optimum Picture:

D1, the additivity of marginal error measure function and domain error measure function is set up combined error measure functionWherein, λ_kRepresent weight coefficient；K=1,2,3,4 Represent cut zone；z_i,jRepresent the collimation error；w_i,jRepresent model printer error；

D2, the eight neighborhood region of each pixel in normal grayscale image and coarse half tone image is utilized direct two-value search calculate Convert in method and jump exchange, makes combined error measure function minimum；

D3, when entire image combined error measure function minimum, algorithm terminate, obtain optimum half tone image.

Obtained half-tone gradation image is carried out quality judging, to evaluate whether half tone image is optimum.Utilize quality Evaluating square mean error amount (Means Squares Error value, MSEv) and Y-PSNR (Peak Signal-to-Noise Ratio, PSNR), gained half tone image is carried out quality evaluation, square mean error amount (MSEv) represents half tone image vision Matrix, for measuring the visual deformation between original-gray image and two-value half tone image.Y-PSNR (PSNR) is one Individual expression signal maximum possible power and affect its ratio of destructive noise power of expression precision.Evaluation result such as Fig. 8 and Shown in Fig. 9, it can be seen that along with cluster areas increases to 4 from 2, MSEv value reduces 0.0129～0.2102, PSNR adds 0.53～3.17.Illustrating that the visual deformation of half tone image reduces, visual effect is preferable.Halftoning duplicating image Higher with the degree of closeness of original image.The noise that in halftoning process, image is introduced is less.

Above in conjunction with accompanying drawing, the detailed description of the invention of the present invention is explained in detail, but the present invention is not limited to above-mentioned embodiment party Formula, in the ken that those of ordinary skill in the art are possessed, it is also possible to make on the premise of without departing from present inventive concept Various changes.

Claims (5)

1. a digital picture halftoning method based on wavelet multi-scale information fusion, it is characterised in that: the tool of described method Body step is as follows:

A, continuous toned image is converted into gray level image and judges whether gray level image is the 2 of standardⁿ×2ⁿGray level image；

B, employing two-dimensional discrete wavelet conversion obtain four dimensional information of the wavelet field of normal grayscale image, little between recycling yardstick The autocorrelation of wave system number merges the wavelet details Relationship of Coefficients of interlayer sets up marginal error measure function；

Normal grayscale image is divided into k region by C, employing K-means clustering procedure respectively, recycles each Local Deviation reciprocal As weight, set up domain error measure function；

D, the additivity of marginal error measure function and domain error measure function is set up combined error measure function, then use Directly two-value searching method minimizes normal grayscale image and the error of coarse half tone image, obtains the digital halftone figure of optimum Picture.

Digital picture halftoning method based on wavelet multi-scale information fusion the most according to claim 1, its feature exists In: in described step B, what marginal error measure function was set up specifically comprises the following steps that

B1, use and there is the Harr small echo of orthogonality carry out two-dimensional discrete wavelet conversion, it is achieved four multi-scale wavelet of gray level image become Change and obtain four scale wavelet transform images；

B2, obtain level according to four scale wavelet transform images, be vertically distributed with diagonal angle subband wavelet coefficient；

It is B3, independent it is assumed that to level, be vertically normalized with three, diagonal angle direction wavelet coefficient according to wavelet sub-band, Obtain the distribution of the wavelet coefficient after normalized；

B4, the autocorrelation of wavelet coefficient between yardstick is utilized to merge each multi-scale wavelet coefficient after normalization, little after being merged Wave system number；

B5, will merge after wavelet coefficient as weight, set up marginal error measure functionWherein, l represents wavelet transformation progression；LH, HL, HH table respectively Show multi-scale wavelet territory vertical direction, horizontal direction and diagonally opposed wavelet coefficient fuse information；I, j represent image pixel；B_i,j Represent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent model printer error.

Digital picture halftoning method based on wavelet multi-scale information fusion the most according to claim 1, its feature exists In: in described step C, what domain error measure function was set up specifically comprises the following steps that

C1, use K-means clustering procedure that normal grayscale image is divided into k area image, wherein k=1,2,3,4；

C2, the average calculating each region and variance；Using the inverse of variance as weight, weighted least squares is utilized to set up district Territory error metric functionWherein, k=1,2,3,4 represent cut zone；λ_kRepresent Weight coefficient；I, j represent image pixel；B_i,jRepresent pixel eight neighborhood；z_i,jRepresent the collimation error；w_i,jRepresent that model printer is by mistake Difference.

Digital picture halftoning method based on wavelet multi-scale information fusion the most according to claim 1, its feature exists In: in described step D, direct two-value searching algorithm minimizes specifically comprising the following steps that of combined error measure function

D1, the additivity of marginal error measure function and domain error measure function is set up combined error measure functionWherein, λ_kRepresent weight coefficient；K=1,2,3,4 Represent cut zone；z_i,jRepresent the collimation error；w_i,jRepresent model printer error；

D2, the eight neighborhood region of each pixel in normal grayscale image and coarse half tone image is utilized direct two-value search calculate Convert in method and jump exchange, makes combined error measure function minimum；

D3, when entire image combined error measure function minimum, algorithm terminate, obtain optimum half tone image.

5., according to the digital picture halftoning method based on wavelet multi-scale information fusion described in claim 1 or 4, it is special Levy and be: described coarse half tone image uses error-diffusion method process to obtain.