CN1852392A - Printing net-point-image dividing method based on moveable contour - Google Patents

Abstract

The invention works out following improvements for classical C-V method in model of geometrical moving contour line: considering relativity among channels, the disclosed method uses spatial distance between pixel vector values in color image instead of difference value of gray scale in gray scale method; using C-V method for color separation so as to use color information in each color channel effectively; adding global factor to PDE equation in C-V method makes C-V method become global method; using transform of color space makes utilized color information even so as to accord with perception of human eyes. The invention is applicable to segmentation of color halftone dot image. Experimental verificates that comparing with classical color C-V method, the disclosed color C-V method enhances capacity for detecting color edge and blur edge, as well as capacity for anti noise, and for detecting hole with thick wall. The invention can obtain edges of halftone dots in color image.

Description

Printing net-point-image dividing method based on moveable contour

Technical field

The present invention relates to a kind of method of technical field of image processing, specifically is a kind of printing net-point-image dividing method based on moveable contour, can directly apply to cutting apart the halftone dot image in the colored printing reproduction process.

Background technology

The identification of halftone dot image should be carried out in conjunction with the concrete property of site in the printing image.Traditional site dividing method has index entropy method, Otsu method etc., and these methods are based on all that gray scale cuts apart, and for background complexity, edge blurry, the more color screen dot image of noise, the result is unsatisfactory.Cut apart for the ideal that realizes halftone dot image, obtain edge, accurate site, the problem that at first needs to solve is how to choose suitable partitioning algorithm and model, and how traditional gray scale dividing method is expanded on the coloured image.Color images utilizes colour information with specific in the correspondence image, and the zone with peculiar property is the process of separation and Extraction in addition.Although the dividing method of gray level image and strategy reach its maturity, wherein many methods also are not suitable for directly cutting apart coloured image.In recent years, a lot of theoretical tools and model all once were effectively applied to cutting apart of coloured image.But how the concrete image of being studied is chosen suitable color property space and its corresponding theory instrument, be still coloured image is carried out the key that success is cut apart.Active contour line method (having another name called the snakes method) has the incomparable advantage of some classical image partition methods: pictorial data, initial estimation, objective contour and be unified in the characteristic extraction procedure based on the constraint of knowledge; Through after the initialization suitably, can independently converge on the minimum state of energy.But the Snake model also has the defective of himself: to the initial position sensitivity, need to rely on other mechanism and place it near the interested image feature; It might converge to local extremum, even disperses; Can't handle the change in topology of curve.

Find through literature search prior art, in order to solve these problems that traditional snakes model exists, Osher and Sethian are at " Journal of Computational Physics " 1998,79 (1), the article of delivering on the pp.12-49 " Fronts propagating with curvature-dependent speed:Algorithmbased on Hamilton-Jacobi formulations " (" the curvature speed based on Hamilton-Jacobi equation develops ", " Computational Physics ") in a kind of Level Set Method has been proposed, the closed curve of two dimension is embedded the curved surface of a three-dimensional, realize the evolution of curve by the evolution of curved surface.But, only utilize image edge information based on the level set algorithm of geometric active contour line model, to edge blurry or exist the target at discrete shape edge to be difficult to obtain desirable segmentation effect.Chan and Vese are at " IEEE Trans Image Processing " 2001,10 (2), " Tony F.Chan and Luminita A.Vese. (2001): " Activecontours without edges " (" unskirted moveable contour "; " IEEE image processing ") proposed a kind of Chan-Vese method to the article of delivering on the pp.266-277; be called for short the C-V method; this method is utilized the global information of the homogeneous region of image, can be partitioned into the target of obscurity boundary or divergent boundary preferably.But the C-V method only is confined to gray scale to be cut apart, and also is not applied to coloured image preferably, and several methods that proposed can not keep its advantage of cutting apart to fuzzy edge and divergent margin in color images.

Summary of the invention

The objective of the invention is at the deficiencies in the prior art, a kind of printing net-point-image dividing method based on moveable contour is provided, it is improved the Chan-Vese method in the geometric active contour line (hereinafter to be referred as the C-V method), it is expanded to the color space.And Color C-V Method is applied to halftone dot image.Method after the improvement has strengthened the anti-noise ability cut apart and the detectability of fuzzy edge, and has avoided gray scale to cut apart the problem that causes, and has improved the segmentation effect to halftone dot image greatly.

The present invention is achieved by the following technical solutions, the present invention is to the classical C-V method in the geometric active contour line model, done following improvement: consider interchannel correlation, with the gray scale difference value in the alternative gray scale method of the space length between the pixel vector value in the coloured image, the C-V method is generalized to colour cuts apart up, make algorithm can really effectively utilize colour information in each color channel; By in the PDE of C-V method equation, adding global factor, really make the C-V method become the method for globalize; And the conversion by color space, the color information even that makes algorithm as far as possible and utilized meets perception of human eyes.The inventive method is applied to during the color screen dot image cuts apart, can either make full use of image information, again can the advantage of inheriting tradition C-V method on noise, weak edge treated, offer help for follow-up print production.

Described with the gray scale difference value in the alternative gray scale method of the space length between the pixel vector value in the coloured image, be meant: with each pixel pixel value in the coloured image in some fixed color space (GRB for example, YUV) Euclidean distance in, the difference of pixel value between the description object vegetarian refreshments, and in the Mathematical Modeling of original gray scale C-V method, gray scale difference value between the pixel that substitutes with this Euclidean distance has so just been finished the C-V model has been cut apart expansion on colour is cut apart from gray scale.By this replacement, colour information has been joined in the parted pattern effectively, has strengthened segmentation effect.

Described by in the PDE of C-V method equation, adding global factor, make the C-V method become the method for globalize, be meant: in the Mathematical Modeling of C-V method, be in the energy functional PDE equation of C-V method, add an item that comprises image overall information, make these global informations participate in whole cutting procedure, have influence on last segmentation result, the realization globalize is cut apart.By adding global factor, strengthened the global segmentation ability of C-V method, and avoided leaking the situation appearance of cutting apart.

Described conversion by color space, make the color information even of being utilized, be meant: the conversion formula that utilizes color space, the RGB pixel value of color pixel is converted into the YUV pixel value, and in the Mathematical Modeling of C-V method, replace rgb value with the YUV pixel value, so just finished the conversion of color space.Because yuv space more meets and perception of human eyes, and colouring information is more even, the conversion of color space has strengthened the image segmentation effect undoubtedly.

The inventive method is applied to during the color screen dot image cuts apart, the color screen dot image that occurs in the printing process is carried out image segmentation, to ask for the edge, site in the image.C-V method after the improvement has very strong segmentation ability for the image that contains noise, weak edge, and owing to comprised global information, can not occur leaking and cut apart, and is applicable to that just in time noise is more, edge blurry, and multiobject halftone dot image.

The present invention really combines the C-V method effectively with colour information, remedy gray scale and cut apart deficiency to coloured image, makes full use of the big characteristics of color image information amount, extracts edge feature, improves segmentation effect.And it is applied to colored site cut apart, obtain gratifying segmentation result.

Description of drawings

Fig. 1 is the noise schematic diagram that contains the heavy wall hole

Wherein: (a) being the noise pattern that a width of cloth contains the heavy wall hole, (b) is the segmentation result of original C-V method to it, is the segmentation result of Color C-V Method (c), (d)-(h) position of the different initial curves of expression.

Fig. 2 is that LUV and RGB segmentation effect compare schematic diagram

Wherein: (a) being original image, (b) is the RGB segmentation effect, (c) LUV segmentation effect.

Fig. 3 is the comparison schematic diagram of Color C-V Method of the present invention and additive method

Wherein: (a) being original image, (b) is the Otsu method, (c) is index entropy method, (d) is the FCM method, (e) is Color C-V Method.

Fig. 4 is the halftone dot image and the segmentation effect schematic diagram of fuzzy edge

Wherein: (a) being former figure, (b) is segmentation result, (c) is former figure, (d) segmentation result.

Embodiment

Below in conjunction with specific embodiment technical scheme of the present invention is described in further detail.

The specific implementation method of Color C-V Method of the present invention is carried out as follows:

1. set up the Chan-Vese model

Chan and Vese have proposed a kind of moveable contour image partition method based on the Mumford-shah model, this method stop the partial gradient that function is no longer dependent on image, and be based on the Mumford parted pattern.Like this, this method is applicable to the detection of the profile of gradient simultaneously, the very level and smooth or discontinuous image segmentation in edge such as those edges.In addition, therefore the model definition of this method can detect the interior zone that has empty target in all level sets.

Below be the Mumford-Shah Image Segmentation Model of simplifying:

If image I (x, domain of definition y) is Ω, and the image border C of current investigation (x y) is divided into some approximate regionally with values, obtains split image I with image I _o(x, y), then the Mumford-Shah Image Segmentation Model is exactly to seek real image boundary C _o(x y) is divided into some homogeneous regions, and gained split image I with image I _o ^MS(x, y) and I (x, the error of all split images of error ratio y) and original image is all little.Structure energy functional F ^MS(I _o, C), make and work as F ^MS(I _o, C) hour, gained border C _oWith image segmentation is some smooth regions, and keeps sharp borders.F ^MS(I _o, expression formula C) is

$F^{\mathrm{MS}}(I_o,C)=\mathrm{\μLength}\left(C\right)+\mathrm{\λ}\underset{\mathrm{\Ω}}{\∫}{|I-I_o|}^2\mathrm{dxdy}+\underset{\mathrm{\Ω}/C}{\∫}{|\▿I_o|}^2\mathrm{dxdy}$

Chan and Vese have proposed based on above simplified models parted pattern, establish promptly that the gray scale of each homogeneous region is a constant in the image, then the minimization of energy function F ^MS(I _o, purpose C) transforms in order to seek optimum segmentation C _o, make difference minimum between split image and the original image.The mathematical expression of Chan-Vese model is as follows:

If (x y) is divided into target ω by moveable contour C to original image I _oWith background ω _bTwo zones.Each regional average gray is C _oAnd C _b, the match energy function of Chan-Vese model is as follows:

$F\left(C\right)=\mathrm{\μL}\left(C\right)+v{S}_o\left(C\right)+{\mathrm{\λ}}_o{\∫}_{\mathrm{inside}\left(C\right)}{|I-c_o|}^2\mathrm{dxdy}+{\mathrm{\λ}}_b{\∫}_{\mathrm{outside}\left(C\right)}{|I-c_b|}^2\mathrm{dxdy}$

C in the following formula is any closed moveable contour, and L (C) is the length of closed contour line C; S _o(C) be the interior zone area of C, μ, v 〉=0, λ _o, λ _bThe＞0th, the heavy coefficient of each energy necklace; Two is flat picture item.As can be seen from the above equation, be not positioned at the border C of two homogeneous regions as closed moveable contour C _oThe time, F (C) can not reach minimum value.

2.C-V method is to the expansion of N dimensional vector value image

The core concept of this expansion is exactly the gray value in the alternative traditional C of Euclidean distance-V method of utilizing between pixel value, effectively utilizes three relevant informations between Color Channel to reach, and improves colored segmentation effect.Make u _{O, i}Be i channel value of image pixel value, i=1 ... N, C are moveable contour.For the N dimension space, order $\stackrel{\‾}{c^{+}}=({c_1}^{+},...,{c_N}^{+}),$ And $\stackrel{\‾}{c^{-}}=({c_1}^{-},...,{c_N}^{-})$ Be the N n dimensional vector n.Euclidean distance between amount of orientation replaces gray value, and obtaining the C-V model that the N n dimensional vector n cuts apart is that its expression formula is as follows:

$F\left(\stackrel{\‾}{c^{+}},\stackrel{\‾}{c^{-}},\mathrm{\φ}\right)=\mathrm{\μ}\·L\left(C\right)+v{S}_0\left(C\right)+{\∫}_{\mathrm{inside}\left(C\right)}\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left({{\mathrm{\λ}}_i}^{+}\right)}^2{(u_{0,i}(x,y)-{c_i}^{+})}^2}\mathrm{dxdy}$

$+{\∫}_{\mathrm{outside}\left(C\right)}\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left({{\mathrm{\λ}}_i}^{-}\right)}^2{(u_{0,i}(x,y)-{c_i}^{-})}^2}\mathrm{dxdy}$

The λ here _i ⁺＞0, λ _i ^-＞0, be the parameter weight of each passage.Its value is between (0,1); Parameter μ and λ are representing the sensitivity of rim detection, and μ and λ are big, and then model is easy to filter away high frequency noise, and on the contrary, μ and λ are little, and then model is good to the edge details segmentation effect, and its value is also between (0,1).

As can be seen from the above equation, model is exactly to seek vector

Optimal approximation so that

Minimum.Moveable contour C is the edge of two homogeneous regions, and preceding two of equation the right is level and smooth, and two of back are energy term.Under this form, the present invention utilizes the Euclidean distance of vector space that N passage combined, and effectively utilizes the marginal information of each passage, and the distance that obtains like this meets the physical significance of hyperspace more, and the problem of having avoided simple weighted to cause.

Under the level set form, utilize Euler-Lagrange equation to find the solution the formula energy function and finally obtain:

$\frac{\∂\mathrm{\φ}}{\∂t}={\mathrm{\δ}}_{\mathrm{\ϵ}}[\mathrm{\μ}\·\mathrm{div}\left(\frac{\▿\mathrm{\φ}}{|\▿\mathrm{\φ}|}\right)-\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{{\mathrm{\λ}}_i}^{+}{(u_{0,i}-{c_i}^{+})}^2}+\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{{\mathrm{\λ}}_i}^{-}{(u_{0,i}-{c_i}^{-})}^2}]\·$

As can be seen from the above equation, the effective range of δ function directly has influence on the global property of model.

3. adding global factor

Because the definition of the δ value in the original C-V method causes its sphere of action to only limit to around it, for this reason, the present invention replaces with the δ in the energy function |  _φ|, to strengthen the of overall importance of equation, make the effective range of equation expand to the entire image zone.

For coloured image, N=3, therefore, the level set form that has added the coloured image of global factor is:

$\frac{\∂\mathrm{\φ}}{\∂t}=|\▿\mathrm{\φ}|[\mathrm{\μ}\·\mathrm{div}\left(\frac{\▿\mathrm{\φ}}{|\▿\mathrm{\φ}|}\right)-\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{{\mathrm{\λ}}_i}^{+}{(u_{0,i}-{c_i}^{+})}^2}+\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{{\mathrm{\λ}}_i}^{-}{(u_{0,i}-{c_i}^{-})}^2}]$

This Color C-V Method based on the M-S model is not only respond well to the image segmentation that is subjected to noise pollution or obscurity boundary; Can also solve the problem that has the hole time-division to cut in the image well, realize global optimization.Fig. 1 (a) is the noise pattern that a width of cloth contains the heavy wall hole, and Fig. 1 (b) is the segmentation result of original C-V method to it, and Fig. 1 (c) is the segmentation result of Color C-V Method, can see: the Color C-V algorithm can detect the hole of heavy wall, has improved segmentation ability.In addition, this method also has the robustness to the initial curve position, and as the position of the different initial curves of Fig. 1 (d)-(h) expression, they can both obtain the effect as Fig. 1 (c), the location independent of segmentation result and initial curve.

4. the selection of color space

Under many circumstances, the RGB color space can not really reflect the color distortion on the human eye vision.Often appear to have two kinds of colors than big difference, the distance in rgb space is but very little, is easy to when cutting apart miss.In Fig. 2, (a) be original image, (b) in below square do not detect.In order to make the image that splits meet perception of human eyes more.The present invention has selected for use more even than the distribution of RGB chrominance space, meets the LUV chrominance space of human eye vision perception more.By the conversion of RGB-LUV, can see that the LUV segmentation result after the conversion can identify following square, as Fig. 2 (c), improved segmentation effect.

5. numerical solution

Selective entropy averaging method of the present invention is asked for numerical solution to final level set expression formula.The concrete form of numerical solution is:

φ _i，j ⁿ⁺¹＝φ _i，j ⁿ+Δt{-[max(F _m-s，0) ⁺+min(F _m-s，0) ⁺]

-[max(u _ij ⁿ，0)D _ij ^-x+min(u _ij ⁿ，0)D _ij ^+x]

-[max(v _ij ⁿ，0)D _ij ^-y+min(v _ij ⁿ，0)D _ij ^+y]

+μk _i，j[(D _i，j ^0y) ²+(D _i，j ^0x) ²] ^1/2}

Here u _Ij ⁿ, v _Ij ⁿBe g _ColorAt the partial differential of x and y direction, k _IjBe level set point (i, the curvature of j) locating, and

F _m-s＝-v-λ _o[I(x，y)-c _o] ²+λ _b[I(x，y)-c _b] ²，

 ⁺＝[max(D _ij ^-x，0) ²+min(D _ij ^+x，0) ²+max(D _ij ^-y，0) ²+min(D _ij ^+y，0) ²] ^1/2，

 ^-＝[min(D _ij ^-x，0) ²+max(D _ij ^+x，0) ²+min(D _ij ^-y，0) ²+max(D _ij ^+y，0) ²] ^1/2。

The present invention is applied to halftone dot image with above-mentioned Color C-V Method, has strengthened the anti-noise ability cut apart and the detectability of fuzzy edge; To cut apart and expand to the view picture halftone dot image, avoid leaking segmentation problem; And make full use of colour information, avoided gray scale to cut apart the problem that causes, improved segmentation effect greatly halftone dot image.Fig. 3 has provided Color C-V Method and additive method to the segmentation effect of complex background halftone dot image relatively.(a) being original halftone dot image, (b) is the segmentation result of Otsu method, (c) is the segmentation result of index entropy method, (d) is the segmentation result of FCM method, (e) is the segmentation result of Color C-V Method.

Fig. 4 is the segmentation effect to ill-defined halftone dot image.As seen from Figure 4, Color C-V Method can be good at identifying fuzzy edge, reflects the shape and the size of site truly.

Claims (4)

1, a kind of printing net-point-image dividing method based on moveable contour, it is characterized in that, to the classical C-V method in the geometric active contour line model, done following improvement: consider interchannel correlation, with the gray scale difference value in the alternative gray scale method of the space length between the pixel vector value in the coloured image, the C-V method is applied to colour cuts apart up, effectively utilize the colour information in each color channel; By in the PDE of C-V method equation, adding global factor, make the C-V method become the method for globalize; And the conversion by color space, make the color information even of being utilized, meet perception of human eyes.

2, printing net-point-image dividing method based on moveable contour according to claim 1, it is characterized in that, described with the gray scale difference value in the alternative gray scale method of the space length between the pixel vector value in the coloured image, be meant: with the Euclidean distance of each pixel pixel value in some fixed color space in the coloured image, the difference of pixel value between the description object vegetarian refreshments, and in the Mathematical Modeling of original gray scale C-V method, gray scale difference value between the pixel that substitutes with this Euclidean distance has so just been finished the C-V model has been cut apart expansion on colour is cut apart from gray scale.

3, the printing net-point-image dividing method based on moveable contour according to claim 1, it is characterized in that, described by in the PDE of C-V method equation, adding global factor, make the C-V method become the method for globalize, be meant: in the Mathematical Modeling of C-V method, promptly in the energy functional PDE equation of C-V method, add an item that comprises image overall information, make these global informations participate in whole cutting procedure, have influence on last segmentation result, the realization globalize is cut apart.

4, the printing net-point-image dividing method based on moveable contour according to claim 1, it is characterized in that, described conversion by color space, make the color information even of being utilized, be meant: utilize the conversion formula of color space, the RGB pixel value of color pixel is converted into the YUV pixel value, and in the Mathematical Modeling of C-V method, replace rgb value with the YUV pixel value, so just finished the conversion of color space.

Images