CN102999886A - Image edge detector and ruler raster grid line precision detection system - Google Patents

Abstract

The invention relates to an image edge detector and a ruler raster grid line precision detection system comprising the same. The detector comprises an image filtering module, a primary edge extraction module, a threshold segmentation module, a secondary edge extraction module and an edge thinning module, wherein the image filtering module is used for filtering image noise to obtain a filtered image, the primary edge extraction module uses a double-trapezoidal edge extraction operator for extracting the edge of a filtered image to obtain a new image, the threshold segmentation module selects an optimal segmentation threshold T of the new image obtained by the primary edge extraction module and then performs binarization processing for the image, the secondary edge extraction module extracts the edge of the segmented image again by the aid of the double-trapezoidal edge extraction operator, and the edge thinning module thins the edge extracted by the secondary edge extraction module and removes burrs. The edge of the image can be extracted clearly and accurately, and edge contrast is quite high.

Description

Image Edge Detector and scale grating grid precision detection system

Technical field

The invention belongs to technical field of image processing, relate to a kind of Image Edge Detector and comprise the scale grating grid precision detection system of this detecting device.

Background technology

The edge is the most basic feature of image, and rim detection plays an important role in the application such as computer vision, graphical analysis, is the important step of graphical analysis and identification, and this is because the edge of image has comprised the useful information that is used for identification.So being the principal character of graphical analysis and pattern-recognition, rim detection extracts means.

Classical, the simplest edge detection method is to original image certain neighborhood structure boundary operator according to pixels, because original image often contains noise, and edge and noise show as gray scale in spatial domain larger rising and falling arranged, then react for being both high fdrequency component at frequency domain, this just brings difficulty to rim detection.

Long grating has been widely used in various surveying instruments, lathe digital display, the Numeric Control Technology as a kind of novel measuring element.The making precision of long grating will directly have influence on the measuring accuracy of testing tool (such as three coordinate machine etc.), and the machining precision of workpiece.Therefore to improve the making precision of grating scale and the reliability that long grating is used, must the precision index of grating scale be detected, analyze the error component in the long grating manufacture process.For guaranteeing the work of optical grating measuring system accurate stable ground, large, the sinusoidal property of electrical signal amplitude of requirement optical grating Moire fringe will be got well, the ratio of bright black level is wanted large (being that contrast will be got well), require simultaneously measuring on the total length, the variation of grating signal amplitude is little, the variation of DC level and drift variation little, the biphase signaling phase differential are little.Practical application shows: the quality of grating signal depends primarily on the quality of scale grating.For this reason, with the variation of grating signal amplitude and DC level on the total length, the variation of biphase signaling phase differential and grating grid precision etc. are three on the total length usually, as the leading indicator of estimating the grating scale quality.Practical application also shows: the precision of optical grating measuring system depends primarily on the precision of scale grating grid.Detect the precision of grating grid and at first will extract accurately the Moire fringe edge line, then by calculating the pitch of grating.

Summary of the invention

The technical matters that the present invention will solve provides a kind ofly can extract the Image Edge Detector of image border clear, exactly.

In order to solve the problems of the technologies described above, Image Edge Detector of the present invention comprises:

The image filtering module: filtering image noise obtains filtered image;

An edge extracting module: utilize the double trapezoid arithmetic operators to extract the edge of image, the convolution mask of this operator is as follows:

$G_x=\left[\begin{array}{ccc}& m_7& \\ m_1& & m_4\\ m_2& m_8& m_5\\ m_3& & m_6\\ & m_9& \end{array}\right]$ $G_y=\left[\begin{array}{ccccc}& m_1& m_2& m_3& \\ m_7& & m_8& & m_9\\ & m_4& m_5& {\mathrm{<figure-callout\; id="6"\; label="m"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">m</figure-callout>}}_6& \end{array}\right]$

Wherein: m ₁~ m ₉All be integer, G _xThe arithmetic operators of horizontal direction, G _yThe arithmetic operators of vertical direction, m ₁=m ₃=-m ₄=-m ₆, m ₂=-m ₅, m ₂〉=2 * m ₁, m ₅〉=2 * m ₄, m ₇=m ₉＜0, m ₈=-(m ₇+ m ₉=)=-2 * m ₇=-2 * m ₉, | m ₇|≤| m ₁|; If the minor increment in the image between the adjacent two edges is d, then when the pixel of d＜20, | m ₁|=1 or 2,2≤| m ₂|≤5; When the pixel of d 〉=20,2|m ₁|≤5≤, 5≤| m ₂|≤10;

Filtered image and convolution mask are obtained a width of cloth new images as convolution algorithm;

Threshold segmentation module: select the optimal segmenting threshold T of the new images that edge extracting module obtains, then new images is made binary conversion treatment, will put maximum gradation value N greater than the pixel of optimal segmenting threshold T _h, will set to 0 less than the pixel gray scale of optimal segmenting threshold T, thereby obtain a width of cloth binary image;

Second edge extraction module: utilize the method identical with edge extracting module that Threshold segmentation image is afterwards carried out edge extracting again;

The edge thinning module: the edge that the second edge extraction module is extracted carries out refinement, removes burr.

Image filtering is the pretreatment stage of Image Edge-Detection, mainly is filtering image noise, for follow-up edge extracting is prepared.

Edge extracting is normally realized by convolution by the spatial domain derivative operator.In fact this process is similar to by difference method and finishes.The corresponding single order of gradient or second derivative, people have proposed the different operator of many kinds at present, such as Robert Cross operator, Prewitt operator, Kirsch operator and Sobel operator etc.Because the contrast on border of these operator extraction is lower, is unfavorable for follow-up Threshold segmentation, therefore the present invention proposes a kind of new arithmetic operators-double trapezoid operator.This operator can extract the edge of image clear, accurately, and the contrast at edge is very high.

The contrast formula: $C=\sqrt{\frac{1}{M\&times;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{[f(i,j)-\stackrel{\&OverBar;}{f}]}^2}$

The average gradient formula: $G=\sqrt{\frac{1}{M\&times;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}[{(f(i,j)-f(i-1,j))}^2+{(f(i,j)-f(i,j-1))}^2]}$

Wherein f (i, j) is the capable j row of the i pixel of original image,

Be the average gray value of original image, original image size is the capable N row of M.

Utilize above several arithmetic operators that original image is carried out edge extracting, picture contrast and average gradient correlative value after the edge extracting, as shown in the table:

	Robert Cross	Prewitt	Kirsch	Sobel	The double trapezoid operator
						Contrast
	5	16	43	21	65
						Average gradient	2	6	18	8	25

Average gradient is larger, and the edge of key diagram picture is more clear, and contrast is larger, illustrates that the contrast at edge is higher.Double trapezoid operator contrast of the present invention and average gradient illustrate that this operator can extract the edge of image clear, accurately, and the contrast at edge are very high all greater than other operators as can be seen from the table.

Because an edge extracting result has only described the rough local edge of piece image, therefore need to do further aftertreatment to rough edge image.This aftertreatment comprises the processing such as Threshold segmentation, edge thinning.

After the Threshold segmentation, the edge line of image is thicker, and for the refinement edge line, the present invention utilizes the double trapezoid arithmetic operators that the image after the Threshold segmentation is carried out edge extracting again, makes edge line be refined as single pixel.Because the edge line after the refinement is subsidiary many " burrs " simultaneously.The edge that the present invention utilizes the edge thinning module that the second edge extraction module is extracted is further processed and removes these " burrs ", obtains thus clear, image border line accurately.

Described image filtering module adopts gaussian filtering method filtering noise.

Image filtering is the pretreatment stage that carries out the grating grid accuracy detection, mainly is filtering image noise, for follow-up edge extracting is prepared.In the present invention, adopt effectively filtering noise of gaussian filtering method.

Described Threshold segmentation module is calculated the new images statistic histogram, and the statistic histogram envelope is fitted to a smooth curve, then with the minimum point of the smooth curve that finds as a setting with the optimal segmenting threshold T of edge line.

Described Threshold segmentation module adopts the Research on threshold selection based on the histogram envelope line.The basic thought of the method is that the histogrammic envelope of image statistics is fitted to a smooth curve, and the minimum point of the smooth curve that finds is the optimal segmenting threshold of background and edge line.The present invention is by the histogrammic characteristic distributions of image statistics, with the minimal value of image histogram envelope representing background and target intersection gray-scale value as optimal segmenting threshold, image is carried out binary conversion treatment, thereby can be partitioned into exactly the edge line of image.

Described Threshold segmentation module is carried out smothing filtering with the new images statistic histogram, adopts the single order differential method to obtain the local maximum value set of filtering statistic histogram afterwards; Utilize the local maximum value set to carry out curve fitting, obtain the curve minimum point after the match, and the gray-scale value that this curve minimum point is corresponding is as optimal segmenting threshold T.

Described Threshold segmentation module utilizes the matlab programming to carry out curve fitting.

Described image filtering module, edge extracting module, Threshold segmentation module, second edge extraction module, an edge thinning module realize by the com component of VC++ and matlab hybrid programming.

The present invention does not directly realize the least square fitting process by VC++ when carrying out curve fitting, computation process is very complicated like this, and calculated amount is very large.But adopt the matlab programming, and realize by the com component of VC++ and matlab hybrid programming, greatly reduced calculated amount.Calculate very simply, and it is accurate to ask for threshold value for the larger situation of background and target contrast.

Described edge thinning module stores has a plurality of 4 * 3 to eliminate template, and 4 * 3 elimination templates are as follows:

$\left[\begin{array}{ccc}{P}_1& P_2& P_3\\ P_4& P_5& P_6\\ P_7& P_8& P_9\\ P_{10}& P_{11}& P_{12}\end{array}\right]$

Eliminate template and satisfy simultaneously following four conditions:

A, P ₅Eight neighborhood elements in 2 ~ 6 elements are arranged is 1, all the other elements are 0, i.e. N=P ₁+ P ₂+ P ₃+ P ₄+ P ₆+ P ₇+ P ₈, 2≤N≤6;

B, P ₂And P ₈Have at least one to be zero, i.e. P ₂* P ₈=0;

C, P ₅Eight neighborhood elements circulate in the direction of the clock or by counterclockwise circulation only have one 0,1 discontinuous point;

D, P ₄, P ₆And P ₈In have at least one to be zero, i.e. P ₄* P ₆* P ₈=0; Perhaps P ₈Eight neighborhood elements circulate in the direction of the clock or do not have 0,1 discontinuous point or have greater than 10,1 discontinuous point by counterclockwise circulation;

Begin search from the binary image top left corner pixel, if the current pixel gray-scale value is 0, then skip; If the current pixel gray-scale value is N _h, then make this pixel corresponding to the element P that eliminates template ₅, with other pixels around this pixel with eliminate on the template correspondence position element and compare, template is identical then to be set to 0 the current pixel gray scale if eliminate with one of them, otherwise the current pixel gray-scale value is constant; Repeat said process, until the neither one grey scale pixel value is changed in the binary image, edge thinning finishes.

The kind of edge thinning method is a lot, can be divided into according to the order of refinement: serial refinement, parallel thinning and mixing refinement.The thinning method that the present invention adopts belongs to the serial refinement, and its principle is a plurality of elimination templates of structure, and binary image and elimination template are compared, and determines whether to delete certain point.It is thorough that the method that the present invention adopts not only can refinement, and the single pixel line after the refinement is at the center line of edge line, and the Glabrous thorn, can guarantee the accuracy of next step calculating.

Edge thinning refers on the basis that does not affect the edge line connectedness, and the edge pixel of deletion edge line removes " burr " on the straight line, make edge line be single pixel wide till.Edge line skeleton after the desirable refinement should be the centre position of original edge line, and the connectivity of keep the edge information line, topological structure and minutia.A kind of good thinning algorithm should satisfy following condition:

(1) convergence: iteration must restrain

(2) connectivity: the connectivity of not destroying edge line

(3) Topological: do not cause progressively eating of edge line, keep the basic structure characteristic of original image

(4) retentivity: the minutia of protection edge line

(5) refinement: the width of skeleton edge line is 1 pixel, and namely single pixel is wide.

(6) axis: skeleton is as far as possible near stripe centerline

(7) rapidity: algorithm is simple, and speed is fast.

The edge thinning method that the present invention adopts can satisfy above 7 conditions simultaneously, extracts accurately the skeleton of edge line, removes " burr ".

Another technical matters that the present invention will solve provides a kind of scale grating grid precision detection system that comprises above-mentioned Image Edge Detector.

In order to solve the problems of the technologies described above, scale grating grid precision detection system of the present invention also comprises VTOL (vertical take off and landing) platform 1, be fixed in the CCD camera 2 on the VTOL (vertical take off and landing) platform 1, be connected to the enlarging lens 5 on the CCD camera 2, be installed in the coaxial light source 6 on the enlarging lens 5, image pick-up card 3, grid line width computing module; The directional light that coaxial light source 6 sends out impinges upon the grid line district of scale grating 8 from the lens barrel of enlarging lens 5; CCD camera 2 gathers the grating grid area image and sends image pick-up card 3 to; Image pick-up card 3 with the raster image data transmission that gathers to Image Edge Detector, Image Edge Detector is extracted the grid line of raster image, every grid line evenly is divided into M part in the image that grid line width computing module detects Image Edge Detector, tries to achieve the grid line width of every portion; Then M grid line width removed N maximal value, remove N minimum value, M-2N value of centre is averaging the width that obtains each bar grid line; All grid line width of entire image are removed Q maximal value, remove Q minimum value, then average, draw the average grating grid width of this image; M wherein〉2N, the grid line number〉2Q.

Description of drawings

Below in conjunction with the drawings and specific embodiments the present invention is described in further detail.

Fig. 1 is Image Edge Detector functional module framework figure of the present invention.

Fig. 2 is scale grating grid precision detection system structural representation of the present invention.

Fig. 3 is scale grating grid precision detection system functional module framework figure of the present invention.

Fig. 4 is the statistic histogram of the new images that obtains of edge module.

Fig. 5 is the original image in the scale grating grid zone arrived of collected by camera.

Fig. 6 is the superimposed image of detected grating edge line image and original image.

Embodiment

As shown in Figure 1, Image Edge Detector of the present invention comprises image filtering module, edge extracting module, Threshold segmentation module, second edge extraction module and an edge thinning module.

Image is in the process that gathers and transmit, various noises tend to mix, cause the Quality Down of image, this edge extracting for image has caused very large difficulty, in order better to carry out edge extracting, must carry out first filtering, the image filtering module adopts preferably gaussian filtering of denoising effect among the present invention.The discrete Gaussian function expression formula of two dimension zero-mean is:

$g[i,j]=e^{-\frac{({\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2)}{2{\mathrm{\&sigma;}}^2}}$

Gather image and Gaussian function convolution and obtain filtered image.

One time the edge extracting module adopts a kind of new double trapezoid arithmetic operators that the image after the filtering is carried out edge extracting, and this operator can extract the edge of image clear, accurately, and the contrast at edge is very high, is conducive to follow-up Threshold segmentation.The convolution mask of this operator is as follows:

$G_x=\left[\begin{array}{ccc}& m_7& \\ m_1& & m_4\\ m_2& m_8& m_5\\ m_3& & m_6\\ & m_9& \end{array}\right]$ $G_y=\left[\begin{array}{ccccc}& m_1& m_2& m_3& \\ m_7& & m_8& & m_9\\ & m_4& m_5& {\mathrm{<figure-callout\; id="6"\; label="m"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">m</figure-callout>}}_6& \end{array}\right]$

Wherein: m ₁~ m ₉All be integer, G _xThe arithmetic operators of horizontal direction, G _yIt is the arithmetic operators of vertical direction.m ₁=m ₃=-m ₄=-m ₆, m ₂=-m ₅, m ₂〉=2 * m ₁, m ₅〉=2 * m ₄, m ₇=m ₉＜0, m ₈=-(m ₇+ m ₉=)=-2 * m ₇=-2 * m ₉, | m ₇|≤| m ₁|; If the minor increment in the image between the adjacent two edges is d, then when the pixel of d＜20, | m ₁|=1 or 2,2≤| m ₂|≤5; When the pixel of d 〉=20,2|m ₁|≤5≤, 5≤| m ₂|≤10.

Filtered image and convolution mask are obtained a width of cloth new images as convolution algorithm.

The Threshold segmentation module is carried out Threshold segmentation to new images.

Although the new images contrast that edge extracting module obtains is higher, but there is larger transitional region between background and the edge line, want to be partitioned into accurately edge line, to such an extent as to make too slightly too carefully fracture of edge line, must select suitable threshold value.The method that Threshold segmentation module of the present invention adopts is based on the Research on threshold selection of histogram envelope line.The basic thought of the method is that the histogrammic envelope of image statistics is fitted to a smooth curve, finds the minimum point of smooth curve to be the optimal segmenting threshold of background and edge line.The Threshold segmentation module at first adopts the method for differential to obtain statistic histogram envelope local maximum, then these local maximums is fitted to smooth curve again.When carrying out curve fitting, directly do not realize the least square fitting process by VC++, computation process is very complicated like this, and calculated amount is very large.But adopt the matlab programming, and realize by the com component of VC++ and matlab hybrid programming, greatly reduced calculated amount.The method is very simple, and it is accurate to ask for threshold value for the larger situation of background and target contrast.The specific implementation process is as follows:

A, at first draw the statistic histogram (referring to Fig. 4) of new images, P in the statistic histogram (i) is the pixel number of i for gray level on the image;

B, statistic histogram is carried out smothing filtering, that is:

$\begin{array}{c}P\left(i\right)=(P(i-u)+P(i-u+1)+......P(i-2)+P(i-1)+P\left(i\right)+P(i+1)+P(i+2)+......\\ +P(i+u))/(2u+1)\end{array}$

U is natural number in the formula, and the selection of its numerical value does not have strict regulation, and the larger filter effect of numerical value is better, but 3≤u≤7 are generally selected in the corresponding increase of calculated amount;

C, the employing single order differential method are obtained the local maximum value set M (j) of the statistic histogram after the filtering, and wherein j is gradation of image value corresponding to Local modulus maxima;

D, local maximum value set M (j) is delivered to matlab by com component carries out curve fitting, obtain the curve minimum point after the match, transmit back again VC++, namely draw optimal segmenting threshold T.

After drawing optimal segmenting threshold T, utilize binary conversion treatment $f(i,j)=\left\{\begin{array}{cc}{N}_h& f(i,j)>T\\ 0& f(i,j)<T\end{array}\right.$ Realize Threshold segmentation, wherein N _hBe generally the image maximum gray scale.

Described Threshold segmentation module can also adopt maximum variance between clusters to obtain optimal segmenting threshold T.Maximum variance between clusters is to derive out on the basis of the principle of least square, and it is as follows that its threshold value is asked for process:

A, at first find out high grade grey level L in the image;

B, then get respectively from each gray level of 0 to L as threshold value th, calculate this threshold value and separate two class C ₀, C ₁Probability w separately ₀, w ₁And average value mu ₀, μ ₁If the gradation of image value is the pixel count of i is n _i, then total pixel number is: , the probability of each gray-scale value is: p _i=n _i/ N.

C ₀The probability of group is:

C ₁The probability of group is:

C ₀The mean value of group is:

C ₁The mean value of group is:

C, the total average gray of computed image are: μ=ω ₀μ ₀+ ω ₁μ ₁, the variance of calculating between two classes is: σ ²=ω ₀(μ ₀-μ) ²+ ω ₁(μ ₁-μ) ²

D, the variance of finding out between two classes are peaked threshold value T, i.e. σ ²(T)=max (σ ²(th)).

After threshold calculations is finished, utilize binary conversion treatment $f(i,j)=\left\{\begin{array}{cc}{N}_h& f(i,j)>T\\ 0& f(i,j)<T\end{array}\right.$ Realize Threshold segmentation.

The edge line that obtains image after the Threshold segmentation is thicker, for the refinement edge line, the second edge extraction module has adopted a kind of very simple effective method, and namely the method identical with edge extracting module carried out edge extracting again to the image after the Threshold segmentation.

Through behind the second edge extraction, although the edge of the image that draws almost is single pixel, simultaneously subsidiary " burr ".In order to remove these " burrs ", then take edge thinning to process.The kind of edge thinning algorithm is a lot, can be divided into according to the order of refinement: serial refinement, parallel thinning and mixing refinement.The thinning method that the edge thinning module adopts among the present invention belongs to the serial refinement, and its principle is a plurality of elimination templates of structure, and the edge is connected image afterwards and eliminates template relatively, determines whether to delete certain point.It is thorough that the method that the present invention adopts not only can refinement, and the single pixel line after the refinement is at the center line of grating grid edge line, and the Glabrous thorn, can guarantee the accuracy of next step calculating.

The edge thinning module adopts a plurality of 4 * 3 to eliminate template among the present invention, and is as follows:

$\left[\begin{array}{ccc}{P}_1& P_2& P_3\\ P_4& P_5& P_6\\ P_7& P_8& P_9\\ P_{10}& P_{11}& P_{12}\end{array}\right]$

By being compared, image and this elimination template decide current point whether deleted.The condition that this elimination template is satisfied is respectively:

A, suppose that pixel value in the image behind the second edge extraction is 0 usefulness 0 representative, pixel value is N _hUsefulness 1 representative.P ₅Eight neighborhoods in 1 number between 2 to 6, i.e. N=P ₁+ P ₂+ P ₃+ P ₄+ P ₆+ P ₇+ P ₈, 2≤N≤6;

B, P ₂And P ₈Have at least one to be zero, i.e. P ₂* P ₈=0.

C, P ₅Eight neighborhood elements circulate in the direction of the clock or counterclockwise circulation only has one 0,1 discontinuous point.

D, P ₄, P ₆And P ₈In have at least one to be zero, i.e. P ₄* P ₆* P ₈=0; Perhaps at P ₄, P ₅, P ₆, P ₉, P ₁₂, P ₁₁, P ₁₀, P ₇Eight elements circulate in the direction of the clock or counterclockwise circulation does not have 0,1 discontinuous point or has greater than 10,1 discontinuous point.

Begin search from the binary image top left corner pixel, if current pixel value is 0, then skip, if current pixel value is N _h, then make this point corresponding to P ₅Compare with eliminating template, if eliminate identical then this point deletion of template (even this pixel value sets to 0) with one of them, otherwise keep.Repeat said process, until the neither one pixel value is changed, edge thinning finishes.

Elaborate below in conjunction with the refinement of several example edge.

If a certain 4 * 3 pixel gray matrixs are on the image:

$\left[\begin{array}{ccc}{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">P</figure-callout>}}_1}^{\&prime;}& {{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">P</figure-callout>}}_2}^{\&prime;}& {{\mathrm{<figure-callout\; id="3"\; label="P"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">P</figure-callout>}}_3}^{\&prime;}\\ {P_4}^{\&prime;}& {{\mathrm{<figure-callout\; id="5"\; label="P"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">P</figure-callout>}}_5}^{\&prime;}& {{\mathrm{<figure-callout\; id="6"\; label="P"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">P</figure-callout>}}_6}^{\&prime;}\\ {P_7}^{\&prime;}& {P_8}^{\&prime;}& {P_9}^{\&prime;}\\ {P_{10}}^{\&prime;}& {P_{11}}^{\&prime;}& {P_{12}}^{\&prime;}\end{array}\right]$

P ＇ wherein ₅Be current pixel point.

$\left(1\right),\left[\begin{array}{ccc}1& 1& 1\\ 0& 1& 0\\ 0& 0& 0\\ 1& 1& 1\end{array}\right]$ $\left(2\right),\left[\begin{array}{ccc}1& 0& 1\\ 1& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$ $\left(3\right),\left[\begin{array}{ccc}1& 1& 1\\ 1& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$ $\left(5\right),\left[\begin{array}{ccc}1& 1& 1\\ 0& 1& 0\\ 0& 1& 0\\ 0& 1& 0\end{array}\right]$

$\left(5\right),\left[\begin{array}{ccc}1& 1& 1\\ 0& 1& 0\\ 0& 1& 0\\ 0& 1& 0\end{array}\right]$ $\left(6\right),\left[\begin{array}{ccc}0& 0& 0\\ 1& 1& 1\\ 1& 1& 1\\ 1& 0& 1\end{array}\right]$ $\left(7\right),\left[\begin{array}{ccc}1& 0& 0\\ 1& 1& 1\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$ $\left(8\right),\left[\begin{array}{ccc}1& 1& 1\\ 0& 1& 0\\ 0& 1& 0\\ 0& 0& 0\end{array}\right]$

Suppose that element in above-mentioned eight matrixes is corresponding to the pixel of eight parts on the image.Must satisfy simultaneously above-mentioned four conditions owing to eliminate template, thereby just can determine that this matrix is identical with one of them elimination template as long as the element in the matrix satisfies above-mentioned four conditions simultaneously, namely corresponding to P ₅Pixel be " burr ", then this pixel gray-scale value is set to 0.

Matrix (1): current some P ₅' eight neighborhoods in 1 number be 3; P ₈' be zero;

P ₅' eight neighborhood elements circulate in the direction of the clock only have one 0,1 discontinuous point; P ₄', P ₆' and P ₈' all be zero, i.e. P ₄* P ₆* P ₈=0.Satisfy simultaneously four conditions, P ₅' be burr, so deletion.

Matrix (2): P ₅' eight neighborhood elements circulated in the direction of the clock two 0,1 discontinuous point, c does not satisfy condition.P ₅' the some reservation.

Matrix (3) b that do not satisfy condition, P ₅' the some reservation.

Matrix (4): satisfy simultaneously four conditions, P ₅' deletion.

P in the matrix (5) ₅' neighborhood has two discontinuous points, the c that do not satisfy condition, P ₅' the some reservation.

Matrix (6): P ₈' neighborhood only has 0,1 discontinuous point, the d that do not satisfy condition, P ₅' the some reservation.

Matrix (7): P ₈' neighborhood has two 0,1 discontinuous points, satisfies simultaneously four conditions, P ₅' deletion.

Matrix (8): P ₈' neighborhood do not have 0,1 discontinuous point, satisfies simultaneously four conditions, P ₅' deletion.

Referring to Fig. 2, Fig. 3, scale grating grid detection system of the present invention comprises VTOL (vertical take off and landing) platform 1, be fixed in the high definition 1394 interface CCD cameras 2 on the VTOL (vertical take off and landing) platform 1, the equipment that is connected to CCD camera 2 rear ends has image pick-up card 3, computing machine 4, be connected to the magnification at high multiple camera lens 5 on the CCD camera 2, be installed in the coaxial light source 6 on the magnification at high multiple camera lens 5; Comprise Image Edge Detector and grid line width computing module in the described computing machine 4.The directional light that coaxial light source 6 sends out impinges upon the grid line district of the scale grating 8 on the horizontal location platform 7 from the lens barrel of magnification at high multiple camera lens 5.By CCD collected by camera grating grid area image, extract accurately by the edge line of Image Edge Detector with grid line, then calculated the pitch of grid line by grid line width computing module.

It is 20um that the present invention adopts pitch, black and white is selected to be of a size of 2/3 ' than being the scale grating 8 of 11:9 ' the high definition CCD camera 2 of (pixel dimension is 6.45um*6.45um), the collocation enlargement factor is 8 times magnification at high multiple camera lens 5, when its object distance was 87mm, the visual field reached 1mm.

As shown in Figure 3, scale grating grid precision detection system functional module of the present invention comprises image filtering module, edge extracting module, Threshold segmentation module, second edge extraction module and an edge thinning module, grid line width computing module; Concrete testing process is as follows: (1) carries out filtering to the image that collects; (2) image after the filtering is carried out the edge extracting first time; (3) edge extracting first time image is afterwards carried out Threshold segmentation; (4) image after the Threshold segmentation is carried out edge extracting again; (5) image after edge extracts carries out edge thinning; (6) image after the edge refinement calculates.

Each step is specific as follows:

Step (1): the image that collects is carried out filtering

The original image in the scale grating grid district that CCD camera 2 collects as shown in Figure 5.Image gather and the process of transmission in, the various noises that tend to mix cause the Quality Down of image, this edge extracting for image has caused very large difficulty, in order better to carry out edge extracting, must carry out first filtering, the image filtering module adopts preferably gaussian filtering of denoising effect.The discrete Gaussian function expression formula of two dimension zero-mean is:

$g[i,j]=e^{-\frac{({\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2)}{2{\mathrm{\&sigma;}}^2}}$

Gather image and Gaussian function convolution and obtain image after the filtering.

Step (2): the image after the filtering is carried out edge extracting

One time the edge extracting module adopts a kind of new double trapezoid arithmetic operators.This operator can extract the edge of image clear, accurately, and the contrast at edge is very high, is conducive to follow-up Threshold segmentation.The convolution mask of this operator is as follows:

$G_x=\left[\begin{array}{ccc}& m_7& \\ m_1& & m_4\\ m_2& m_8& m_5\\ m_3& & m_6\\ & m_9& \end{array}\right]$ $G_y=\left[\begin{array}{ccccc}& m_1& m_2& m_3& \\ m_7& & m_8& & m_9\\ & m_4& m_5& {\mathrm{<figure-callout\; id="6"\; label="m"\; filenames="HDA00002336124600012.png"\; state="\{\{state\}\}">m</figure-callout>}}_6& \end{array}\right]$

Wherein: m ₁~ m ₉All be integer, G _xThe arithmetic operators of horizontal direction, G _yIt is the arithmetic operators of vertical direction.m ₁=m ₃=-m ₄=-m ₆, m ₂=-m ₅, m ₂〉=2 * m ₁, m ₅〉=2 * m ₄, m ₇=m ₉＜0, m ₈=-(m ₇+ m ₉=)=-2 * m ₇=-2 * m ₉, | m ₇|≤| m ₁|; If the minor increment in the image between the adjacent two edges is d, then when the pixel of d＜20, | m ₁|=1 or 2,2≤| m ₂|≤5; When the pixel of d 〉=20,2|m ₁|≤5≤, 5≤| m ₂|≤10.

In grating grid accuracy detection of the present invention, edge extracting module of Image Edge Detector and second edge extraction module can only adopt the arithmetic operators G of horizontal direction _xObtain a width of cloth new images with filtered image as convolution algorithm.Extract the horizontal direction arithmetic operators G of raster image informal voucher line left hand edge _XzWith the horizontal direction arithmetic operators G that extracts raster image informal voucher line right hand edge _XyConcrete selection is as follows:

$G_{\mathrm{xy}}=\left[\begin{array}{ccc}& -1& \\ 2& & -2\\ 8& 2& -8\\ 2& & -2\\ & -1& \end{array}\right]$ $G_{\mathrm{xz}}=\left[\begin{array}{ccc}& -1& \\ -2& & 2\\ -8& 2& 8\\ -2& & -2\\ & -1& \end{array}\right]$

Step (3): edge extracting first time image is afterwards carried out Threshold segmentation

Although the image contrast after the edge extracting is higher, there is larger transitional region between background and the edge line, want to be partitioned into accurately edge line, to such an extent as to make too slightly too carefully fracture of edge line, must select suitable threshold value.The Threshold segmentation module can adopt the Research on threshold selection based on the histogram envelope line, also can adopt maximum variance between clusters or other threshold segmentation methods.Basic thought based on the Research on threshold selection of histogram envelope line is that the histogrammic envelope of image statistics is fitted to a smooth curve, finds the minimum point of smooth curve to be the optimal segmenting threshold of background and edge line.The Threshold segmentation module at first adopts the method for differential to obtain local maximum, then these local maximums is fitted to smooth curve again.When carrying out curve fitting, directly do not realize the least square fitting process by VC++, computation process is very complicated like this, and calculated amount is very large.But adopt the matlab programming, and realize by the com component of VC++ and matlab hybrid programming, greatly reduced calculated amount.The method is very simple, and it is accurate to ask for threshold value for the larger situation of background and target contrast.Referring to Fig. 4, the specific implementation process is as follows:

A, at first draw the statistic histogram of the new images that edge extracting module obtains, P in the statistic histogram (i) is the pixel number of i for gray level on the image;

B, statistic histogram is carried out smothing filtering, that is:

$\begin{array}{c}P\left(i\right)=(P(i-u)+P(i-u+1)+......P(i-2)+P(i-1)+P\left(i\right)+P(i+1)+P(i+2)+......\\ +P(i+u))/(2u+1)\end{array}$

Wherein u is natural number, and its numerical value determines that according to actual needs the larger filter effect of u is better, but also corresponding increase of calculated amount.General selection 1≤u≤() selected u=2 among the present invention.

C, the employing single order differential method are obtained the local maximum value set M (j) of the statistic histogram after the filtering, and wherein j is gradation of image value corresponding to Local modulus maxima.

D, local maximum value set M (j) is delivered to matlab by com component carries out curve fitting, obtain the curve minimum point after the match, transmit back again VC++, namely draw optimal segmenting threshold T.

After threshold calculations is finished, utilize binary conversion treatment $f(i,j)=\left\{\begin{array}{cc}{N}_h& f(i,j)>T\\ 0& f(i,j)<T\end{array}\right.$ Realize Threshold segmentation.N wherein _hBe generally the image maximum gray scale.

Step (4): the image after the Threshold segmentation is carried out edge extracting again

The edge line that obtains image after the Threshold segmentation is thicker, and for the refinement edge line, the second edge extraction module has adopted a kind of very simple effective method, namely again carries out the edge extracting of above-mentioned steps (2) and processes.

Step (5): the image after edge connects carries out edge thinning

Through after the above-mentioned steps, although the edge of the image that draws almost is single pixel, simultaneously subsidiary " burr ".In order to remove these " burrs ", then take edge thinning to process.The kind of edge thinning algorithm is a lot, can be divided into according to the order of refinement: serial refinement, parallel thinning and mixing refinement.The thinning method that the present invention adopts belongs to the serial refinement, and its principle is a plurality of elimination templates of structure, and the image after the edge is connected and template relatively determine whether to delete certain point.It is thorough that the method that the present invention adopts not only can refinement, and the single pixel line after the refinement is at the center line of grating grid edge line, and the Glabrous thorn, has guaranteed the accuracy of next step calculating.

The method adopts 4*3 to eliminate template, and is as follows, when image is compared with the elimination template, eliminates the current point in the P5 correspondence image in the template.This elimination template need meet some requirements.The condition that this template satisfied is respectively:

$\left[\begin{array}{ccc}{P}_1& P_2& P_3\\ P_4& P_5& P_6\\ P_7& P_8& P_9\\ P_{10}& P_{11}& P_{12}\end{array}\right]$

The 4*3 formwork structure

A, suppose that pixel value in the image behind the second edge extraction is 0 usefulness 0 representative, pixel value is N _hUsefulness 1 representative.1 number is between 2 to 6 in eight neighborhoods of P5, i.e. N=P ₁+ P ₂+ P ₃+ P ₄+ P ₆+ P ₇+ P ₈, 2≤N≤6;

B, P2 and P8 have at least one to be zero, i.e. P2 * P8=0.

The eight neighborhood elements of c, P5 circulate in the direction of the clock or counterclockwise circulation only has one 0,1 discontinuous point.

Have at least one to be zero among d, P4, P6 and the P8, i.e. P4 * P6 * P8=0; Perhaps circulate in the direction of the clock or counterclockwise circulation does not have 0,1 discontinuous point or has greater than 10,1 discontinuous point at P4, P5, P6, P9, P12, P11, eight elements of P10, P7.

Begin search from the binary image top left corner pixel, if current pixel value is 0, then skip, if current pixel value is N _h, then should compare with eliminating template corresponding to P5 by point, if identical with one of them elimination template, then this point deletion is about to this pixel value and sets to 0, otherwise keeps.Repeat said process, until the neither one pixel value is changed, edge thinning finishes.

Step (6): the image after the edge refinement calculates

Finally extract accurately the edge line of grating grid by the image processing method of above-mentioned five steps, with image after the original image stack as shown in Figure 6.The image size of the CCD collected by camera that the present invention adopts is 1392 pixel *, 1040 pixels, comprises several grating grids in the image, and every grid line evenly is divided into 20 parts, asks the grid line width of every portion.Then 20 grid line width are removed 5 maximal values, remove 5 minimum value, 10 values of centre are averaging the width that namely gets each bar grid line.All grid line width of entire image are removed 20 maximal values, remove 20 minimum value, then average, draw the average grating grid width of this image.The average pitch of scale grating that gets among Fig. 5 by above-mentioned computation process is 19.35um.

Claims (8)

1. Image Edge Detector is characterized in that comprising:

The image filtering module: filtering image noise obtains filtered image;

An edge extracting module: utilize the double trapezoid arithmetic operators to extract the edge of image, the convolution mask of this operator is as follows:

$G_x=\left[\begin{array}{ccc}& m_7& \\ m_1& & m_4\\ m_2& m_8& m_5\\ m_3& & m_6\\ & m_9& \end{array}\right]$ $G_y=\left[\begin{array}{ccccc}& m_1& m_2& m_3& \\ m_7& & m_8& & m_9\\ & m_4& m_5& m_6& \end{array}\right]$

Wherein: m ₁~ m ₉All be integer, G _xThe arithmetic operators of horizontal direction, G _yThe arithmetic operators of vertical direction, m ₁=m ₃=-m ₄=-m ₆, m ₂=-m ₅, m ₂〉=2 * m ₁, m ₅〉=2 * m ₄, m ₇=m ₉＜0, m ₈=-(m ₇+ m ₉=)=-2 * m ₇=-2 * m ₉, | m ₇|≤| m ₁|; If the minor increment in the image between the adjacent two edges is d, then when the pixel of d＜20, | m ₁|=1 or 2,2≤| m ₂|≤5; When the pixel of d 〉=20,2|m ₁|≤5≤, 5≤| m ₂|≤10;

Filtered image and convolution mask are obtained a width of cloth new images as convolution algorithm;

Threshold segmentation module: select the optimal segmenting threshold T of the new images that edge extracting module obtains, then new images is made binary conversion treatment, will put maximum gradation value N greater than the pixel of optimal segmenting threshold T _h, will set to 0 less than the pixel gray scale of optimal segmenting threshold T, thereby obtain a width of cloth binary image;

Second edge extraction module: utilize the method identical with edge extracting module that Threshold segmentation image is afterwards carried out edge extracting again;

The edge thinning module: the edge that the second edge extraction module is extracted carries out refinement, removes burr.

2. Image Edge Detector according to claim 1, it is characterized in that described Threshold segmentation module calculating new images statistic histogram, the statistic histogram envelope is fitted to a smooth curve, then with the minimum point of the smooth curve that finds as a setting with the optimal segmenting threshold T of edge line.

3. Image Edge Detector according to claim 2 is characterized in that described Threshold segmentation module carries out smothing filtering with the new images statistic histogram, adopts the single order differential method to obtain the local maximum value set of the statistic histogram after the filtering; Utilize the local maximum value set to carry out curve fitting, obtain the curve minimum point after the match, and the gray-scale value that this curve minimum point is corresponding is as optimal segmenting threshold T.

4. Image Edge Detector according to claim 3 is characterized in that described Threshold segmentation module utilizes the matlab programming to carry out curve fitting.

5. Image Edge Detector according to claim 1 is characterized in that described edge thinning module stores has a plurality of 4 * 3 to eliminate template, and 4 * 3 elimination templates are as follows:

$\left[\begin{array}{ccc}{P}_1& P_2& P_3\\ P_4& P_5& P_6\\ P_7& P_8& P_9\\ P_{10}& P_{11}& P_{12}\end{array}\right]$

Eliminate template and satisfy simultaneously following four conditions:

A, P ₅Eight neighborhood elements in 2 ~ 6 elements are arranged is 1, all the other elements are 0, i.e. N=P ₁+ P ₂+ P ₃+ P ₄+ P ₆+ P ₇+ P ₈, 2≤N≤6;

B, P ₂And P ₈Have at least one to be zero, i.e. P ₂* P ₈=0;

C, P ₅Eight neighborhood elements circulate in the direction of the clock or by counterclockwise circulation only have one 0,1 discontinuous point;

D, P ₄, P ₆And P ₈In have at least one to be zero, i.e. P ₄* P ₆* P ₈=0; Perhaps P ₈Eight neighborhood elements circulate in the direction of the clock or do not have 0,1 discontinuous point or have greater than 10,1 discontinuous point by counterclockwise circulation;

Begin search from the binary image top left corner pixel, if the current pixel gray-scale value is 0, then skip; If the current pixel gray-scale value is N _h, then make this pixel corresponding to the element P that eliminates template ₅, with other pixels around this pixel with eliminate on the template correspondence position element and compare, template is identical then to be set to 0 the current pixel gray scale if eliminate with one of them, otherwise the current pixel gray-scale value is constant; Repeat said process, until the neither one grey scale pixel value is changed in the binary image, edge thinning finishes.

6. scale grating grid precision detection system that comprises such as Image Edge Detector as described in the arbitrary claim of claim 1 ~ 5, characterized by further comprising VTOL (vertical take off and landing) platform (1), be fixed in the CCD camera (2) on the VTOL (vertical take off and landing) platform (1), be connected to the enlarging lens (5) on the CCD camera (2), be installed in the coaxial light source (6) on the enlarging lens (5), image pick-up card (3), grid line width computing module; The directional light that coaxial light source (6) sends out impinges upon the grid line district of scale grating (8) from the lens barrel of enlarging lens (5); CCD camera (2) gathers the grating grid area image and sends image pick-up card (3) to; Image pick-up card (3) with the raster image data transmission that gathers to Image Edge Detector, Image Edge Detector is extracted the grid line of raster image, every grid line evenly is divided into M part in the image that grid line width computing module detects Image Edge Detector, tries to achieve the grid line width of every portion; Then M grid line width removed N maximal value, remove N minimum value, M-2N value of centre is averaging the width that obtains each bar grid line; All grid line width of entire image are removed Q maximal value, remove Q minimum value, then average, draw the average grating grid width of this image; M wherein〉2N, the grid line number〉2Q.

7. scale grating grid precision detection system according to claim 6 is characterized in that edge extracting module of described Image Edge Detector and the arithmetic operators G that the second edge extraction module adopts horizontal direction _xObtain a width of cloth new images with filtered image as convolution algorithm.

8. scale grating grid precision detection system according to claim 7 is characterized in that extracting the horizontal direction arithmetic operators G of raster image informal voucher line left hand edge _XzWith the horizontal direction arithmetic operators G that extracts raster image informal voucher line right hand edge _XyAs follows:

$G_{\mathrm{xz}}=\left[\begin{array}{ccc}& -1& \\ -2& & 2\\ -8& 2& 8\\ -2& & -2\\ & -1& \end{array}\right]$ $G_{\mathrm{xy}}=\left[\begin{array}{ccc}& -1& \\ 2& & -2\\ 8& 2& -8\\ 2& & -2\\ & -1& \end{array}\right]$

Images