CN103673872A

CN103673872A - Measurement method and measurement system of liquid drop volume

Info

Publication number: CN103673872A
Application number: CN201210333220.7A
Authority: CN
Inventors: 于强; 童方圆
Original assignee: National Space Science Center of CAS
Current assignee: National Space Science Center of CAS
Priority date: 2012-09-10
Filing date: 2012-09-10
Publication date: 2014-03-26

Abstract

The present invention relates to a method and system for measuring the volume of a droplet, wherein the method comprises: an image smoothing step for reducing noise; a step for calculating gradient magnitude and direction; a non-maximum suppression step for adjusting the amplitude The ridge band in the value image is thinned, and only the pixels with the largest local variation in amplitude are retained; the double thresholding and edge connection steps are used to remove false edges and connect the real edges; the image thinning step is used to An image whose edge is at a single pixel level is obtained, so as to obtain the outer contour of the droplet; finally, the volume of the droplet is obtained according to the outer contour of the droplet. By adopting the method provided by the invention, the measurement accuracy of the droplet volume is improved.

Description

The measuring method of droplet size and measuring system

Technical field

The present invention relates to fields of measurement, especially relate to a kind of measuring method and system of droplet size.

Background technology

The method of measuring at present drop mainly contains two kinds.Wherein a kind of is fitting process (referring to: Xu Zhi button rule Fang Cheng etc., considering the contact angle fitting algorithm [J] of droplet size, High-Voltage Technology, 6 phases in 2010), and the profile of equivalent drop with a circle, then calculates round volume.Because drop itself is not circular, the droplet size deviation in this way calculating is larger.This method is easily understood, and is easy to realize, and when liquid volume is very little, applicability is better; But for the larger situation of droplet size, not too applicable, thereby do not there is versatility.In the art, the volume that also often calculates drop according to the contour edge of drop is (referring to Song Qing, Zhang Guoxiong etc., image processing techniques is used for calculating droplet size [J], optical technology, 30 3 phases of volume in 2004): first utilize Sobel operator to carry out rim detection to gray-scale map, obtain the region that grey scale change is violent, the false profile in inside that this region comprises true profile, the interference noise of drop and causes because drop is reflective; Then, carry out thresholding processing, according to statistic histogram choose reasonable thresholding, remove interference noise; Again by the connected component labeling algorithm process to binary map, each connected region in image is made a distinction by being marked as different labels, select the region of area maximum in region, filtering, due to the false profile in reflective inside of causing and noise, has improved the robustness of system acquisition edge contour; Finally, carry out peripheral edge line drawing, and adopt the mode of similar rectangular integration to carry out volume calculated.Although this method has provided the conventional method that utilizes image processing techniques to calculate droplet size, and respond well.But, because needs are calculated a plurality of parameters, thereby the very difficult accuracy that guarantees volume result of calculation in practical application.

Summary of the invention

The technical problem to be solved in the present invention is to provide the droplet size measuring method that a kind of measuring accuracy is high.In addition, also provide a kind of droplet size measuring system.

For solving the problems of the technologies described above, the invention provides a kind of measuring method of droplet size, the method comprises:

Image smoothing step, it is for reducing noise, to obtain smoothed image f ₁(x, y);

Compute gradient amplitude and direction step;

Non-maximum value suppresses step, for the ridge band to magnitude image, carries out refinement, only retains the pixel of amplitude localized variation maximum in magnitude image;

Dual threshold is processed and edge Connection Step, for removing false edge and real edges being coupled together, obtains drop edge image;

Image thinning step, for obtaining the image that image border is single pixel scale, thereby obtains the peripheral profile of drop;

According to the peripheral profile of described drop, obtain the take pixel drop height H that is unit, the number of pixels of every a line, and cross-sectional diameter, obtain the volume of drop by following formula:

V = Σ_{i = 1}^{N} A_{i} = \frac{π}{4} Σ_{i = 1}^{N} D_{i}^{2} ΔH,

Wherein, A _irepresent cross-sectional area; Δ H refers to two height values between continuous drop xsect.

Preferably, the measuring method of droplet size according to claim 1, is characterized in that, described image smoothing step refers to carries out process of convolution by original image and Gaussian function.

Preferably, the measuring method of droplet size according to claim 1, is characterized in that, in described compute gradient amplitude and direction step, establishes the x direction of two-dimensional Gaussian function and the first order derivative of y direction as shown in formula (1):

\{\begin{matrix} \frac{&PartialD; g}{&PartialD; x} = kxexp (- \frac{x^{2}}{2 σ^{2}}) \exp (- \frac{y^{2}}{2 σ^{2}}) \\ \frac{&PartialD; g}{&PartialD; y} = kyexp (- \frac{y^{2}}{2 σ^{2}}) \exp (- \frac{x^{2}}{2 σ^{2}}) \end{matrix} - - - (1)

By described formula (1) respectively with described smoothed image f ₁(x, y) carries out convolution, obtains Grad in x direction and the Grad in y direction:

\{\begin{matrix} E_{x} (x, y) = \frac{&PartialD; g}{&PartialD; x} * f_{1} (x, y) \\ E_{y} (x, y) = \frac{&PartialD; g}{&PartialD; y} * f_{1} (x, y) \end{matrix}

Amplitude and the direction that can obtain image gradient are respectively:

\{\begin{matrix} M (x, y) = \sqrt{{E_{x}}^{2} (x, y) + {E_{y}}^{2} (x, y)} \\ θ (x, y) = \arctan [\frac{E_{y} (x, y)}{E_{x} (x, y)}] \end{matrix}

Wherein, (x, y) represents certain pixel; M (x, y) is defined as the amplitude of image gradient; θ (x, y) is defined as the direction of image gradient; E _x(x, y) represents the Grad in x direction; E _y(x, y) represents the Grad in y direction; K is constant.

Preferably, the measuring method of droplet size according to claim 1, is characterized in that, described non-maximum value suppresses step and comprises: the scope of determining the gradient direction θ (x, y) of pixel (x, y); The variation range of described gradient direction θ (x, y) is divided into 4 sectors; By described pixel (x, y) the described gradient amplitude M (x locating, y) with described gradient direction θ (x, y) gradient magnitude of two of sector, place neighbor pixels compares, if described pixel (x, y) the described gradient amplitude M (x locating, y) be less than or equal to this pixel (x, y) gradient magnitude of two of sector, gradient direction place consecutive point, described pixel (x, y) being labeled as non-marginal point, is 0 by described M (x, y) assignment; Otherwise described pixel (x, y) is labeled as candidate marginal, the value of described M (x, y) remains unchanged.

Preferably, the measuring method of droplet size according to claim 1, is characterized in that, described dual threshold process and edge Connection Step in, by the accumulation histogram of magnitude image, obtain a high threshold T _h, then obtain a low threshold value T _l=0.4T _h; Each pixel (x, y) suppressing in the image after step process through described non-maximum value is detected, if (x, y) gradient magnitude is greater than high threshold T _h, think that this pixel one is decided to be marginal point; If the gradient magnitude of pixel (x, y) is less than low threshold value T _l, think that this pixel is marginal point scarcely; Pixel for gradient magnitude between two threshold values, sees in eight adjacent pixels points of this pixel whether be greater than the pixel of high threshold, if had, this pixel is edge so, otherwise is not just edge; Finally, the real edges after dual threshold processing is coupled together.

Preferably, the measuring method of droplet size according to claim 1, is characterized in that, described image thinning step refers to the thinning method based on concordance list, whether it looks into described concordance list according to the situation of wanting eight neighborhoods of refinement pixel, decide this pixel should delete.

In addition, the present invention also provides a kind of droplet size measuring system, and it comprises CCD camera, computing machine with camera lens; Described computing machine is carried out said method.

Compared with prior art, the present invention has the following advantages:

The present invention combines the feature that image is processed, adopt image smoothing step, compute gradient amplitude and direction step, non-maximum value to suppress step, dual threshold processing and edge Connection Step and image thinning step, can rely on image processing means to obtain the accurate numerical value of droplet size.

Accompanying drawing explanation

Fig. 1 is for processing input picture and the final schematic diagram that obtains image border;

Fig. 2 is that schematic diagram is divided in gradient direction sector;

Fig. 3 is the schematic diagram of image thinning;

Fig. 4 is for being obtained the schematic diagram of the step of liquid profile by drop image;

Fig. 5 is the volume of drop and the integration schematic diagram of surface area

Fig. 6 is droplet size measuring system schematic diagram.

Embodiment

Below in conjunction with concrete drawings and Examples, the present invention is described in detail.

The droplet size measuring method of the embodiment of the present invention detects drop image border based on Canny criterion, then according to the drop edge (being also drop profile) obtaining, finally obtains the volume of drop.Wherein, Canny criterion is three criterions that a best edge detection operator of Canny proposition should meet, also: want accurately and single edges response criteria good testing result, the location of edge.Canny edge detection operator is the multistage edge detection algorithm that John F.Canny developed in 1986.

Fig. 1 has shown that the present invention processes input picture and the final schematic diagram that obtains image border.Comprising image smoothing step, compute gradient amplitude and direction step, non-maximum value, suppress step, dual threshold processing and edge Connection Step and image thinning step.For each step, be described in detail below.

image smoothing step

The object of picture smooth treatment is that the original image of input is processed to reduce noise, to be subject to during computed image gradient noise minimum.Smooth function is used Gaussian function, and its form is as follows:

g (x, y) = \frac{1}{2 {πσ}^{2}} \exp (- \frac{x^{2} + y^{2}}{2 σ^{2}})

Wherein, the mean square deviation that σ is Gaussian function, σ gets 1 in embodiments of the present invention.

If original image is f (x, y), smoothed image f ₁(x, y) represents, is the convolution of original image and Gaussian function, that is:

f ₁(x,y)＝f(x,y)*g(x,y)

Due to symmetry and the decomposability of Gaussian function, g (x, y) can be decomposed into two one dimension Gaussian functions of x direction and y direction.The convolution of original image and two-dimensional Gaussian function can be simplified with following methods, and original image first carries out convolution with the one dimension Gaussian function of x direction, and then carries out convolution with the one dimension Gaussian function of y direction.

the amplitude of computed image gradient and the step of direction

Because two-dimensional Gaussian function has symmetry and decomposability, so can obtain by calculating the convolution of the directional derivative of Gaussian function in either direction and image amplitude and the direction of image gradient.

Adopt ranks wave filter the first order derivative of Gaussian function to be decomposed into the ranks wave filter of two one dimensions, can improve arithmetic speed like this.The x direction of two-dimensional Gaussian function and the first order derivative of y direction are as shown in formula (1):

\{\begin{matrix} \frac{&PartialD; g}{&PartialD; x} = kx exp (- \frac{x^{2}}{2 σ^{2}}) \exp (- \frac{y^{2}}{2 σ^{2}}) \\ \frac{&PartialD; g}{&PartialD; y} = ky exp (- \frac{y^{2}}{2 σ^{2}}) \exp (- \frac{x^{2}}{2 σ^{2}}) \end{matrix} - - - (1)

By formula (1) respectively with f ₁(x, y) carries out convolution, obtains Grad in x direction and the Grad in y direction:

\{\begin{matrix} E_{x} (x, y) = \frac{&PartialD; g}{&PartialD; x} * f_{1} (x, y) \\ E_{y} (x, y) = \frac{&PartialD; g}{&PartialD; y} * f_{1} (x, y) \end{matrix}

Amplitude and the direction that can obtain image gradient are respectively:

\{\begin{matrix} M (x, y) = \sqrt{{E_{x}}^{2} (x, y) + {E_{y}}^{2} (x, y)} \\ θ (x, y) = \arctan [\frac{E_{y} (x, y)}{E_{x} (x, y)}] \end{matrix}

Wherein, M (x, y) is defined as the amplitude of image gradient, namely the amplitude of the gradient of original image f (x, y); θ (x, y) is defined as the direction of image gradient, namely the direction of the gradient of original image f (x, y), the i.e. deflection of pixel; E _x(x, y) represents the Grad in x direction; E _y(x, y) represents the Grad in y direction; K is constant.

non-maximum value suppresses step

M (x, y) value shows that more greatly corresponding image gradient value is larger, but this is not enough to determine pixel (x, y) whether be marginal point, in order determining more accurately, must to carry out refinement to the ridge band in magnitude image by marginal point, only to retain the pixel of amplitude localized variation maximum.Wherein, magnitude image (that is: gradient amplitude image) is exactly M (x, y) image.Ridge band refers to the region, image border of similar ridge-shaped.Refinement is to realize by the amplitude of image gradient being carried out to non-maximum value inhibition processing.Non-maximum value suppresses to obtain refinement edge by suppressing the gradient magnitude of all non-ridge peak values on gradient direction.Non-maximum value Restrainable algorithms is as follows:

For each pixel (x, y), first determine the scope of this deflection θ (x, y).The variation range one of deflection is divided into 4 sectors, as shown in Figure 2.

These four sector numbers are 0,1,2,3.By described pixel (x, y) the gradient amplitude M (x, y) locating compares with the gradient magnitude of two neighbor pixels of sector, its gradient direction θ (x, y) place, if described pixel (x, y) the gradient amplitude M (x, y) locating is less than or equal to the gradient magnitude of two consecutive point of this sector, pixel (x, y) gradient direction place, described pixel (x, y) be labeled as non-marginal point, M (x, y) assignment is 0; Otherwise described pixel (x, y) is labeled as candidate marginal, the value of M (x, y) remains unchanged.Result after non-maximum value suppresses processing is designated as N (x, y).Real edge is just included in that in N (x, y), those are worth in non-vanishing pixel, has also comprised a large amount of false edges in these points simultaneously.

dual threshold is processed and edge Connection Step

The object that dual threshold is processed is to remove false edge.In order to judge false edge and real edges, N (x, y) is carried out to thresholding processing, i.e. a given threshold value, is real marginal point higher than the pixel of threshold value, lower than the pixel of threshold value, is false edge.But owing to using a fixing threshold value, threshold value is too high or too low all affects final result of determination.For addressing this problem, the dual threshold method that embodiments of the invention adopt Canny to propose, that is: the accumulation histogram by magnitude image obtains a high threshold T _h, then obtain a low threshold value T _l=0.4T _h.Wherein, accumulation histogram representative image constituent is in the accumulated probability distribution situation of gray level, and each probable value representative is less than or equal to the probability of this gray-scale value; Accumulation histogram is prior art, is also a kind of very basic method in graphical analysis.First dual threshold facture detects each pixel (x, y) in N (x, y), if (x, y) gradient magnitude is greater than high threshold T _h, think that this pixel one is decided to be marginal point; If the gradient magnitude of pixel (x, y) is less than low threshold value T _l, think that this pixel is marginal point scarcely; Pixel for gradient magnitude between two threshold values, sees in eight adjacent pixels points of this pixel whether be greater than the pixel of high threshold, if had, this pixel is edge so, otherwise is not just edge.Finally, the real edges after dual threshold processing is coupled together.

image thinning step

Through above-mentioned treatment step, the image obtaining is that (what pixel value was 1 is the part that needs refinement to bianry image, and what pixel value was 0 is background area.), can obtain a reasonable edge detection results, but export edge, be not single pixel scale, but have certain width.The object of image thinning is to obtain the image that image border is single pixel scale.In order to obtain the edge detection results of single pixel scale, continue the edge of output to carry out image thinning operation, can obtain the edge of extraordinary single pixel scale.A simple image thinning operation as shown in Figure 3.

Fig. 4 has shown the schematic diagram that is obtained the step of liquid profile by drop image.Embodiment provided by the invention adopts the thinning algorithm based on concordance list to process through Canny operator and processes the drop edge image (being profile matching) obtaining.So-called refinement is exactly through peeling off from level to level, is keeping under the prerequisite of image original form, adopts thinning algorithm from original image, to remove some points, until obtain the skeleton of image.Thinning algorithm meets convergence requirement, thereby can guarantee the connectedness of fine rule after refinement, and keeps the basic configuration of former figure, and can also reduce the distortion of stroke intersection; Result after refinement is exactly the center line that has retained original image; Rapidity and the iterations of refinement are few.

Whether the thinning algorithm based on concordance list, according to certain basis for estimation, is made a concordance list, then according to the situation of wanting eight neighborhoods of refinement pixel, looks into concordance list, decide this pixel should delete.Described basis for estimation is described as: (1) internal point can not be deleted; (2) isolated point can not be deleted; (3) straight line end points can not be deleted; (4) if certain some P is frontier point, remove after P, if connected component does not increase, P can delete.Due to 8 neighborhoods of a pixel have in 256 may situation, the size of concordance list is generally 256, in table, element value is set to 1, is illustrated under this kind of neighborhood value condition, pixel should be deleted; Otherwise, should retain.Algorithm based on concordance list need to not judge computing when algorithm is carried out, and only need to table look-up, therefore fast than other algorithms.

Obtain, after the peripheral profile of drop, can solving the volume that obtains drop, as shown in Figure 5, by the volume of each integral unit in cumulative figure, can obtain volume and the surface area of drop.

The drop profile being obtained by profile matching, the height that the poor Δ H(of drop height that to be easy to obtain to take pixel be unit is also profile, and cross-sectional diameter D above most and descend pixel ordinate value to subtract each other most of profile) and the number of pixels of every a line (profile be expert at the rightest and the most left pixel abscissa value subtract each other), _i, i=1,2 ..., N.Thereby can draw cross-sectional area A _i:

A_{i} = \frac{π}{4} Σ_{i = 1}^{N} D_{i}^{2} ΔH

According to the method for numerical integration, cross-sectional area A can superpose _iobtain the volume of drop:

V = Σ_{i = 1}^{N} A_{i} = \frac{π}{4} Σ_{i = 1}^{N} D_{i}^{2} ΔH,

Wherein, Δ H refers to two height values between continuous drop xsect.In actual computation, can directly use 1 replacement, represent to calculate an integrated value at interval of a pixels tall.

What Fig. 5 showed is droplet size measuring system.With the CCD camera of camera lens, catch the image of drop, the numerical information of this image is passed to computing machine.Computing machine utilizes said method to process image, draws accurate droplet size.

checking

Adopt the steel ball of diameter 5mm to replace drop to carry out emulation experiment, outcome measurement error is 0.13%.

It should be noted last that, above embodiment is only unrestricted in order to technical scheme of the present invention to be described.Although the present invention is had been described in detail with reference to embodiment, those of ordinary skill in the art is to be understood that, technical scheme of the present invention is modified or is equal to replacement, do not depart from the spirit and scope of technical solution of the present invention, it all should be encompassed in the middle of claim scope of the present invention.

Claims

1. A method for measuring droplet volume, characterized in that the method comprises:

an image smoothing step for reducing noise to obtain a smoothed image f ₁ (x,y);

Calculate gradient magnitude and direction steps;

The non-maximum value suppression step is used to refine the roof band in the magnitude image, and only keep the pixel points with the largest local variation of the magnitude in the magnitude image;

Double threshold processing and edge connection steps are used to remove false edges and connect real edges to obtain a droplet edge image;

The image thinning step is used to obtain an image whose edge is a single pixel level, thereby obtaining the peripheral contour of the droplet;

According to the outer contour of the droplet, the droplet height H in units of pixels, the number of pixels in each row, and the cross-sectional diameter are obtained, and the volume of the droplet is obtained by the following formula:

V V = = {Σ Σ}_{i i = = 11}^{N N} {A A}_{i i} = = \frac{π π}{44} {Σ Σ}_{i i = = 11}^{N N} {D D.}_{i i}^{22} ΔH ΔH,,

Among them, _Ai represents the cross-sectional area; ΔH refers to the height value between two consecutive droplet cross-sections.

2. The measuring method of droplet volume according to claim 1, characterized in that, said image smoothing step refers to carrying out convolution processing with original image and Gaussian function.

3. the measurement method of droplet volume according to claim 1 is characterized in that, in described calculation gradient magnitude and direction step, the x direction of two-dimensional Gaussian function and the first order derivative of y direction are as formula (1 ) as shown:

\{\begin{matrix} \frac{&PartialD; &PartialD; g g}{&PartialD; &PartialD; x x} = = kxexp kxexp ((- - \frac{{x x}^{22}}{22 {σ σ}^{22}})) exp exp ((- - \frac{{y the y}^{22}}{22 {σ σ}^{22}})) \\ \frac{&PartialD; &PartialD; g g}{&PartialD; &PartialD; y the y} = = kyexp kyexp ((- - \frac{{y the y}^{22}}{22 {σ σ}^{22}})) exp exp ((- - \frac{{x x}^{22}}{22 {σ σ}^{22}})) \end{matrix} - - - - - - ((11))

The formula (1) is respectively convolved with the smooth image f ₁ (x, y) to obtain the gradient value in the x direction and the gradient value in the y direction:

\{\begin{matrix} {E E.}_{x x} ((x x,, y the y)) = = \frac{&PartialD; &PartialD; g g}{&PartialD; &PartialD; x x} * * {f f}_{11} ((x x,, y the y)) \\ {E E.}_{y the y} ((x x,, y the y)) = = \frac{&PartialD; &PartialD; g g}{&PartialD; &PartialD; y the y} * * {f f}_{11} ((x x,, y the y)) \end{matrix}

Then the magnitude and direction of the image gradient can be obtained as follows:

\{\begin{matrix} M m ((x x,, y the y)) = = \sqrt{{E E.}_{x x}^{22} ((x x,, y the y)) + + {E E.}_{y the y}^{22} ((x x,, y the y))} \\ θ θ ((x x,, y the y)) = = arctan arctan [[\frac{{E E.}_{y the y} ((x x,, y the y))}{{E E.}_{x x} ((x x,, y the y))}]] \end{matrix}

Among them, (x, y) represents a certain pixel point; M(x, y) is defined as the magnitude of the image gradient; θ(x, y) is defined as the direction of the image gradient; E _x (x, y) represents the direction in the x direction The gradient value of ; E _y (x, y) represents the gradient value in the y direction; k is a constant.

4. The measuring method of droplet volume according to claim 1, is characterized in that, described non-maximum suppression step comprises: determine the scope of the gradient direction θ (x, y) of pixel point (x, y); The variation range of the gradient direction θ(x, y) is divided into 4 sectors; the gradient amplitude M(x, y) at the pixel point (x, y) is compared with the gradient direction θ( The gradient magnitude values of two adjacent pixel points in the sector where x, y) are located are compared, if the gradient magnitude M(x, y) at the pixel point (x, y) is less than or equal to the pixel point (x , y) the gradient magnitude values of two adjacent points in the sector where the gradient direction is located, then the pixel point (x, y) is marked as a non-edge point, and the M(x, y) is assigned a value of 0; otherwise, The pixel point (x, y) is marked as a candidate edge point, and the value of M(x, y) remains unchanged.

5. The measuring method of droplet volume according to claim 1, is characterized in that, in described two-threshold processing and edge connection step, obtains a high threshold _Th by the accumulative histogram of magnitude image, then obtains a Low threshold T _l =0.4T _h ; each pixel point (x, y) in the image processed by the non-maximum suppression step is detected, if the (x, y) gradient magnitude value is greater than the high threshold T _h , it is considered that the pixel point must be an edge point; if the gradient amplitude value of the pixel point (x, y) is less than the low threshold T _l , it is considered that the pixel point must not be an edge point; for the gradient amplitude value between the two thresholds If there is any pixel larger than the high threshold in the eight adjacent pixels of the pixel, if there is, then the pixel is an edge, otherwise it is not an edge; finally, the real edge after double thresholding connect them.

6. The measuring method of droplet volume according to claim 1, is characterized in that, described image thinning step refers to the thinning method based on index table, and it according to the situation of eight neighborhoods of desired thinning pixel point Check the index table to determine whether the pixel should be deleted.

7. A liquid droplet volume measurement system, which comprises a CCD camera with a lens, a computer; the computer executes the method according to claim 1.