CN104867117A - Flow field image preprocessing method and system thereof - Google Patents

Abstract

The invention provides a flow field image preprocessing method and a system thereof, belongs to the field of image preprocessing methods and devices, aims at the problem in the prior art that preprocessing time is relatively long is flow field measurement, and improves the flow field image preprocessing efficiency. The flow field image preprocessing method comprises the steps of equalization, image stretching, noise judgment, fixed noise removing, binary mean value image construction and mask operation. The flow field image preprocessing system comprises an input Ethernet port, an Ethernet input control chip, a digital signal processor, a synchronous dynamic random access memory, a dynamic storage chip, a live-line erasable programmable read only memory, an output Ethernet port and an Ethernet output control chip. According to the invention, the size and the power consumption are relatively small, and the flow field image preprocessing system can be embedded into an existing flow field measuring instrument to carry out flow field measured image preprocessing and output flow filed measured images relatively low in noise, so that the efficiency and the processing speed of a whole rear end processor are improved.

Description

A kind of flow field image pre-processing method and system thereof

Technical field

The invention belongs to image pre-processing method and device, be specifically related to a kind of flow field image pre-processing method and system thereof, for ethernet communication.

Background technology

The research of High Speed Flow Field has become and has more and more been widely used in engineering practice (as the maneuvering flight of aircraft, the complex appearance flowing etc. of space shuttle), multiple query is also there is in current research, mainly be that the environment residing for flow field is comparatively special, the Particles Moving speed of its inside is fast, and skewness, under the environment of laser, the uneven illumination that whole flow field can show again is even, (may there is bubble when moving in particle to add the interference of other factors extraneous, water smoke etc. may be there is in observation wall), the particle picture obtaining stable motion is made to become more difficult, and Particles Moving speed is high, need the acquisition performance of camera higher.Method relatively more conventional is at present by high speed camera acquisition stream field picture, is transferred to the enterprising row relax of PC.Owing to obtaining there is a large amount of noise (uneven illuminations in original particle picture, bubble, water smoke etc.), make to need the cost regular hour to carry out Image semantic classification work the most at first, because data volume is very large, (current flow observation camera frame per second used is at more than 100fps, resolution is more than 100w pixel), emphasizing the flow field survey field that real-time handling property is higher, its time spending on Image semantic classification is very high, the speed of whole process that what such meeting was serious have influence on, for traditional flow field survey Image semantic classification, spend in the 30%-40% that the pretreated time often occupies the whole processing time, this has had a strong impact on the raising of processing speed.

Summary of the invention

The invention provides a kind of flow field image pre-processing method, its Image semantic classification system is provided simultaneously, for the problem that pre-service Occupation time in current flow field survey is longer, improve flow field Image semantic classification efficiency.

A kind of flow field provided by the present invention image pre-processing method, for the flow field that will observe, utilizes Ethernet camera to obtain N width flow field figure picture, it is characterized in that, for any kth width flow field figure picture, comprise the steps:

(1) equalization step:

To kth width flow field figure as I _kupper all pixels are slided image window line by line, calculate the equalization value N of each pixel x in the jth image window that formed _k ^j(x):

${N_k}^j\left(x\right)=\frac{{I_k}^j\left(x\right)-{I^j}_{\min }}{{I^j}_{\max }-{I^j}_{\min }},$ k＝1、…、N，N＝1000～50000，j≥1，

0≤N _k ^j(x)≤1；

Wherein, I _k ^j(x) for kth width flow field figure is as the pixel value of pixel x in a jth window, I ^j _maxand I ^j _minbe respectively max pixel value and minimum pixel value in a jth image window:

${I^j}_{\max }=\underset{x\∈\mathrm{\Ω}}{\max }\left({I_k}^j\left(x\right)\right),{I^j}_{\min }=\underset{x\∈\mathrm{\Ω}}{\min }\left({I^j}_k\left(x\right)\right),$

Ω is the territory at image window place, and described image window size is M × M, M=8,16 or 32 pixels;

(2) image stretch step, comprises following sub-step:

(2.1) for a jth image window of kth width image, the tension values S of wherein each pixel x is calculated ^j _k(x):

${S^j}_k\left(x\right)=\frac{{N^j}_k\left(x\right)-\stackrel{\‾}{I_{k\min }}}{\stackrel{\‾}{I_{k\max }}-\stackrel{\‾}{I_{k\min }}},$

Wherein be respectively the minimum pixel value in kth width image and max pixel value;

(2.2) for kth width image, the tension values S of each pixel x ^j _kx (), by the position of its respective pixel x, forms kth width stretching image S _k;

Image equalization step and image stretch step, each window in piece image carries out;

In flow field survey, because particle is very little, flowing velocity is very fast again, therefore needs to throw light on laser, generally to wanting the particle in viewing area to throw light on, due to the problem of image capturing angle, collected by camera also exists the phenomenon of uneven illumination to particle figure sector-meeting, shows as some local brightness very high, some place is darker, carry out image stretch operation be exactly get rid of uneven illumination phenomenon, make the brightness of image evenly, be convenient to post-processed;

(3) noise determining step, comprises following sub-step:

(3.1) naked eyes judge that kth width flow field figure is as I _kon whether there is water smoke, be carry out step (4); Otherwise carry out sub-step (3.2);

(3.2) judge that kth width flow field figure is as I _kon whether there is bubble, be carry out step (5), otherwise export S _k, terminate;

(4) remove steady noise step, comprise following sub-step:

(4.1) first template image B is obtained _ref:

$B_{\mathrm{ref}}=\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{S}_k;$

The implication of above formula is, the pixel value that N width stretching image is corresponding is added, and addition result, divided by the position by its respective pixel x after N, forms template image B _ref;

(4.2) kth width stretching image S _kwith template image B _refcarry out image subtraction, obtain denoising image

${\stackrel{\‾}{S}}_k=S_k-B_{\mathrm{ref}};$

Described image subtraction refers to that the pixel value that two width images are corresponding subtracts each other, and subtracts each other the position of result by its respective pixel x, forms denoising image;

Image subtraction operation is to get rid of noise fixing in flow field, and in the experiment of carried out flow field survey, mainly observe the water smoke of locular wall, water smoke is static in the picture, and particle is motion, therefore can remove water smoke by the method for background subtracting;

(4.3) judge that kth width flow field figure is as I _kon whether there is bubble, be carry out step (5), otherwise export terminate;

(5) construct binaryzation average image step, comprise following sub-step:

(5.1) the flow field figure picture image I as a setting that does not comprise bubble is selected _nomal, by kth width flow field figure as I _kwith background image I _nomalcarry out image subtraction, obtain kth width error image

$\stackrel{\‾}{I_k}=I_k-I_{\mathrm{nomal}},$

(5.2) equalization operation is carried out to all pixels of kth width width error image, obtain the difference equalization value G of each pixel x _k(x):

$G_k\left(x\right)=\frac{\stackrel{\‾}{I_k}-\stackrel{\‾}{I_{k\min }}}{{\stackrel{\‾}{I}}_{k\max }-\stackrel{\‾}{I_{k\min }}},$ 0≤G _k(x)≤1；

Wherein, represent kth width error image minimum pixel value, represent kth width error image max pixel value;

(5.3) the difference equalization value G of each pixel x _kx (), by the position of its respective pixel x, forms kth width average image G _k;

(5.4) to G _kcarry out binaryzation operation, form kth width binary image the wherein pixel value of each pixel x

${\stackrel{\‾}{G}}_k\left(x\right)=\left\{\begin{array}{c}0,(G_k\left(x\right)\≤G_{\mathrm{kth}})\\ 1,(G_k\left(x\right)>G_{\mathrm{kth}})\end{array}\right.;$

Wherein, 0≤binary-state threshold G _kth≤ 1; The laser energy being irradiated to flow field is larger, G _kththe closer to 1, otherwise, G _kththe closer to 0;

The pixel value of such pictures non-zero namely 1, the flow field due to reality is very dark, and in the process of practical laser polishing, alveolate local bubble can obvious reflected light photograph, shows as brightness higher, middle pixel value is 1;

(6) mask operation step, comprises following sub-step:

(6.1) exist in the pixel that all pixel values are 1 is judged and is processed, formed mask images

(6.2) by kth width stretching image S _kor denoising image with mask images carry out image subtraction, obtain pretreated result images, Output rusults image, terminate; Result images eliminates the picture noises such as bubble, water smoke, uneven illumination.

Described sub-step (3.2) and step (4.3) judge that kth width flow field figure is as I _kon whether there is bubble, comprise following process:

A. naked eyes select the flow field figure picture image I as a setting that does not comprise bubble _nomal, by kth width flow field figure as I _kwith background image I _nomalcarry out image subtraction, obtain kth width error image

$\stackrel{\‾}{I_k}=I_k-I_{\mathrm{nomal}},$

B. for error image upper all pixels slip error image window line by line, for each error image window, compares each pixel value and bubble threshold value that comprise in it size, judge to be greater than in error image window pixel value whether to exceed (Q/2+0.5) individual, be judge that this error image window comprises bubble, otherwise judge that this error image window does not comprise bubble;

Described error image window size is Q × Q, Q=9,11 or 13 pixels;

it is suitable to select according to different illumination conditions when illumination is stronger time, value near 255, when illumination is more weak, value near 0, generally value is 100;

C. error image any one error image window upper comprises bubble, then think that this width flow field figure is as I _kcomprise bubble; Only has error image upper all error image windows all do not comprise bubble, just think that this width flow field figure is as I _kdo not comprise bubble.

Described sub-step (6.1) forms mask images time, in be the pixel of 1 to all pixel values line by line, carry out following operation successively, until travel through the pixel that whole pixel value is 1:

(6.1.1). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that pixel value is 1 in 3 × 3 pixel regions around it, be, carry out process (6.1.2), otherwise the pixel value of this central point is set to 0, move to in next pixel value be the pixel of 1, turn over journey (6.1.1);

(6.1.2). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 5 × 5 pixel regions around it, be carry out process (6.1.3), otherwise the pixel value of all pixels in described 5 × 5 pixel regions is set to 0 entirely, move to in next pixel value be the pixel of 1, turn over journey (6.1.1);

(6.1.3). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 7 × 7 pixel regions around it, be carry out process (6.1.4), otherwise the pixel value of all pixels in described 7 × 7 pixel regions is set to 0 entirely, move to in next pixel value be the pixel of 1, turn over journey (6.1.1);

(6.1.4). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 9 × 9 pixel regions around it, be, carry out process (6.1.6), otherwise the process of carrying out (6.1.5);

(6.1.5). by pixel values all in 7 × 7 regions be 1 pixel with described central point for the center of circle, with decentering point distance pixel value farthest in 5 × 5 pixel regions around it be 1 pixel be that radius does circle to the distance of central point, circumferentially the pixel value of all pixels is set to 1, form closed interval, turn over journey (6.1.1);

(6.1.6) by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 11 × 11 pixel regions around it, be, then judge whether there is the pixel that new pixel value is 1 in 13 × 13 pixel regions around it, otherwise by pixel values all in 9 × 9 regions be 1 pixel with described central point for the center of circle, with decentering point distance pixel value farthest in 7 × 7 pixel regions around it be 1 pixel be that radius does circle to the distance of central point, circumferentially the pixel value of all pixels is set to 1, forms closed interval, so repeatedly, until by current pixel value be 1 pixel centered by point, no longer exist till new pixel value is the pixel of 1 around it in (2P+1) × (2P+1) pixel region, by all pixel values in (2P-1) × (2P-1) region be 1 pixel with described central point for the center of circle, with decentering point distance pixel value farthest in (2P-3) around it × (2P-3) pixel region be 1 pixel be that radius does circle to the distance of central point, circumferentially the pixel value of all pixels is set to 1, form closed interval, turn over journey (6.1.1), wherein P is for being less than the positive integer of one half-size scale,

(6.1.7) exist in aforesaid operations is completed to all pixel values pixel that is 1 after, form mask intermediate image ? the middle pixel value by all for each circumferential inner pixels is set to 1, forms mask images comprising " filled circles " that multiple pixel value is 1; In image, namely the position of bubble and size are all determined.

A kind of flow field figure provided by the present invention is as pretreatment system, comprise input Ethernet interface, Ethernet input control chip, digital signal processor, synchronous DRAM, dynamic memory chip, band EEPROM (Electrically Erasable Programmable Read Only Memo), export Ethernet interface, Ethernet exports control chip, it is characterized in that:

Ethernet input control chip controls input Ethernet interface and the communication of Ethernet camera, gather the view data will measuring flow field, deliver to synchronous DRAM by digital signal processor to keep in, dynamic memory chip is for storing Image Pretreatment Algorithm, band EEPROM (Electrically Erasable Programmable Read Only Memo) is used for storage configuration parameter, and configuration parameter comprises flow field figure film size number N, image window size parameter M; During work, Ethernet collected by camera high-speed motion flow field raw image data, be transferred in input Ethernet interface by gigabit Ethernet, digital signal processor reads the parameter in EEPROM, again the Image Pretreatment Algorithm that FLASH stores is loaded into self, then from SDRAM, reads image data carries out Image semantic classification, and the result of Image semantic classification exports control chip from exporting Ethernet interface by network cable transmission to PC by Ethernet.

Described flow field figure is as pretreatment system, and it is further characterized in that:

Described digital signal processor is configured with a joint test mouth, is convenient to user and carries out secondary development to Image Pretreatment Algorithm in digital signal processor.

Pretreated demand during the present invention measures mainly for High Speed Flow Field.This Image semantic classification system can be powered with small-sized 12V direct-current switch power supply, and size is no more than 13cm × 10cm, and weight is no more than 400g (not containing power supply), and volume is little, lightweight, is easy to carry; What the present invention selected is that gigabit Ethernet carries out communication, data rate can reach very high, the Ethernet camera collaborative work that pixel and frame per second can be coordinated very high, relatively be applicable to the large flow field survey occasion of volume of transmitted data, selected DSP is 8 core dsp processors, every core highest point reason speed is 1.25GHz, utilize 8 core parallel processings, the flow field figure of big data quantity can be realized under this platform as preprocessing function, all have better treatment effect for the phenomenon such as uneven illumination, bubble, water smoke occurred in the shooting of flow field; Selected FLASH capacity is 32MB, can store some Image Pretreatment Algorithms, and coordinate EEPROM can realize the selection selecting different Image Pretreatment Algorithms according to flow field actual conditions, adaptive faculty improves greatly; Result selects gigabit Ethernet to transmit, transmission speed is fast, transmitted data amount is large, not only can transmit pretreated flow field figure picture and also can transmit primary flow field picture, and can long-distance transmissions be carried out, mainly can solve the problem that Image semantic classification spended time in existing PIV instrument is longer, improve the speed of Image semantic classification.

Flow field figure of the present invention as pretreatment system and Ethernet camera with the use of, flow field survey image acquisition and Image semantic classification are placed on flow field figure of the present invention as the enterprising row relax of pretreatment system, final output be the picture of the flow field survey that noise is less, improve efficiency and the processing speed of whole back-end processor, flow field figure of the present invention is micro embedded disposal system as pretreatment system simultaneously, volume and power consumption are all smaller, can be good be embedded in existing flow field survey instrument, improve the Image semantic classification function of existing flow field survey instrument.

Accompanying drawing explanation

Fig. 1 is flow field of the present invention image pre-processing method FB(flow block);

Fig. 2 is for forming mask images fB(flow block);

Fig. 3 is that flow field figure of the present invention is as pretreatment system structural representation.

Embodiment

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

As shown in Figure 1, flow field of the present invention image pre-processing method, comprise: (1) equalization step, (2) image stretch step, (3) noise determining step, (4) steady noise step is removed, (5) structure binaryzation average image step, (6) mask operation step.

As an embodiment, in equalization step (1), N=3000, M=8 pixel; In structure binaryzation average image step (5), binary-state threshold G _kth=0.4.

The sub-step (3.2) of described noise determining step and the sub-step (4.3) of removal steady noise step judge that kth width flow field figure is as I _kon whether there is bubble, in the present embodiment, described error image window size is Q × Q, Q=11 pixel, bubble threshold value

Figure 2 shows that in mask operation step and form mask images fB(flow block);

As shown in Figure 3, flow field figure of the present invention, as pretreatment system, comprises input Ethernet interface, Ethernet input control chip, digital signal processor DSP, synchronous DRAM SDRAM, dynamic memory chip FLASH, band EEPROM (Electrically Erasable Programmable Read Only Memo) EEPROM, exports Ethernet interface, Ethernet exports control chip;

Ethernet input control chip controls input Ethernet interface and the communication of Ethernet camera, gather the view data will measuring flow field, deliver to synchronous DRAM SDRAM by digital signal processor DSP to keep in, dynamic memory chip FLASH is for storing Image Pretreatment Algorithm, band EEPROM (Electrically Erasable Programmable Read Only Memo) EEPROM is used for storage configuration parameter, and configuration parameter comprises flow field figure film size number N, image window size parameter M; During work, Ethernet collected by camera high-speed motion flow field raw image data, be transferred in input Ethernet interface by gigabit Ethernet, digital signal processor DSP reads the parameter in EEPROM, again the Image Pretreatment Algorithm that FLASH stores is loaded into self, then from SDRAM, reads image data carries out Image semantic classification, and the result of Image semantic classification exports control chip from exporting Ethernet interface by network cable transmission to PC by Ethernet.

Described digital signal processor DSP can be configured with a joint test mouth JTAG, is convenient to user and carries out secondary development to Image Pretreatment Algorithm in digital signal processor.

The Image Pretreatment Algorithm of exploitation can be burnt in FLASH by JTAG mouth, then in EEPROM, carries out corresponding parameter adjustment, selects the independently developed Image Pretreatment Algorithm of user to carry out the process of image when next time starts.

Because dsp processor is higher to power requirement, therefore power module should ensure that the ripple of voltage is less, ensures that the electrifying timing sequence of dsp processor carries out in strict accordance with handbook again.

In the present embodiment, selected dsp processor is TMS320C6678, FLASH is NAND512R3A2D, EEPROM is M24M01, JTAG is general 10P jtag interface, and Ethernet interface is general gigabit Ethernet mouth, ethernet control chip is 88E1111, SDRAM is K4B2G1646C.

Claims (5)

1. a flow field image pre-processing method, for the flow field that will observe, utilizes Ethernet camera to obtain N width flow field figure picture, it is characterized in that, for any kth width flow field figure picture, comprise the steps:

(1) equalization step:

To kth width flow field figure as I _kupper all pixels are slided image window line by line, calculate the equalization value N of each pixel x in the jth image window that formed _k ^j(x):

k＝1、…、N，N＝1000～50000，j≥1，0≤N _k ^j(x)≤1；

Wherein, I _k ^j(x) for kth width flow field figure is as the pixel value of pixel x in a jth window, I ^j _maxand I ^j _minbe respectively max pixel value and minimum pixel value in a jth image window:

${I^j}_{\max }=\underset{x\∈\mathrm{\Ω}}{\max }\left({I_k}^j\left(x\right)\right),$ ${I^j}_{\min }=\underset{x\∈\mathrm{\Ω}}{\min }\left({I^j}_k\left(x\right)\right),$

Ω is the territory at image window place, and described image window size is M × M, M=8,16 or 32 pixels;

(2) image stretch step, comprises following sub-step:

(2.1) for a jth image window of kth width image, the tension values S of wherein each pixel x is calculated ^j _k(x):

${S^j}_k\left(x\right)=\frac{{N^j}_k\left(x\right)-\stackrel{\‾}{I_{k\min }}}{\stackrel{\‾}{I_{k\max }}-\stackrel{\‾}{I_{k\min }}},$

Wherein be respectively the minimum pixel value in kth width image and max pixel value;

(2.2) for kth width image, the tension values S of each pixel x ^j _kx (), by the position of its respective pixel x, forms kth width stretching image S _k;

Image equalization step and image stretch step, each window in piece image carries out;

(3) noise determining step, comprises following sub-step:

(3.1) naked eyes judge that kth width flow field figure is as I _kon whether there is water smoke, be carry out step (4); Otherwise carry out sub-step (3.2);

(3.2) judge that kth width flow field figure is as I _kon whether there is bubble, be carry out step (5), otherwise export S _k, terminate;

(4) remove steady noise step, comprise following sub-step:

(4.1) first template image B is obtained _ref:

$B_{\mathrm{ref}}=\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{S}_k;$

The implication of above formula is, the pixel value that N width stretching image is corresponding is added, and addition result, divided by the position by its respective pixel x after N, forms template image B _ref;

(4.2) kth width stretching image S _kwith template image B _refcarry out image subtraction, obtain denoising image

${\stackrel{\‾}{S}}_k=S_k-B_{\mathrm{ref}};$

Described image subtraction refers to that the pixel value that two width images are corresponding subtracts each other, and subtracts each other the position of result by its respective pixel x, forms denoising image;

(4.3) judge that kth width flow field figure is as I _kon whether there is bubble, be carry out step (5), otherwise export terminate;

(5) construct binaryzation average image step, comprise following sub-step:

(5.1) the flow field figure picture image I as a setting that does not comprise bubble is selected _nomal, by kth width flow field figure as I _kwith background image I _nomalcarry out image subtraction, obtain kth width error image

${\stackrel{\‾}{I}}_k=I_k-I_{\mathrm{nomal}},$

(5.2) equalization operation is carried out to all pixels of kth width width error image, obtain the difference equalization value G of each pixel x _k(x):

$G_k\left(x\right)=\frac{{\stackrel{\‾}{I}}_k-I_{k\min }^{-}}{{\stackrel{\‾}{I}}_{k\max }-I_{k\min }^{-}},$ 0≤G _k(x)≤1；

Wherein, represent kth width error image minimum pixel value, represent kth width error image max pixel value;

(5.3) the difference equalization value G of each pixel x _kx (), by the position of its respective pixel x, forms kth width average image G _k;

(5.4) to G _kcarry out binaryzation operation, form kth width binary image the wherein pixel value of each pixel x

${\stackrel{\‾}{G}}_k\left(x\right)=\left\{\begin{array}{cc}0,& (G_k\left(x\right)\≤G_{\mathrm{kth}})\\ 1,& (G_k\left(x\right)>G_{\mathrm{kth}})\end{array}\right.;$

Wherein, 0≤binary-state threshold G _kth≤ 1; The laser energy being irradiated to flow field is larger, G _kththe closer to 1, otherwise, G _kththe closer to 0;

(6) mask operation step, comprises following sub-step:

(6.1) exist in the pixel that all pixel values are 1 is judged and is processed, formed mask images

(6.2) by kth width stretching image S _kor denoising image with mask images carry out image subtraction, obtain pretreated result images, Output rusults image, terminate.

2. flow field as claimed in claim 1 image pre-processing method, is characterized in that:

Described sub-step (3.2) and sub-step (4.3) judge that kth width flow field figure is as I _kon whether there is bubble, comprise following process:

A. naked eyes select the flow field figure picture image I as a setting that does not comprise bubble _nomal, by kth width flow field figure as I _kwith background image I _nomalcarry out image subtraction, obtain kth width error image

${\stackrel{\‾}{I}}_k=I_k-I_{\mathrm{nomal}},$

B. for error image upper all pixels slip error image window line by line, for each error image window, compares each pixel value and bubble threshold value that comprise in it size, judge to be greater than in error image window pixel value whether to exceed (Q/2+0.5) individual, be judge that this error image window comprises bubble, otherwise judge that this error image window does not comprise bubble;

Described error image window size is Q × Q, Q=9,11 or 13 pixels;

it is suitable to select according to different illumination conditions when illumination is stronger time, value near 255, when illumination is more weak, value near 0, generally value is 100;

C. error image any one error image window upper comprises bubble, then think that this width flow field figure is as I _kcomprise bubble; Only has error image upper all error image windows all do not comprise bubble, just think that this width flow field figure is as I _kdo not comprise bubble.

3. flow field as claimed in claim 1 image pre-processing method, is characterized in that:

Described sub-step (6.1) forms mask images time, in be the pixel of 1 to all pixel values line by line, carry out following operation successively, until travel through the pixel that whole pixel value is 1:

(6.1.1). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that pixel value is 1 in 3 × 3 pixel regions around it, be, carry out process (6.1.2), otherwise the pixel value of this central point is set to 0, move to in next pixel value be the pixel of 1, turn over journey (6.1.1);

(6.1.2). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 5 × 5 pixel regions around it, be carry out process (6.1.3), otherwise the pixel value of all pixels in described 5 × 5 pixel regions is set to 0 entirely, move to in next pixel value be the pixel of 1, turn over journey (6.1.1);

(6.1.3). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 7 × 7 pixel regions around it, be carry out process (6.1.4), otherwise the pixel value of all pixels in described 7 × 7 pixel regions is set to 0 entirely, move to in next pixel value be the pixel of 1, turn over journey (6.1.1);

(6.1.4). by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 9 × 9 pixel regions around it, be, carry out process (6.1.6), otherwise the process of carrying out (6.1.5);

(6.1.5). by pixel values all in 7 × 7 regions be 1 pixel with described central point for the center of circle, with decentering point distance pixel value farthest in 5 × 5 pixel regions around it be 1 pixel be that radius does circle to the distance of central point, circumferentially the pixel value of all pixels is set to 1, form closed interval, turn over journey (6.1.1);

(6.1.6) by current pixel value be 1 pixel centered by point, judge whether there is the pixel that new pixel value is 1 in 11 × 11 pixel regions around it, be, then judge whether there is the pixel that new pixel value is 1 in 13 × 13 pixel regions around it, otherwise by pixel values all in 9 × 9 regions be 1 pixel with described central point for the center of circle, with decentering point distance pixel value farthest in 7 × 7 pixel regions around it be 1 pixel be that radius does circle to the distance of central point, circumferentially the pixel value of all pixels is set to 1, forms closed interval, so repeatedly, until by current pixel value be 1 pixel centered by point, no longer exist till new pixel value is the pixel of 1 around it in (2P+1) × (2P+1) pixel region, by all pixel values in (2P-1) × (2P-1) region be 1 pixel with described central point for the center of circle, with decentering point distance pixel value farthest in (2P-3) around it × (2P-3) pixel region be 1 pixel be that radius does circle to the distance of central point, circumferentially the pixel value of all pixels is set to 1, form closed interval, turn over journey (6.1.1), wherein P is for being less than the positive integer of one half-size scale,

(6.1.7) exist in aforesaid operations is completed to all pixel values pixel that is 1 after, form mask intermediate image ? the middle pixel value by all for each circumferential inner pixels is set to 1, forms mask images comprising " filled circles " that multiple pixel value is 1; In image, namely the position of bubble and size are all determined.

4. a flow field figure is as pretreatment system, comprise input Ethernet interface, Ethernet input control chip, digital signal processor (DSP), synchronous DRAM (SDRAM), dynamic memory chip (FLASH), band EEPROM (Electrically Erasable Programmable Read Only Memo) (EEPROM), export Ethernet interface, Ethernet exports control chip, it is characterized in that:

Ethernet input control chip controls input Ethernet interface and the communication of Ethernet camera, gather the view data will measuring flow field, deliver to synchronous DRAM (SDRAM) by digital signal processor (DSP) to keep in, dynamic memory chip (FLASH) is for storing Image Pretreatment Algorithm, band EEPROM (Electrically Erasable Programmable Read Only Memo) (EEPROM) is for storage configuration parameter, and configuration parameter comprises flow field figure film size number N, image window size parameter M; During work, Ethernet collected by camera high-speed motion flow field raw image data, be transferred in input Ethernet interface by gigabit Ethernet, digital signal processor (DSP) reads the parameter in EEPROM, again the Image Pretreatment Algorithm that FLASH stores is loaded into self, then from SDRAM, reads image data carries out Image semantic classification, and the result of Image semantic classification exports control chip from exporting Ethernet interface by network cable transmission to PC by Ethernet.

5. flow field figure as claimed in claim 4 is as pretreatment system, it is characterized in that:

Described digital signal processor (DSP) is configured with a joint test mouth (JTAG), is convenient to user and carries out secondary development to Image Pretreatment Algorithm in digital signal processor.