CN107665350A

CN107665350A - Image-recognizing method and system and autofocus control method and system

Info

Publication number: CN107665350A
Application number: CN201610617161.4A
Authority: CN
Inventors: 盛司潼; 冀高
Original assignee: Guangzhou Kangxinrui Gene Health Technology Co Ltd
Current assignee: Guangzhou Kangxinrui Gene Health Technology Co Ltd
Priority date: 2016-07-29
Filing date: 2016-07-29
Publication date: 2018-02-06

Abstract

The present invention relates to field of image recognition, there is provided a kind of image-recognizing method and system, and its application.It the described method comprises the following steps：The system includes images match step：Using images match method, all area arrays to be matched in array of templates and testing image are subjected to similarity mode respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtain Similarity value image；First filtration step：The first filtering is carried out to Similarity value image using similarity threshold, obtains the first filtering image；Region recognition step：Continuous Similarity value region contour in first filtering image is obtained by region-growing method, and then obtains the subject matter profile in testing image.The method and system of the present invention can accurately identify subject matter profile in image, during especially for including a variety of articles of different shapes in testing image, can accurately identify subject matter profile in image.

Description

Image-recognizing method and system and autofocus control method and system

Technical field

The present invention relates to field of image recognition, and more specifically to image-recognizing method and system, and automatic focus on is controlled Method and system processed.

Background technology

Refer to a kind of method of the extraction SAR image objective contour based on Snakes models in the prior art, including based on Least bit noise model, using image local statistical property, a window is selected to enter as regional area using Lee filtering algorithms Row filtering；Condition judgment and amplitude adjustment are carried out to the gray value of each pixel of the image after clutter recognition, use segmented line Property greyscale transformation method to after clutter recognition image carry out greyscale transformation；Scheme using the spontaneous feature of profile and after greyscale transformation The feature of picture establishes energy function, and by solving energy function minimization problem, obtains objective contour curve.

This method is related to filtering, gray value condition judgment and amplitude adjustment, greyscale transformation, the foundation of energy function and asked Solution, finally just obtains objective contour curve, method and step is numerous and diverse.

Therefore a kind of method and system that can quickly identify subject matter profile in testing image is needed.

The content of the invention

It is an object of the invention to provide a kind of image-recognizing method and system, it is intended to solves prior art and obtains in image The problem of subject matter contour method step is numerous and diverse.

In order to realize goal of the invention, a kind of image-recognizing method, the described method comprises the following steps：

Images match step：Using images match method, by all area arrays to be matched point in array of templates and testing image Similarity mode is not carried out, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtains Similarity value figure Picture；

First filtration step：The first filtering is carried out to Similarity value image using similarity threshold, obtains the first filtering image；

Region recognition step：Continuous Similarity value region contour in first filtering image is obtained by region-growing method, and then Obtain the subject matter profile in testing image.

Wherein, above-mentioned image-recognizing method also includes array of templates acquisition step：Calculated according to Gauss nuclear operator and obtain mould Plate array；Or processing acquisition array of templates is carried out to the numerical matrix of subject matter region in real image.

Wherein, described image matching step comprises the following steps：

Normalization step：All area arrays to be matched in testing image are normalized according to array of templates, obtain normalizing Area array to be matched after change；

First matching step：Using images match method, array of templates and the area array to be matched after normalization are carried out respectively Similarity mode, and each Similarity value is invested to the pixel corresponded in region to be matched respectively, obtain Similarity value image.

Wherein, described image matching method is normalized crosscorrelation matching algorithm.

Wherein, above-mentioned image-recognizing method, in addition to the second filtration step：Using signal threshold value to each picture in testing image Vegetarian refreshments carries out the second filtering, obtains the second filtering image；

Described image matching step is using images match method, by all areas to be matched in array of templates and the second filtering image Domain array carries out similarity mode respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtains phase Like angle value image.

Wherein, the region recognition step obtains continuous similar in the first filtering image specifically by region-growing method Angle value region contour, while area size screening, the company after must screening are carried out to continuous Similarity value region using region threshold Continuous Similarity value region contour；And then obtain the subject matter profile in testing image.

Wherein, the testing image is high flux gene sequencing image.

Further, the testing image is the gray level image of high flux gene sequencing image.

Wherein, the subject matter is sphere.

Further, the subject matter is magnetic bead or microballon.

Further, the subject matter is luminous magnetic bead or microballon.

In order to further realize the purpose of the present invention, present invention also offers a kind of image identification system, including：

Images match module, for utilizing images match method, by all region battle arrays to be matched in array of templates and testing image Row carry out similarity mode respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtain similarity It is worth image；

First filtering module, for carrying out the first filtering to Similarity value image using similarity threshold, obtain the first filtering image；

Region identification module, for obtaining the continuous Similarity value region contour in the first filtering image by region-growing method, And then obtain the subject matter profile in testing image.

Wherein, above-mentioned image identification system also includes array of templates acquisition module, is calculated according to Gauss nuclear operator and obtains mould Plate array；Or processing acquisition array of templates is carried out to the numerical matrix of subject matter region in real image.

Wherein, described image matching module includes normalization unit and the first matching unit；

The normalization unit, for carrying out normalizing to all area arrays to be matched in testing image according to array of templates Change, the area array to be matched after must normalizing；

First matching unit, for utilizing images match method, by array of templates and the area array to be matched after normalization Similarity mode is carried out respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtains Similarity value Image.

Wherein, above-mentioned image identification system also includes the second filtering module, for utilizing signal threshold value in testing image Each pixel carries out the second filtering, obtains the second filtering image；

Described image matching module, for utilizing images match method, array of templates is filtered in image with second and needed Similarity mode is carried out respectively with area array, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, Obtain Similarity value image.

Wherein, the region identification module, it is continuous similar in the first filtering image for being obtained by region-growing method Angle value region contour, while area size screening, the company after must screening are carried out to continuous Similarity value region using region threshold Continuous Similarity value region contour；And then obtain the subject matter profile in testing image.

Wherein, the testing image is high flux gene sequencing image.

Wherein, the subject matter is sphere.

Further, the subject matter is magnetic bead or microballon.

Further, the subject matter is luminous magnetic bead or microballon.

In order to further realize the purpose of the present invention, present invention also offers a kind of image-recognizing method, including following step Suddenly：

Image recognizing step, identify the subject matter wheel in multiple sequencing images respectively using any of the above-described kind of image-recognizing method It is wide；

Definition judgment step, the subject matter quantity each contained in the multiple sequencing image is counted, then according to subject matter The number of quantity determines definition highest sequencing image in the multiple sequencing image.

Picture recognition module, identify the subject matter wheel in multiple sequencing images respectively using any of the above-described kind of image identification system It is wide；

Definition judgment module, the subject matter quantity each contained in the multiple sequencing image is counted, then according to subject matter The number of quantity determines definition highest sequencing image in the multiple sequencing image.

In order to further realize the purpose of the present invention, present invention also offers a kind of autofocus control method, including with Lower step：

Focusing step, repeatedly adjust imaging device and adopt the distance of figure position, record the focal position after each regulation, will gather Image command, which is issued, adopts module, and after instruction of the focusing to best focus position is received, regulation imaging device is moved to most Good focal position；

Figure step is adopted, after collection image command is received, image is gathered using imaging device, obtains the multiple of different focal positions Sequencing image；

Image processing step, after multiple sequencing images of the different focal positions are received, using any of the above-described kind of image Recognition methods, determine definition highest sequencing image in the multiple sequencing image, using definition highest sequencing image as Best focus position, and focusing module is issued into the instruction focused to best focus position.

In order to further realize the purpose of the present invention, present invention also offers a kind of auto focus control system, including：

Focusing module, for repeatedly adjusting imaging device and adopting the distance of figure position, the focal position after regulation every time is recorded, will Collection image command, which is issued, adopts module, after instruction of the focusing to best focus position is received, regulation imaging device movement To best focus position；

Module is adopted, for after collection image command is received, gathering image using imaging device, obtaining different focal positions Multiple sequencing images；

Image processing module, for after multiple sequencing images of the different focal positions are received, using any of the above-described kind Image identification system, definition highest sequencing image in the multiple sequencing image is determined, with definition highest sequencer map Focusing module is issued as being best focus position, and by the instruction focused to best focus position.

From the foregoing, it will be observed that the present invention method and system by the way that array of templates is matched with area array to be matched, so The Similarity value of acquisition is filtered afterwards, and then the subject matter profile in identification acquisition testing image is increased by region, because This, method and system of the invention can accurately identify subject matter profile in image, include especially for testing image When there are a variety of articles of different shapes, subject matter profile in image can be accurately identified.Based on target in above-mentioned identification image The method and system of object area, present invention also offers another image for being capable of accurate more multiple sequencing image definition Recognition methods and system, and can accurately realize the autofocus control method and system of the automatic focusing of sequenator.

Brief description of the drawings

Fig. 1 is the method flow diagram of first embodiment of the invention.

Fig. 2 is the Method And Principle schematic diagram of first embodiment of the invention.

Fig. 3 is another Method And Principle schematic diagram of first embodiment of the invention.

Fig. 4 is that filtering diagram is intended to obtained by the step S02 of first embodiment of the invention.

Fig. 5 is several examples of the array of templates of the present invention.

Fig. 6 is curve map corresponding to the Gauss nuclear operator of the present invention.

Fig. 7 is the gray level image of the high flux gene sequencing image in the specific embodiment of the present invention.

Fig. 8 is the numerical matrix calculated based on Gauss nuclear operator obtained by Fig. 7 adjustment.

Fig. 9 is the array of values in several regions to be matched in one embodiment of the invention in the front and rear comparison diagram of normalization.

Figure 10 is the schematic diagram of the image identification system in one embodiment of the present of invention.

Figure 11 is the flow chart of the image-recognizing method in one embodiment of the present of invention.

Figure 12 is the schematic diagram of the image identification system in one embodiment of the present of invention.

Figure 13 is the flow chart of the autofocus control method in one embodiment of the present of invention.

Figure 14 is the schematic diagram of the auto focus control system in one embodiment of the present of invention.

Embodiment

In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples The present invention is further elaborated.

The present invention proposes first embodiment, as shown in Figure 1, Figure 2, Figure 3 and Figure 4, a kind of image-recognizing method, including it is following Step：

S01, images match step：Using images match method, by all region battle arrays to be matched in array of templates and testing image Row carry out similarity mode respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtain similarity It is worth image；

S02, the first filtration step：The first filtering is carried out to Similarity value image using similarity threshold, obtains the first filtering image；

S03, region recognition step：Continuous Similarity value region contour in first filtering image is obtained by region-growing method, And then obtain the subject matter profile in testing image.

It should be noted that the array of templates is the numerical value with definite shape and size being made up of series of values Array；The actual Similarity value image is Similarity value array.As shown in Fig. 2 area array 1 to be matched is defined to wait to be sequenced Any one in image and the array of values of each pixel numerical value composition in array of templates size, shape identical region, respectively There may be lap between area array to be matched.First filtering is according to similarity threshold, identifies Similarity value Similarity value is more than or equal to the pixel of similarity threshold in image.The first filtering image, it is defined as walking through the first filtering After rapid, it is identified in Similarity value image, Similarity value is more than or equal to the image that the pixel of similarity threshold forms.Institute Continuous Similarity value region is stated, it is the phase being identified in the step S02 to be defined as all Similarity values in the region Like angle value, and the spacing between each adjacent Similarity value is identical.

Fig. 3 is only to illustrate, and the Similarity value image in Fig. 3 only lists several representative higher similarities Value, remaining Similarity value are not shown in figure 3.

In the procedure of reality, the size and shape of array of templates can carry out accommodation as needed, such as Size and shape of the subject matter in testing image.

Fig. 4 only illustrates, and one of them that illustrate only in Fig. 4 in the first filtering image of the gained after step S02 connects Continuous Similarity value region and its enlarged drawing, remaining continuous Similarity value region are not shown.

In addition, quantity without concrete restriction of the present invention to the subject matter in testing image.

The testing image can be common photo, or professional equipment is taken pictures gained photo, such as utilize high flux base Because of high flux gene sequencing image obtained by sequenator shooting.

The subject matter is depending on the object for needing to identify in specific testing image, including but not limited in testing image Lived thing or the effigurate object of tool.The subject matter is preferably the object with special construction or shape, special The requirement of structure or shape for images match method is relatively low, it is necessary to which the factor considered is less, and Similarity value differs greatly, and passes through The subject matter profile that this method identifies is more accurate.When testing image is high flux gene sequencing image, the subject matter Preferably sphere or like ball thing, such as magnetic bead or microballon.

The method of the present embodiment is by the way that array of templates is matched with area array to be matched, then to the similar of acquisition Angle value is filtered, and then increases the continuous Similarity value region contour in identification acquisition the first filtering image by region, will The continuous Similarity value region contour is considered as the subject matter profile in testing image.Therefore, the method for the present embodiment can be accurate Identification testing image in subject matter profile, during especially for including a variety of articles of different shapes in testing image, energy The profile of subject matter in enough accurately identification images.

In the specific embodiment of the present invention, the Similarity value is invested in corresponding area array to be matched respectively Imago vegetarian refreshments., also can be by these phases in the alternate embodiment that the central pixel point of area array to be matched is invested in Similarity value Invested respectively like angle value adjacent above pixel, such as central pixel point corresponding to the other positions in area array to be matched Adjacent pixel below pixel or central pixel point.

In another specific embodiment, testing image size is M × N, and array of templates size is P × Q, the area to be matched Domain array is to treat that any one in sequencing image forms with each pixel numerical value in array of templates size, shape identical region Array of values, then area array size to be matched is also P × Q, then in testing image share（M-P+1）×（N-Q+1） Individual area array to be matched.Now, the array of templates is carried out respectively with all area arrays to be matched in testing image Similarity mode, carried out successively with the movement of individual element point.It is preferred that from wantonly 1 corner in 4 corners of testing image Start, and terminate from wantonly 1 corner in remaining 3 corner, if the motion track in whole matching process for snakelike route or Dry parallel lines.The present embodiment can realize Rapid matching, and ensure the accuracy rate of matching.

In the alternate embodiment of above-mentioned specific embodiment, the array of templates and all areas to be matched in testing image Similarity mode that domain array is carried out respectively or with every time it is mobile 2,3,5 or more pixel movements according to Secondary progress.Accordingly, in this alternate embodiment, the area array to be matched is to treat any one in sequencing image and template The array of values of each pixel numerical value composition in array size, shape identical region, and in each area array to be matched Spacing between the central point of heart point area array to be matched adjacent thereto is 2,3,5 or more pixels.Originally replace Matching efficiency can be significantly improved for embodiment.

For similarity threshold it should be noted that the similarity threshold can be set as needed.Treated for example, working as Altimetric image is high flux gene sequencing image, and when the subject matter is magnetic bead or microballon, the similarity threshold is preferably in 30%- Between 100%.In the present embodiment because high flux gene sequencing image more rule, be substantially not present in image except subject matter with Other outer articles, so similarity threshold can be set relatively low.

In one embodiment, the similarity threshold is preferentially between 55%-100%.This specific embodiment can be compared with It is good to avoid that partly overlapping subject matter is identified in testing image being between each other present.Because phase in testing image Partly overlapping subject matter be present between mutually, can be interfered with each other in the subsequent processes of high flux gene sequencing image, dropped The accuracy of reduction process result, so this programme can improve the accuracy of whole high flux gene sequencing processing result image.

For array of templates, it is necessary to illustrate, the shapes and sizes of the array of templates preferably with testing image The shapes and sizes of subject matter are consistent or almost consistent.The array of templates that part can use in the present invention is shown in Fig. 5.Institute Array of values or numerical matrix with definite shape and size can be set to as needed by stating array of templates.The template battle array Numerical value in row can be light signal strength, and the light signal strength can be to be represented by rgb value, i.e. colour picture signal value； Can be to be represented by gray value, i.e. black-and-white image signal value.The corresponding testing image can be that coloured image is alternatively black and white Image.

Preferably, each numerical value in the array of templates corresponds to the artwork master of one or more pixels of the position The gray value of picture.When the region to be matched that this programme can be substantially reduced in array of templates and testing image carries out similarity mode Operand, improve efficiency.

On the basis of any of the above-described embodiment, the present invention proposes one embodiment, also includes mould before the step S01 Plate array obtains step S00：Processing acquisition array of templates is carried out to the numerical matrix of subject matter region in real image.

The present embodiment obtains array of templates based on actually detected obtained numerical matrix after processing, and method is simple, right The relatively regular or smooth object of subject matter is especially suitable in image.It should be noted that the processing in the present embodiment can have A variety of implementations, it will be specifically described below by numerous embodiments.

In one embodiment of the invention, the array of templates is method manually or automatically, to having schemed Subject matter as in is detected, and then obtains its corresponding array of values, and the numerical matrix can be directly as array of templates.

In another embodiment of the present invention, the acquisition of the array of templates is similar with a upper embodiment, Bu Guoqi By being detected to the subject matter in existing image, and then multiple array of values are obtained, then pass through the multiple numerical value battle array After row carry out average value processing, the array of values after average value processing is obtained as array of templates.

In the present embodiment, by average value processing, it can make that array of templates is more representative, and matching result is more accurate.

In addition, the array of templates can according to subject matter represented characteristic under light illumination, by corresponding algorithm and .For example, when the gray level image that the testing image is high flux gene sequencing image, when subject matter is magnetic bead or microballon, treat The intensity of reflected light distribution of magnetic bead is similar with the curve distribution of Gauss nuclear operator in altimetric image.The mathematical table of the Gauss nuclear operator It is up to formula：

The figure of above-mentioned mathematic(al) representation is as shown in Figure 6.Wherein a represents the height of curve, and b refers to curve in the center of x-axis, c Refer to the width of curve.

As shown in fig. 6, function of region value is maximum centered on the characteristics of above-mentioned Gauss nuclear operator, the equidistant function with central point It is worth identical, and functional value is relative to the specific exponentially type downward trend of central point and the point.With in testing image magnetic bead it is anti- It is very much like to penetrate light intensity distributions.According to the size of magnetic bead in testing image, the gray-value variation trend of each pixel of magnetic bead, adjust A, b, c value in whole above-mentioned mathematic(al) representation, you can obtain the Gauss nuclear operator highly similar to magnetic bead in testing image.This reality Magnetic bead or microballon profile that the method for example is particularly suitable for use in identification high flux gene sequencing image are applied, and result is accurate.

In one particular embodiment of the present invention, Fig. 7 is the gray level image of high flux gene sequencing image, and Fig. 8 is base In the numerical matrix that Gauss nuclear operator obtained by Fig. 7 adjustment calculates, the numerical matrix can be used as array of templates, for similarity Matching, testing image be with shooting under Fig. 7 the same terms obtained by high flux gene sequencing image gray level image, and then pass through The profile of magnetic bead in the testing image that subsequent treatment obtains.It is corresponding, a in the expression formula of the Gauss nuclear operator= 35100、b=0、c=4/3。

Because during the similarity mode of the step S01, what Similarity value considered is array of templates with treating Subfield value to be matched and numerical value change trend in altimetric image, and influenceed by illumination variation, the difference in same testing image The numerical value of the subject matter of position may differ greatly, but numerical value change trend is identical；Obtained by the same terms shooting not Above mentioned problem is equally existed with the subject matter in testing image.

For influence of the numerical value difference to Similarity value caused by solving illumination variation, the present invention is in any of the above-described embodiment On the basis of propose an embodiment, methods described be additionally included in the step S01 it is specific before carrying out matching primitives each time, also The step of including the area array to be matched in testing image is normalized.This programme can reduce illumination variation to similarity The influence of value, accuracy rate of the image-recognizing method of the present invention in similarity mode is improved, and then more accurately obtained to be measured All subject matter profiles in image.

On the basis of above-described embodiment, the present invention proposes another embodiment, and the step S01 comprises the following steps：

S011, normalization step：All area arrays to be matched in testing image are normalized, treated after must normalizing Matching area array；

S012, the first matching step：Using images match method, array of templates and the area array to be matched after normalization are carried out Similarity mode, and each Similarity value is invested to the pixel corresponded in region to be matched respectively, obtain Similarity value image.

It should be noted that the normalization step there are a variety of implementations, will be carried out below by multiple embodiments It is expanded on further.

In one embodiment of the invention, the normalization step is using array of templates as normalization array of templates, institute The each numerical value needed in matching area array is multiplied by z；The z is coefficient of variation, for example, the center of normalization array of templates The ratio of point value and the centerpoint value in region to be matched.In one particular embodiment of the present invention, z=a/b；The a is normalizing Change the centerpoint value of array of templates, the b is the centerpoint value in region to be matched.

Fig. 9 shows in one embodiment of the invention that the array of values in several regions to be matched is in front and rear pair of normalization Than.Wherein, for masterplate array as normalization array of templates, A, B, C are the array of values before region normalization to be matched respectively, A ', B ', C ' are the array of values after region A, B, C normalization to be matched respectively.

For normalization array of templates it should be noted that it is not limited to array of templates, in another implementation of the present invention In example, the normalization array of templates is the array of values separately built, different from array of templates.In normalization, by template The array of values in array and region to be matched is normalized by the normalization array of templates.

For the coefficient of variation, it is not limited to central point of the centerpoint value with region to be matched for normalizing array of templates The ratio of value, as long as this ratio is the ratio for normalizing the numerical value of same position in array of templates and region to be matched. Such as normalize the ratio of point value adjacent above the array of templates central point point value adjacent with above regional center point to be matched.

It is high flux gene sequencing image for testing image, when subject matter is magnetic bead, because subject matter and array of templates In be that central point is most bright place；That is numerical value highest place；Using centerpoint value ratio as coefficient of variation, can improve similar Spend the accuracy of matching.

It is high flux gene sequencing image for testing image, subject matter is the concrete condition of magnetic bead, in any of the above-described reality On the basis of applying example, the present invention proposes a specific embodiment, and described image matching method is to carry out matching primitives based on Euclidean distance.

In the another specific embodiment of the present invention, described image matching method is normalized crosscorrelation matching algorithm.

In one particular embodiment of the present invention, the normalized crosscorrelation matching algorithm is：

Wherein, NCC is Similarity value, and w (x, y) is array of templates, and g (x, y) is region to be matched, is array of templates average, is Regional average value to be matched, m are the length of template area, and n is the width of template area.

This specific embodiment while matching primitives are carried out, realizes normalizing by normalized crosscorrelation matching algorithm Change, both reduced influence of the illumination variation to Similarity value, improve standard of the image-recognizing method of the present invention in similarity mode True rate, and then all subject matter profiles in testing image are more accurately obtained, calculating process is in turn simplify, improves efficiency.

In order to avoid influence of the noise to subject matter profile in acquisition testing image in testing image, the present invention proposes another A kind of embodiment, image-recognizing method, comprises the following steps：

S001, the second filtration step：The second filtering is carried out to each pixel in testing image using signal threshold value, obtains the second filtering Image；

S01, images match step：Using images match method, by all areas to be matched in array of templates and the second filtering image Domain array carries out similarity mode respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtains phase Like angle value image；

It should be noted that second filtering is according to signal threshold value, identify in testing image that signal value is more than In the pixel of signal threshold value.The second filtering image, is defined as after the second filtration step, is identified in testing image , signal value is more than or equal to the image that the pixel of signal threshold value forms.In the present embodiment, area array to be matched is defined as the Any one in two filtering images and the numerical value battle array of each pixel numerical value composition in array of templates size, shape identical region Row.Other necessity define identical with first embodiment.

By handling in advance testing image in the present embodiment, it is to be measured to obtaining noise in testing image had both been avoided The influence of subject matter profile in image, reduces the operand in S01 steps again, improves the thing profile to testing image acceptance of the bid Recognition accuracy and recognition efficiency.

In the present embodiment, the signal threshold value can be set as needed.For example, when testing image is gray-scale map During picture, the signal threshold value can be 100,300,500,1000 or higher.

When testing image is high flux gene sequencing image, and subject matter is luminous magnetic bead, because treating the back of the body of sequencing image Jing Zhonghui cannot not remain luminously the false signal of magnetic bead and the weak signal of random distribution, these noises in images match step, Larger matching factor may also be calculated, and then by the first follow-up filtration step, and in follow-up region recognition step It is identified, and is considered as subject matter, so as to exports corresponding profile.And often signal value is relatively low for above-mentioned noise, therefore increase Adaptation is carried out plus the second filtration step stated, and to images match step, can effectively remove above-mentioned noise Influence, improve the recognition accuracy and recognition efficiency of the thing profile to testing image acceptance of the bid.

In one particular embodiment of the present invention, when the gray level image that testing image is high flux gene sequencing image When, when subject matter is luminous magnetic bead, the signal threshold value is preferably between 300 to 1000.In this section, it can preferably remove and treat Noise in sequencing image.

In order to further improve the recognition accuracy of the thing profile to testing image acceptance of the bid, the present invention is in any of the above-described implementation Another embodiment is proposed on the basis of example, in the embodiment, the region recognition step obtains specifically by region-growing method Continuous Similarity value region contour in first filtering image, while using region threshold to continuous Similarity value region carry out area Domain size screening, the continuous Similarity value region contour after must screening；And then obtain the subject matter profile in testing image.

The present embodiment by region-growing method obtain first filtering image in continuous Similarity value region contour it is same When, area size screening is carried out to continuous Similarity value region using region threshold, retains and meets the continuous of region threshold requirement Similarity value region contour, that is, the continuous Similarity value region contour after screening；Continuous Similarity value region after screening is taken turns Exterior feature is considered as the subject matter profile in testing image；So as to exclude the mistake obtained in the region recognition step of above-described embodiment Small and/or excessive continuous Similarity value region, further improve the recognition accuracy of the thing profile to testing image acceptance of the bid.

For above-described embodiment, the present invention proposes another alternate embodiment, in the embodiment, the region recognition step tool Body is that the continuous Similarity value region contour in the first filtering image is obtained by region-growing method, then utilizes region threshold pair Continuous Similarity value region carries out area size screening, the continuous Similarity value region contour after must screening；And then obtain to be measured Subject matter profile in image.

The differing only in of the present embodiment and a upper embodiment carry out region threshold screen the time is different；It can equally enter One step improves the recognition accuracy of the thing profile to testing image acceptance of the bid.

In one particular embodiment of the present invention, when the gray level image that testing image is high flux gene sequencing image When, when subject matter is luminous magnetic bead, the region threshold is preferably between 3 to 100.In this section, it can preferably remove and treat The continuous Similarity value region that background noise in altimetric image is identified in region recognition step, it is typically small；And can is gone Except the getting together too closely of multiple luminous magnetic beads present in testing image, or even there is the luminous magnetic bead of lap；Drop The intractability of low follow-up possible data analysis.

Present invention also offers second embodiment, a kind of image identification system, as shown in Figure 10, the system 100 include： Images match module 110, the first filtering module 120 and region identification module 130；

Described image matching module 110, for utilize images match method, by array of templates with it is all to be matched in testing image Area array carries out similarity mode respectively, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, obtains Similarity value image；

First filtering module 120, for carrying out the first filtering to Similarity value image using similarity threshold, obtain the first mistake Filter image；

The region identification module 130, for obtaining the continuous Similarity value area in the first filtering image by region-growing method Domain profile, and then obtain the subject matter profile in testing image.

It should be noted that in the present embodiment, the module or system can perform the integrated circuit of specific function, It can be stored in storage device and the software program of corresponding function is completed by computing device.

The array of templates is the array of values with definite shape and size being made up of series of values；It is described similar Actual angle value image is Similarity value array.Area array to be matched is defined to treat any one in sequencing image and array of templates The array of values of each pixel numerical value composition in size, shape identical region.First filtering is according to similarity threshold Value, identify that Similarity value is more than or equal to the pixel of similarity threshold in Similarity value image.The first filtering image, it is fixed Justice is after the processing of the first filtering module, is identified in Similarity value image, Similarity value is more than or equal to similarity threshold Pixel composition image.The continuous Similarity value region, it is by first to be defined as all Similarity values in the region The Similarity value that filtering module 120 identifies, and the spacing between each adjacent Similarity value is identical.

In the present embodiment, the size and shape of array of templates can carry out accommodation as needed, such as subject matter exists Size and shape in testing image.In the present embodiment, to the quantity without concrete restriction of the subject matter in testing image.It is described to treat Altimetric image can be common photo, or professional equipment is taken pictures gained photo, for example, using the shooting of high flux gene sequencer and The high flux gene sequencing image obtained.

The subject matter is depending on the object for needing to identify in specific testing image, including but not limited in testing image Lived thing or the effigurate object of tool.The subject matter is preferably the object with special construction or shape, special The requirement of structure or shape for images match module 110 it is relatively low, it is necessary to consider factor it is less, and Similarity value difference compared with Greatly, the subject matter profile identified by the system is more accurate.It is described when testing image is high flux gene sequencing image Subject matter is preferably sphere or like ball thing, such as magnetic bead or microballon.

In the present embodiment, regarded the continuous Similarity value region contour in the first filtering image is obtained by region-growing method For subject matter profile in testing image；The image identification system 100 of the present embodiment is filtered by images match module 110, first The mutual cooperation of module 120 and region identification module 130, subject matter profile in testing image can be accurately identified, especially During for including a variety of articles of different shapes in testing image, subject matter profile in testing image can be accurately identified.

In another specific embodiment, testing image size is M × N, and array of templates size is P × Q, then region to be matched Array size is also P × Q, then is shared in testing image（M-P+1）×（N-Q+1）Individual area array to be matched.Now, institute The similarity mode that array of templates is carried out respectively with all area arrays to be matched in testing image is stated, is with individual element point What movement was carried out successively.In this specific embodiment, the area array to be matched is to treat any one in sequencing image and template The array of values of each pixel numerical value composition in array size, shape identical region.

For similarity threshold, it is necessary to illustrate, the similarity threshold can be set as needed.For example, work as Testing image is high flux gene sequencing image, and when the subject matter is magnetic bead or microballon, the similarity threshold preferably exists Between 30%-100%.In the present embodiment, because high flux gene sequencing image more rule, it is substantially not present in image except target Other articles beyond thing, so similarity threshold can be set relatively low.

On the basis of any of the above-described embodiment, the present invention proposes one embodiment, and described image identifying system also includes Array of templates obtains module：Processing acquisition array of templates is carried out to the numerical matrix of subject matter region in real image.

The present embodiment obtains array of templates based on actually detected obtained numerical matrix after processing, simple, efficient, right The relatively regular or smooth object of subject matter is especially suitable in image.It should be noted that the processing in the present embodiment can have A variety of implementations, it will be specifically described below by numerous embodiments.

In one embodiment of the invention, the array of templates obtains module, for the target in existing image Thing is detected, and then obtains its corresponding array of values, and the numerical matrix can be directly as array of templates.

In another embodiment of the present invention, the array of templates acquisition module is similar with a upper embodiment, but It is specially：For being detected to the subject matter in existing image, and then multiple array of values are obtained, then by described more After individual array of values carries out average value processing, the array of values after average value processing is obtained as array of templates.

For above-described embodiment, the present invention proposes another alternate embodiment, and described image identifying system also includes template battle array Row obtain module：For according to subject matter represented characteristic under light illumination, passing through corresponding algorithm and obtaining array of templates.

For example, when the gray level image that the testing image is high flux gene sequencing image, subject matter is magnetic bead or microballon When, the intensity of reflected light distribution of magnetic bead is similar with the curve distribution of Gauss nuclear operator in testing image.The Gauss nuclear operator Mathematic(al) representation is：

As shown in fig. 6, function of region value is maximum centered on the characteristics of above-mentioned Gauss nuclear operator, the equidistant function with central point It is worth identical, and functional value is relative to the specific exponentially type downward trend of central point and the point.With in testing image magnetic bead it is anti- It is very much like to penetrate light intensity distributions.According to the size of magnetic bead in testing image, the gray-value variation trend of each pixel of magnetic bead, adjust A, b, c value in whole above-mentioned mathematic(al) representation, you can obtain the Gauss nuclear operator highly similar to magnetic bead in testing image.This reality Magnetic bead or microballon profile that the system of example is particularly suitable for use in identification high flux gene sequencing image are applied, and result is accurate.

Now, the array of templates obtains module, and array of templates is obtained for being calculated according to Gauss nuclear operator.

In one particular embodiment of the present invention, Fig. 7 is the gray level image of high flux gene sequencing image, and Fig. 8 is base In the numerical matrix that Gauss nuclear operator obtained by Fig. 7 adjustment calculates, the numerical matrix can be used as array of templates, for similarity Matching, testing image be with shooting under Fig. 7 the same terms obtained by high flux gene sequencing image gray level image, and then pass through The region of magnetic bead in the testing image that subsequent treatment obtains.It is corresponding, a in the expression formula of the Gauss nuclear operator= 35100、b=0、c=4/3。

Because described image matching module is in the course of the work, what Similarity value considered is array of templates and treats mapping Subfield value to be matched and numerical value change trend as in, and influenceed by illumination variation, the diverse location in same testing image The numerical value of subject matter may differ greatly, but numerical value change trend is identical；Difference obtained by the same terms shooting is treated Subject matter in altimetric image equally exists above mentioned problem.

For influence of the numerical value difference to Similarity value caused by solving illumination variation, the present invention is in any of the above-described embodiment On the basis of propose an embodiment, described image matching module, be additionally operable to before matching primitives each time are carried out, to testing image In area array to be matched be normalized.This programme can reduce influence of the illumination variation to Similarity value, improve the present invention Accuracy rate of the image identification system in similarity mode, and then more accurately obtain all subject matter wheels in testing image It is wide.

On the basis of above-described embodiment, the present invention proposes another embodiment, and described image matching module includes normalization Unit and the first matching unit；

It should be noted that the normalization unit there are a variety of implementations, will be carried out below by multiple embodiments It is expanded on further.

In one embodiment of the invention, the normalization unit is using array of templates as normalization array of templates, institute The each numerical value needed in matching area array is multiplied by z；The z is coefficient of variation, for example, the center of normalization array of templates The ratio of point value and the centerpoint value in region to be matched.In one particular embodiment of the present invention, z=a/b；The a is normalizing Change the centerpoint value of array of templates, the b is the centerpoint value in region to be matched.

Fig. 9 shows in one embodiment of the invention that the array of values in several regions to be matched is at through normalization unit The front and rear contrast of reason.Wherein, for masterplate array as normalization array of templates, A, B, C are that region to be matched is single through normalization respectively The array of values of first before processing, A ', B ', C ' are the array of values of region A, B, C to be matched after normalization unit is handled respectively.

For normalization array of templates it should be noted that it is not limited to array of templates, in another implementation of the present invention In example, the normalization array of templates is the array of values separately built, different from array of templates.When normalization unit works, The array of values in array of templates and region to be matched is normalized by the normalization array of templates.

It is high flux gene sequencing image for testing image, when subject matter is magnetic bead, because subject matter and array of templates In be that central point is most bright place；That is, numerical value highest place；Using centerpoint value ratio as coefficient of variation, phase can be improved Like the accuracy of degree matching.

This specific embodiment while matching primitives are carried out, realizes normalizing by normalized crosscorrelation matching algorithm Change, both reduced influence of the illumination variation to Similarity value, improve standard of the image identification system of the present invention in similarity mode True rate, and then all subject matter profiles in testing image are more accurately obtained, calculating process is in turn simplify, improves efficiency.

In order to avoid influence of the noise to subject matter profile in acquisition testing image in testing image, the present invention proposes another Embodiment, a kind of image identification system, including：Second filtering module, images match module, the first filtering module and region recognition Module；

Second filtering module, for carrying out the second filtering to each pixel in testing image using signal threshold value, obtain second Filter image；

Described image matching module, for utilizing images match method, array of templates is filtered in image with second and needed Similarity mode is carried out respectively with area array, and each Similarity value is invested to the pixel in corresponding region to be matched respectively, Obtain Similarity value image；

The region identification module, taken turns for obtaining the continuous Similarity value region in the first filtering image by region-growing method Exterior feature, and then obtain the subject matter profile in testing image.

It should be noted that second filtering is according to signal threshold value, identify in testing image that signal value is more than In the pixel of signal threshold value.The second filtering image, is defined as after the processing of the second filtering module, is known in testing image Do not go out, signal value is more than or equal to the image that the pixel of signal threshold value forms.In the present embodiment, area array to be matched is defined For number of any one in the second filtering image with each pixel numerical value composition in array of templates size, shape identical region Value array.Other necessity define identical with second embodiment.

By handling in advance testing image in the present embodiment, it is to be measured to obtaining noise in testing image had both been avoided The influence of subject matter profile in image, reduce the operand of images match module again, improve the thing to testing image acceptance of the bid The recognition accuracy and recognition efficiency of profile.

When testing image is high flux gene sequencing image, and subject matter is luminous magnetic bead, because treating the back of the body of sequencing image Jing Zhonghui cannot not remain luminously the false signal of magnetic bead and the weak signal of random distribution, and these noises work in images match module During, it is also possible to larger matching factor is calculated, and then is identified by the first filtering module, and is considered as subject matter, So as to export corresponding profile.And often signal value is relatively low for above-mentioned noise, therefore increase the second above-mentioned filtering module, and to figure As matching module progress adaptation, the influence of above-mentioned noise can be effectively removed, improves the thing to testing image acceptance of the bid The recognition accuracy and recognition efficiency of profile.

In order to further improve the recognition accuracy of the thing profile to testing image acceptance of the bid, the present invention is in any of the above-described implementation Another embodiment is proposed on the basis of example, in the embodiment, the region identification module, for obtaining the by region-growing method Continuous Similarity value region contour in one filtering image, while region is carried out to continuous Similarity value region using region threshold Size is screened, the continuous Similarity value region contour after must screening；And then obtain the subject matter profile in testing image.

Continuous similarity of the region identification module of the present embodiment in the first filtering image is obtained by region-growing method While being worth region contour, area size screening is carried out to continuous Similarity value region using region threshold, reservation meets region The continuous Similarity value region contour of threshold requirement, that is, the continuous Similarity value region contour after screening；Will be continuous after screening Similarity value region contour is considered as the subject matter profile in testing image；So as to exclude the region recognition in above-described embodiment The too small and/or excessive continuous Similarity value region that module obtains, further improve to the thing profile of testing image acceptance of the bid Recognition accuracy.

For above-described embodiment, the present invention proposes another alternate embodiment, in the embodiment, the region identification module, For obtaining the continuous Similarity value region contour in the first filtering image by region-growing method, region threshold pair is then utilized Continuous Similarity value region carries out area size screening, the continuous Similarity value region contour after must screening；And then obtain to be measured Subject matter profile in image.

The image-recognizing method and system of the present invention, can be applied in multiple directions.Including but not limited to：For image Image-recognizing method and system, the autofocus control method and system of definition judgment.

As shown in figure 11, present invention also offers a kind of image-recognizing method, this method to comprise the following steps：

S11, image recognizing step, using any of the above-described kind of image-recognizing method, identify the subject matter wheel in multiple sequencing images It is wide；

S12, definition judgment step, the subject matter quantity each contained in the multiple sequencing image is counted, then according to mark The number of object area quantity determine definition highest sequencing image in the multiple sequencing image.

Surveyed it should be noted that the multiple sequencing image is the high flux gene shot to same target area Sequence image.The method of the present invention, can be quick to be identified the image containing most subject matters as definition highest image Effective relatively shot to same target area and multiple sequencing images definition.

As shown in figure 12, present invention also offers a kind of image identification system 200, described image identifying system 200 and figure As the difference of identifying system 100 is, image identification system 200 further comprises definition judgment module 220, for uniting The subject matter region quantity each contained in the multiple sequencing image is counted, then according to the number of subject matter quantity to determine State definition highest sequencing image in multiple sequencing images.

Surveyed it should be noted that the multiple sequencing image is the high flux gene shot to same target area Sequence image.In the present embodiment, image identification system 100 is the equal of a picture recognition module, the system of the present embodiment with It is definition highest image to be identified the image containing most subject matters, can fast and effectively be compared to same target area Domain shot and multiple sequencing images definition.

As described in Figure 13, present invention also offers a kind of autofocus control method, comprise the following steps：

S21, focusing step, repeatedly adjust imaging device and adopt the distance of figure position, record the focal position after regulation every time, will Collection image command, which is issued, adopts module, after instruction of the focusing to best focus position is received, regulation imaging device movement To best focus position；

S22, figure step is adopted, after collection image command is received, gather image using imaging device, obtain different focal positions Multiple sequencing images；

S23, image processing step, after multiple sequencing images of the different focal positions are received, using above-mentioned image Recognition methods, determine definition highest sequencing image in the multiple sequencing image, using definition highest sequencing image as Best focus position, and focusing module is issued into the instruction focused to best focus position.

In the control method of the present embodiment, after multiple sequencing images are received, by using above-mentioned image recognition side Method, definition highest sequencing image can be fast and effectively compared, and then judge best focus position, so as to focus on automatically To best focus position.

As described in Figure 14, present invention also offers a kind of auto focus control system 300, including focusing module 310, figure is adopted Module 320 and image processing module 330；

The focusing module 310, for repeatedly adjusting imaging device and adopting the distance of figure position, record the focusing after regulation every time Position, collection image command is issued and adopts module, after instruction of the focusing to best focus position is received, be adjusted to as dress Put and be moved to best focus position；

It is described to adopt module 320, for after collection image command is received, image to be gathered using imaging device, obtain different poly- Multiple sequencing images of burnt position；

Described image processing module 330, for after multiple sequencing images of the different focal positions are received, control to be above-mentioned Image identification system 200, definition highest sequencing image in the multiple sequencing image is determined, be sequenced with definition highest Image is best focus position, and the instruction focused to best focus position is issued into focusing module.

In the control system of the present embodiment, the focusing module by repeatedly adjust imaging device and adopt figure position away from From adopting module control imaging device and collect multiple sequencing images of different focal positions；Image processing module is receiving To after multiple sequencing images of the different focal positions, by controlling above-mentioned comparison system, can fast and effectively compare Go out definition highest sequencing image, and then judge best focus position, so as to focus to best focus position automatically.

The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention All any modification, equivalent and improvement made within refreshing and principle etc., should be included in the scope of the protection.

Claims

1. a kind of image-recognizing method, it is characterised in that the described method comprises the following steps：

2. image-recognizing method according to claim 1, it is characterised in that also obtain step including array of templates：According to Gauss nuclear operator, which calculates, obtains array of templates；Or processing acquisition is carried out to the numerical matrix of subject matter region in real image Array of templates.

3. image-recognizing method according to claim 1, it is characterised in that described image matching step includes following step Suddenly：

4. image-recognizing method according to claim 1, it is characterised in that described image matching method is normalized crosscorrelation Matching algorithm.

5. image-recognizing method according to claim 1, it is characterised in that also including the second filtration step：Utilize signal Threshold value carries out the second filtering to each pixel in testing image, obtains the second filtering image；

6. image-recognizing method according to claim 1, it is characterised in that the region recognition step is specifically by area Domain growth method obtains the continuous Similarity value region contour in the first filtering image, while using region threshold to continuous similarity It is worth region and carries out area size screening, the continuous Similarity value region contour after must screening；And then obtain the mark in testing image Thing profile.

7. the image-recognizing method according to any one of claim 1 to 6, it is characterised in that the testing image is The gray level image of high flux gene sequencing image, the subject matter are sphere.

A kind of 8. image identification system, it is characterised in that including：

9. image identification system according to claim 8, it is characterised in that module also is obtained including array of templates, according to Gauss nuclear operator, which calculates, obtains array of templates；Or processing acquisition is carried out to the numerical matrix of subject matter region in real image Array of templates.

10. image identification system according to claim 8, it is characterised in that described image matching module includes normalization Unit and the first matching unit；

11. image identification system according to claim 8, it is characterised in that described image matching method is mutual for normalization Close matching algorithm.

12. image identification system according to claim 8, it is characterised in that also including the second filtering module, for utilizing Signal threshold value carries out the second filtering to each pixel in testing image, obtains the second filtering image；

13. image identification system according to claim 8, it is characterised in that the region identification module, for passing through area Domain growth method obtains the continuous Similarity value region contour in the first filtering image, while using region threshold to continuous similarity It is worth region and carries out area size screening, the continuous Similarity value region contour after must screening；And then obtain the mark in testing image Thing profile.

14. the image identification system according to any one of claim 8 to 13, it is characterised in that the testing image For the gray level image of high flux gene sequencing image, the subject matter is sphere.

15. a kind of image-recognizing method, it is characterised in that comprise the following steps：

Image recognizing step, the mark in multiple sequencing images is identified using the method any one of claim 1 to 7 respectively Thing profile；

A kind of 16. image identification system, it is characterised in that including：

Picture recognition module, identified respectively in multiple sequencing images using the system any one of claim 8 to 14 Subject matter profile；

17. a kind of autofocus control method, it is characterised in that comprise the following steps：

Image processing step, after multiple sequencing images of the different focal positions are received, using described in claim 15 Image-recognizing method, determine definition highest sequencing image in the multiple sequencing image, with definition highest be sequenced Image is best focus position, and the instruction focused to best focus position is issued into focusing module.

A kind of 18. auto focus control system, it is characterised in that including：

Image processing module, for after multiple sequencing images of the different focal positions are received, using claim 16 Described image identification system, definition highest sequencing image in the multiple sequencing image is determined, with definition highest Sequencing image is best focus position, and the instruction focused to best focus position is issued into focusing module.