CN105760878A - Method and device for selecting urinary sediment microscope image with optimal focusing performance - Google Patents
Method and device for selecting urinary sediment microscope image with optimal focusing performance Download PDFInfo
- Publication number
- CN105760878A CN105760878A CN201410805362.8A CN201410805362A CN105760878A CN 105760878 A CN105760878 A CN 105760878A CN 201410805362 A CN201410805362 A CN 201410805362A CN 105760878 A CN105760878 A CN 105760878A
- Authority
- CN
- China
- Prior art keywords
- image
- urine
- sum
- window
- width
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The invention relates to a method and device for selecting a urinary sediment microscope image with optimal focusing performance. The method includes the following steps that: a plurality of urinary sediment images of one urine sediment sample are acquired through the lens of a fine-tuning microscope; all urine cells in each urinary sediment image are detected, high-pass filtering is performed on each detected urine cell, so that the spectrum energy of the urine cells can be obtained, and the mean value and variance of the spectrum energy of each image are calculated; and based on a rule that the larger the mean value of spectrum energy is, the smaller the variance of the spectrum energy is, and the better the focusing performance of an image is, an image with optimal focusing is selected from the plurality of urinary sediment images. With the method and device of the invention adopted, it can be ensured that a finally-selected image is an image with optimal focusing.
Description
Technical field
The application relates to image analysis technology field, particularly to method and the device of the best Urine sediments analyzer MIcrosope image of selective focus.
Background technology
Urinalysis is a common and important test.After acquiring Urine sediments analyzer sample, urine sediment image to be gathered under the microscope, its purpose is to clearly show urine thin born of the same parents.In order to ensure to collect urine sediment image clearly, generally to a Urine sediments analyzer sample, can by finely tune microscopical camera lens collect several urine sediment images, then artificial selection go out the most clearly image to carry out next step detection analysis.
Artificial selection's urine sediment image the most clearly can be easily subject to subjective impact.
Summary of the invention
The embodiment of the present application provides method and the device of the Urine sediments analyzer MIcrosope image that selective focus is best.
In order to achieve the above object, this application provides following technical scheme:
The method of the Urine sediments analyzer MIcrosope image that a kind of selective focus is best, including:
Several urine sediment images of a Urine sediments analyzer sample are gathered by finely tuning microscopical camera lens；
For each width urine sediment image, detect all urine thin born of the same parents in this width image, detected each urine thin born of the same parents are carried out high-pass filtering, obtain the spectrum energy of this urine thin born of the same parents, and calculate spectrum energy average and the variance of this width image；
More big according to spectrum energy average and variance is more little, the principle that image focusing is more good, select a width to focus on best image in several urine sediment images described.
In the above-mentioned methods, calculate spectrum energy average and the variance of every piece image, then more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, several MIcrosope image described select a width focus on best image, it is ensured that the final image selected is focus on best image.
In one embodiment, farther include before all urine thin born of the same parents in this width image of described detection: gather prospect sample set and background sample set in advance, wherein, prospect sample is the image only comprising urine thin born of the same parents, and background sample is do not comprise urine thin born of the same parents only to comprise the image of background；Extract from each prospect sample and each background sample respectively and preset feature；Adopt default training algorithm that foreground features collection and background characteristics collection are trained, obtain distinguishing the grader of foreground image and background image.
All urine thin born of the same parents in this width image of described detection include: slided on this width urine sediment image to preset the first step-length by the window with foreground and background sample formed objects, for the window after sliding every time, adopt the feature extracting method identical with training process to gather from this window region and preset feature, adopting described grader to distinguish this window region is prospect or background, if prospect, then this window region is labeled as urine thin born of the same parents.
According to a kind of embodiment, described respectively from each prospect sample and each background sample extract preset feature include: for every width prospect or background sample image, the window of default a*b size is slided on this sample image to preset the second step-length, wherein, a=A/m1, the height of b=B/m2, A, B respectively prospect or background sample image and width, m1, m2 are the integer more than 1；
For the window region after sliding every time, extract following four category features:
Feature one, this window region is divided into left and right two parts, feature one=fabs (sum (C1)-sum (D1))/512,
Wherein, C1 represents the left-half of this window region, and sum (C1) represents that D1 represents the right half part of this window region to all pixels summation in C1, sum (D1) represents all pixels in D1 is sued for peace, and fabs represents and takes absolute value；
Feature two, this window region is divided into upper and lower two parts, feature two=fabs (sum (C2)-sum (D2))/512,
Wherein, C2 represents the first half of this window region, and sum (C2) represents that D2 represents the latter half of this window region to all pixels summation in C2, and sum (D2) represents all pixels summation in D2；
Feature three, this window region is from left to right divided into three parts, if respectively C3, D3, E3, wherein, / 4th of the width of C3=this window width, the half of the width of D3=this window width, / 4th of the width of E3=this window width, feature three=fabs (sum (C3)+sum (E3)-sum (D3))/512
Wherein, sum (C3) represents all pixels summation in C3, and sum (D3) represents that all pixels in E3 are sued for peace by sum (E3) expression to all pixels summation in D3；
Feature four, this window region is divided into three parts from top to bottom, if respectively C4, D4, E4, wherein, / 4th of the height of C4=this window height, the half of the height of D4=this window height, / 4th of the height of E4=this window height, feature four=fabs (sum (C4)+sum (E4)-sum (D4))/512
Wherein, sum (C4) represents all pixels summation in C4, and sum (D4) represents that all pixels in E4 are sued for peace by sum (E4) expression to all pixels summation in D4.
In one embodiment, described detected each urine thin born of the same parents are carried out high-pass filtering include: for each urine thin born of the same parents, the window of 8*8 size is slided in this urine thin born of the same parents region to preset the 3rd step-length, for the window after sliding every time, the convolution kernel of the pixel value in this window region Yu default 8*8 size is carried out convolution algorithm, obtains the spectrum energy of this window region；When window slides end in this urine thin born of the same parents region, the spectrum energy of the window region after slip every time is added, obtains the spectrum energy of this urine thin born of the same parents.Wherein, default convolution kernel is: the coefficient in the region of central authorities 4*4 is all+3, and the coefficient in remaining area is all-1.
According to a kind of embodiment, described more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, several urine sediment images described select a width focus on best image to include: when calculating spectrum energy average and the variance of all urine sediment images, judge that whether the maximum spectrum energy average difference with secondary big spectrum energy average is more than preset value, if so, then using image corresponding for maximum spectrum energy average as focus on optimized image；Otherwise, relatively spectrum energy variance maximum, that secondary big spectrum energy average is corresponding, using image corresponding for wherein less spectrum energy variance as focusing on optimized image.
The embodiment of the present application additionally provides the device of the best Urine sediments analyzer MIcrosope image of a kind of selective focus, including:
One urine thin born of the same parents' detection module, for detecting all urine thin born of the same parents in each width urine sediment image；Described urine sediment image is for same Urine sediments analyzer sample, by finely tuning the arbitrary width in several urine sediment images of this sample that microscopical camera lens collects；
One spectrum energy computing module, for for each urine thin born of the same parents in each width urine sediment image, carrying out high-pass filtering to detected each urine thin born of the same parents, obtain the spectrum energy of this urine thin born of the same parents, and calculate spectrum energy average and the variance of this width image；
One Image selection module, is used for more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, selects a width to focus on best image in several urine sediment images described.
In said apparatus, several MIcrosope image for same Urine sediments analyzer sample, by detecting the urine thin born of the same parents in every piece image, and detected each urine thin born of the same parents are carried out high-pass filtering, obtain the spectrum energy of this urine thin born of the same parents, and then calculate spectrum energy average and the variance of every piece image, then more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, several MIcrosope image described select a width focus on best image, it is ensured that the final image selected is focus on best image.
According to a kind of embodiment, described urine thin born of the same parents' detection module includes:
One training module, for gathering prospect sample set and background sample set in advance, wherein, prospect sample is the image only comprising urine thin born of the same parents, and background sample is do not comprise urine thin born of the same parents only to comprise the image of background；Extract from each prospect sample and each background sample respectively and preset feature；Adopt default training algorithm that foreground features collection and background characteristics collection are trained, obtain distinguishing the grader of foreground image and background image；
One detection module, for for each width urine sediment image, the window of the foreground and background sample formed objects gathered with training module is slided on this width urine sediment image to preset the first step-length, for the window after sliding every time, adopt the feature extracting method identical with training module to gather from this window region and preset feature, adopting described grader to distinguish this window region is prospect or background, if prospect, then this window region is labeled as urine thin born of the same parents.
According to a kind of embodiment, described training module extracts default feature respectively from each prospect sample and each background sample and includes:
For the window region after sliding every time, extract following four category features:
Feature one, this window region is divided into left and right two parts, feature one=fabs (sum (C1)-sum (D1))/512,
Wherein, C1 represents the left-half of this window region, and sum (C1) represents that D1 represents the right half part of this window region to all pixels summation in C1, sum (D1) represents all pixels in D1 is sued for peace, and fabs represents and takes absolute value；
Feature two, this window region is divided into upper and lower two parts, feature two=fabs (sum (C2)-sum (D2))/512,
Wherein, C2 represents the first half of this window region, and sum (C2) represents that D2 represents the latter half of this window region to all pixels summation in C2, and sum (D2) represents all pixels summation in D2；
Feature three, this window region is from left to right divided into three parts, if respectively C3, D3, E3, wherein, / 4th of the width of C3=this window width, the half of the width of D3=this window width, / 4th of the width of E3=this window width, feature three=fabs (sum (C3)+sum (E3)-sum (D3))/512
Wherein, sum (C3) represents all pixels summation in C3, and sum (D3) represents that all pixels in E3 are sued for peace by sum (E3) expression to all pixels summation in D3；
Feature four, this window region is divided into three parts from top to bottom, if respectively C4, D4, E4, wherein, / 4th of the height of C4=this window height, the half of the height of D4=this window height, / 4th of the height of E4=this window height, feature four=fabs (sum (C4)+sum (E4)-sum (D4))/512
Wherein, sum (C4) represents all pixels summation in C4, and sum (D4) represents that all pixels in E4 are sued for peace by sum (E4) expression to all pixels summation in D4.
In one embodiment, each urine thin born of the same parents in each width urine sediment image are carried out high-pass filtering and include by described spectrum energy computing module: for each urine thin born of the same parents, the window of 8*8 size is slided in this urine thin born of the same parents region to preset the 3rd step-length, for the window after sliding every time, the convolution kernel of the pixel value in this window region Yu default 8*8 size is carried out convolution algorithm, obtains the spectrum energy of this window region；When window slides end in this urine thin born of the same parents region, the spectrum energy of the window region after slip every time is added, obtains the spectrum energy of this urine thin born of the same parents.Wherein, default convolution kernel is: the coefficient in the region of central authorities 4*4 is all+3, and the coefficient in remaining area is all-1.
According to a kind of embodiment, described Image selection module is more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, several urine sediment images described select a width focus on best image to include: the spectrum energy average of all urine sediment images for same Urine sediments analyzer sample calculated for spectrum energy computing module and variance, judge that whether the maximum spectrum energy average difference with secondary big spectrum energy average is more than preset value, if so, then using image corresponding for maximum spectrum energy average as focus on optimized image；Otherwise, relatively spectrum energy variance maximum, that secondary big spectrum energy average is corresponding, using image corresponding for wherein less spectrum energy variance as focusing on optimized image.
Accompanying drawing explanation
The method flow diagram of the Urine sediments analyzer MIcrosope image of selective focus the best that Fig. 1 provides for the application one embodiment；
The method flow diagram of the Urine sediments analyzer MIcrosope image of selective focus the best that Fig. 2 provides for another embodiment of the application；
The exemplary plot from prospect sample or background sample extraction feature one that Fig. 3 A the embodiment of the present application provides；
The exemplary plot from prospect sample or background sample extraction feature two that Fig. 3 B provides for the embodiment of the present application；
The exemplary plot from prospect sample or background sample extraction feature three that Fig. 3 C provides for the embodiment of the present application；
The exemplary plot from prospect sample or background sample extraction feature four that Fig. 3 D provides for the embodiment of the present application；
The schematic diagram of the convolution kernel of the 8*8 that Fig. 4 provides for the embodiment of the present application；
Fig. 5 is the foreground detection result schematic diagram of the width Urine sediments analyzer MIcrosope image in the application application example for a Urine sediments analyzer sample；
The composition schematic diagram of the device of the Urine sediments analyzer MIcrosope image of selective focus the best that Fig. 6 provides for the embodiment of the present application.
Wherein, accompanying drawing labelling is as follows:
101: for the Urine sediments analyzer sample collected, gather several urine sediment images of this sample by finely tuning microscopical camera lens.
102: for each width urine sediment image, detect all urine thin born of the same parents in this width image.
103: for each urine thin born of the same parents detected, to this urine thin, born of the same parents carry out high-pass filtering, obtain the spectrum energy of this urine thin born of the same parents.
104: the spectrum energy according to urine thin born of the same parents each in this width urine sediment image, calculate spectrum energy average and the variance of this width image.
105: more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, for this Urine sediments analyzer sample collection to all urine sediment images in select a width to focus on best image, in order to carry out next step detection and analyze.
201: for the Urine sediments analyzer sample collected, gather several urine sediment images of this sample by finely tuning microscopical camera lens.
202: for each width urine sediment image, detect all urine thin born of the same parents in this image, wherein, each urine thin born of the same parents define with a rectangle frame presetting the first size.
203: for each urine thin born of the same parents, the window that will preset 8 (pixel) * 8 (pixel) size slides on this urine thin born of the same parents to preset the 3rd step-length, respectively the pixel value in the window region after slip every time is carried out convolution algorithm with default convolution kernel, obtain the spectrum energy of this window region.
204: for each urine thin born of the same parents, the spectrum energy of the window region after all slips obtain 8*8 window in sliding process in this urine thin born of the same parents is added, and obtains the spectrum energy of this urine thin born of the same parents.
205: for every width urine sediment image, the spectrum energy according to urine thin born of the same parents all in this image, calculate spectrum energy average and the variance of this image.
206: more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, select a width to focus on best image in all urine sediment images that step 201 collects.
207: adopt the image focusing on the best to carry out next step detection and analyze.
A, b: the height of window and width
C1, D1: the left-half of window region, right half part
C2, D2: the first half of window region, the latter half
C3, D3, E3: window region three parts from left to right
C4, D4, E4: window region three parts from top to bottom
61: urine thin born of the same parents' detection module
62: spectrum energy computing module
63: Image selection module
Detailed description of the invention
In order to make the purpose of the application, technical scheme and advantage clearly understand, below in conjunction with accompanying drawing and according to embodiment, the technical scheme of the application is described in detail.
The method flow diagram of the Urine sediments analyzer MIcrosope image of selective focus the best that Fig. 1 provides for the application one embodiment, it specifically comprises the following steps that
Step 101: for the Urine sediments analyzer sample collected, gather several urine sediment images of this sample by finely tuning microscopical camera lens.
Here, the rank of " fine setting " is usually micron-sized.
Step 102: for each width urine sediment image, detect all urine thin born of the same parents in this width image.
Step 103: for each urine thin born of the same parents detected, born of the same parents carry out high-pass filtering to this urine thin, obtain the spectrum energy of this urine thin born of the same parents.
Step 104: the spectrum energy according to urine thin born of the same parents each in this width urine sediment image, calculates spectrum energy average and the variance of this width image.
Step 105: more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, for this Urine sediments analyzer sample collection to all urine sediment images in select a width to focus on best image, in order to carry out next step detection and analyze.
The method flow diagram of the Urine sediments analyzer MIcrosope image of selective focus the best that Fig. 2 provides for another embodiment of the application, it specifically comprises the following steps that
Step 201: for the Urine sediments analyzer sample collected, gather several urine sediment images of this sample by finely tuning microscopical camera lens.
Step 202: for each width urine sediment image, detect all urine thin born of the same parents in this image, wherein, each urine thin born of the same parents define with a rectangle frame presetting the first size.
Before urine thin born of the same parents in detection urine sediment image, first carrying out a training process, to obtain distinguishing the grader of urine thin born of the same parents and background, training process is specific as follows:
Step 01: gather prospect sample set and background sample set.
Namely prospect sample only comprises the image of urine thin born of the same parents, and namely background sample does not comprise urine thin born of the same parents and only comprise the image of background.Prospect sample must be identical with background size, is set to A*B, for instance: A=B=96 pixel.
Step 02: extract from each prospect sample and each background sample respectively and preset feature.
The process presetting feature from each prospect sample or background sample extraction is as follows:
1) window of default a (pixel) * b (pixel) size is slided on sample image to preset the first step-length；
A=A/m1, b=B/m2, m1, m2 are the integer more than 1, for instance: m1=m2=3, default first step-length here can be empirically determined, for instance: the first step-length is all 2 pixels in level, vertical direction.Namely when window level is slided, 2 row of being separated by between adjacent twice slip, when window vertical sliding motion, 2 row of being separated by between adjacent twice slip.
2) for the window region after sliding every time, following four category features are extracted:
Feature one, as shown in Figure 3A, is divided into left and right two parts, feature one=fabs (sum (C1)-sum (D1))/512 by window region.
Wherein, C1 represents the left-half of window region, sum (C1) represents all pixels summation in the left-half of window region, D1 represents the right half part of window region, sum (D1) represents all pixels in the right half part of window region is sued for peace, and fabs represents and takes absolute value.
Feature two, as shown in Figure 3 B, is divided into upper and lower two parts by window region, feature two=fabs (sum (C2)-sum (D2))/512,
Wherein, C2 represents the first half of window region, sum (C2) represents all pixels summation in the first half of window region, D2 represents the latter half of window region, sum (D2) represents all pixels in the latter half of window region is sued for peace, and fabs represents and takes absolute value.
Feature three, as shown in Figure 3 C, window region is from left to right divided into three parts, if respectively C3, D3, E3, wherein, / 4th i.e. b/4 of the width=window width of C3, / 4th i.e. b/4 of the half of the width=window width of D3 and the width=window width of b/2, E3, feature three=fabs (sum (C3)+sum (E3)-sum (D3))/512.
Sum (C3) represents all pixels in C3 is sued for peace, and sum (D3) represents all pixels in D3 are sued for peace, and sum (E3) represents all pixels in E3 are sued for peace, and fabs represents and takes absolute value.
Feature four, as shown in Figure 3 D, window region is divided into three parts from top to bottom, if respectively C4, D4, E4, wherein, / 4th i.e. a/4 of the height=window height of C4, / 4th i.e. a/4 of the half of the height=window height of D4 and the height=window height of a/2, E4, feature four=fabs (sum (C4)+sum (E4)-sum (D4))/512.
Sum (C4) represents all pixels in C4 is sued for peace, and sum (D4) represents all pixels in D4 are sued for peace, and sum (E4) represents all pixels in E4 are sued for peace, and fabs represents and takes absolute value.
Step 03: adopt the default training algorithm feature set to extracting from prospect sample set and the feature set extracted from background sample set to be trained, obtain distinguishing the grader of foreground image and background image.
Training algorithm is highly developed algorithm in the prior art, for instance can adopt adaboost algorithm.
After obtaining grader, it is possible to urine sediment image being carried out urine thin born of the same parents and have detected, detailed process is as follows:
Step 04: for every width urine sediment image, slides to preset the second step-length on this image by the window of size A*B, for the window region after sliding every time, adopts the method in step 02 to extract feature from this window region.
Step 05: adopting the grader that step 03 obtains to distinguish this window region is prospect or background, if prospect, then this window region of labelling is urine thin born of the same parents on image.
It can be seen that in step 202 rectangle frame be sized to A*B.
Step 203: for each urine thin born of the same parents, the window that will preset 8 (pixel) * 8 (pixel) size slides on this urine thin born of the same parents to preset the 3rd step-length, respectively the pixel value in the window region after slip every time is carried out convolution algorithm with default convolution kernel, obtain the spectrum energy of this window region.
The 3rd step-length in this step is generally horizontal, vertical direction is all 4 pixels.
Convolution kernel is as shown in Figure 4, convolution kernel be sized to 8*8, wherein, coefficient in the region of central authorities 4*4 is all+3, coefficient in remaining area is all-1, and 4*4=16 pixel of the central authorities being about to the window region after every time sliding is multiplied by 3 respectively, and respectively 8*8-4*4=48 pixel of remaining four edges is multiplied by-1, then by all of product addition, obtain and value is the spectrum energy of this window region.
Above-mentioned convolution kernel is equivalent to the center superposition of two square function boxes.Wherein, a square function box be sized to 8*8, multiplier is-1；Another square function box be sized to 4*4, multiplier is+4 (when, after the square function box superposition of the two, the multiplier after the 4*4 region superposition at center is (+4)+(-1)=+ 3.
The window region of 8*8 and this convolution kernel being carried out convolution, is substantially equivalent to adopt wave filter that the window region of this 8*8 is filtered, this wave filter is high pass filter, and its result is the spectrum energy of this window region.
Step 204: for each urine thin born of the same parents, the spectrum energy of the window region after all slips obtain 8*8 window in sliding process in this urine thin born of the same parents is added, and obtains the spectrum energy of this urine thin born of the same parents.
Step 205: for every width urine sediment image, the spectrum energy according to urine thin born of the same parents all in this image, calculate spectrum energy average and the variance of this image.
The absolute value of the spectrum energy of urine thin born of the same parents all in this image is added, then by with value divided by the urine total cellular score in this image, obtain the spectrum energy average of this image.
Respectively the spectrum energy of each urine thin born of the same parents is deducted the spectrum energy average of this image, the each difference obtained is carried out square operation respectively, all of square operation result is sued for peace, by with divided by the urine total cellular score in this image, quotient is opened radical sign computing, finally gives the spectrum energy variance of this image.
Step 206: more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, selects a width to focus on best image in all urine sediment images that step 201 collects.
The spectrum energy average of image is more big, represents that the focusing quality of image is more good；The spectrum energy variance of image is more little, illustrates that the urine thin born of the same parents being positioned at focus in image are more many.
When calculating spectrum energy average and the variance of all images, it is judged that the difference of maximum spectrum energy average and time big spectrum energy average whether more than preset value, if so, then using image corresponding for maximum spectrum energy average as focusing on optimized image；Otherwise, relatively spectrum energy variance maximum, that secondary big spectrum energy average is corresponding, using image corresponding for less spectrum energy variance as focusing on optimized image.
Step 207: adopt the image focusing on the best to carry out next step detection and analyze.
The application example of the application given below:
Process one, foreground detection training process
This process is referring to step 202.Wherein, prospect sample and background size are all 96 (pixel) * 96 (pixel).
When from prospect sample or background sample extract preset feature time, sliding window be sized to 32 (pixel) * 32 (pixel), sliding step is all 2 pixels in level, vertical direction, and the training algorithm of employing is adaboost algorithm.
Below for the process of the Urine sediments analyzer MIcrosope image of selective focus the best:
Step 01: to a Urine sediments analyzer sample, by finely tuning microscopical camera lens, (q is integer to collect q, and q > 1) width image, the size of each image is m*n (m, n are integer, and m > 1, n > 1), wherein, m is the line number of image, and n is the columns of image.
In this example, m=2592 (pixel), n=2048 (pixel).
The rank of fine setting is micron.
For each image, perform following steps 02～05 respectively.
Step 02: for current width image, the window of 96 (pixel) * 96 (pixel) is slided on this image with horizontal step-length 16 (pixel), vertical step-length 16 (pixel), for the window region after sliding every time, adopt with the identical feature extraction mode of the process of training from this extracted region feature, adopting the grader that training process obtains to distinguish this region is prospect or background, if prospect, then on image, this region of labelling is ROI (RegionofInterest, area-of-interest).
Fig. 5 is the schematic diagram of the foreground detection result of the width Urine sediments analyzer MIcrosope image for a Urine sediments analyzer sample.It should be noted that Fig. 5 is a width schematic diagram, in actual applications, if the focusing of image is preferably, then urine thin born of the same parents can be more visible, and otherwise, urine thin born of the same parents can be fuzzyyer.
Step 03: when current width display foreground is detected complete, ROI for each 96*96, the window being sized to 8*8 is slided on this ROI with horizontal step-length 4, vertical step-length 4, respectively the convolution kernel of the 8*8 window region after slip every time with default 8*8 is carried out convolution algorithm, obtain the spectrum energy of this 8*8 window region.
Convolution kernel is as shown in Figure 4.
Step 04: for each ROI of current width image, the spectrum energy in all 8*8 regions obtained in 8*8 window sliding process is added, obtains the spectrum energy of this ROI.
Step 05: for current width image, the spectrum energy according to all ROI, calculate spectrum energy average and the variance of current width image.
The spectrum energy average of current width image can be represented by equation below:
The spectrum energy variance of current width image can be represented by equation below:
Wherein, M is the sum of the ROI that current width image comprises, fiFor the spectrum energy of i-th ROI, 1≤i≤M, ave is the spectrum energy average of current width image, and var is the spectrum energy variance of current width image.
Step 06: when calculating ave and the var of all images, selects maximum ave wherein, it is judged that the difference of maximum ave and time big ave whether more than preset value, if so, then using image corresponding for maximum ave as focusing on best image, go to step 08；Otherwise, step 07 is performed.
Step 07: select less var in var corresponding for maximum ave and time big var corresponding for ave, using image corresponding for less var as focusing on best image.
Step 08: carry out next step detection analysis to focusing on best image.
It should be noted that, due to when for several MIcrosope image of Urine sediments analyzer sample collection, simply microscopical camera lens is finely tuned with micron order, therefore, the visual field of these several MIcrosope image is regarded as essentially identical, in this way it can be considered that, the position that prospect (namely comprises the ROI of urine thin born of the same parents) is all identical on all images.Thus, in order to reduce the operand of foreground detection, only need to detect for wherein piece image, then the foreground location detected on this width image is copied directly on other image.
The composition schematic diagram of the device of the Urine sediments analyzer MIcrosope image of selective focus the best that Fig. 6 provides for the embodiment of the present application, it specifically includes that 61, spectrum energy computing module 62 of urine thin born of the same parents' detection module and an Image selection module 63, wherein:
Urine thin born of the same parents' detection module 61, for detecting all urine thin born of the same parents in each width urine sediment image；Described urine sediment image is for same Urine sediments analyzer sample, by finely tuning the arbitrary width in several urine sediment images of this sample that microscopical camera lens collects.
Spectrum energy computing module 62, each urine thin born of the same parents being used in each width urine sediment image that urine thin born of the same parents' detection module 61 is detected carry out high-pass filtering, obtain the spectrum energy of this urine thin born of the same parents, and calculate spectrum energy average and the variance of this width urine sediment image.
Image selection module 63, spectrum energy average and variance for several urine sediment images described in calculating according to spectrum energy computing module 62, and it is more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, selects a width to focus on best image in several urine sediment images described.
Preferably, urine thin born of the same parents' detection module 61 includes a training module and a detection module, wherein:
Training module, for gathering prospect sample set and background sample set in advance, wherein, prospect sample is the image only comprising urine thin born of the same parents, and background sample is do not comprise urine thin born of the same parents only to comprise the image of background；Extract from each prospect sample and each background sample respectively and preset feature；Adopt default training algorithm that foreground features collection and background characteristics collection are trained, obtain distinguishing the grader of foreground image and background image.
Detection module, for for each width urine sediment image, the window of the foreground and background sample formed objects gathered with training module is slided on this width urine sediment image to preset the first step-length, for the window after sliding every time, adopt the feature extracting method identical with training module to gather from this window region and preset feature, adopting described grader to distinguish this window region is prospect or background, if prospect, then this window region is labeled as urine thin born of the same parents.
Preferably, training module respectively from each prospect sample and each background sample extract preset feature include:
For the window region after sliding every time, extract following four category features:
Feature one, this window region is divided into left and right two parts, feature one=fabs (sum (C1)-sum (D1))/512,
Wherein, C1 represents the left-half of this window region, and sum (C1) represents that D1 represents the right half part of this window region to all pixels summation in C1, sum (D1) represents all pixels in D1 is sued for peace, and fabs represents and takes absolute value；
Feature two, this window region is divided into upper and lower two parts, feature two=fabs (sum (C2)-sum (D2))/512,
Wherein, C2 represents the first half of this window region, and sum (C2) represents that D2 represents the latter half of this window region to all pixels summation in C2, and sum (D2) represents all pixels summation in D2；
Feature three, this window region is from left to right divided into three parts, if respectively C3, D3, E3, wherein, / 4th of the width of C3=this window width, the half of the width of D3=this window width, / 4th of the width of E3=this window width, feature three=fabs (sum (C3)+sum (E3)-sum (D3))/512
Wherein, sum (C3) represents all pixels summation in C3, and sum (D3) represents that all pixels in E3 are sued for peace by sum (E3) expression to all pixels summation in D3；
Feature four, this window region is divided into three parts from top to bottom, if respectively C4, D4, E4, wherein, / 4th of the height of C4=this window height, the half of the height of D4=this window height, / 4th of the height of E4=this window height, feature four=fabs (sum (C4)+sum (E4)-sum (D4))/512
Wherein, sum (C4) represents all pixels summation in C4, and sum (D4) represents that all pixels in E4 are sued for peace by sum (E4) expression to all pixels summation in D4.
Preferably, each urine thin born of the same parents in each width urine sediment image are carried out high-pass filtering and include by spectrum energy computing module 62:
For each urine thin born of the same parents, the window of 8*8 size is slided in this urine thin born of the same parents region to preset the 3rd step-length, for the window after sliding every time, the convolution kernel of the pixel value in this window region Yu default 8*8 size is carried out convolution algorithm, obtains the spectrum energy of this window region；
When window slides end in this urine thin born of the same parents region, the spectrum energy of the window region after slip every time is added, obtains the spectrum energy of this urine thin born of the same parents,
Wherein, default convolution kernel is: the coefficient in the region of central authorities 4*4 is all+3, and the coefficient in remaining area is all-1.
Preferably, spectrum energy computing module calculates the spectrum energy average of this width image and variance is:
Wherein, M is the sum of the urine thin born of the same parents that this width image comprises, fiFor the spectrum energy of i-th urine thin born of the same parents, 1≤i≤M, ave is the spectrum energy average of this width image, and var is the spectrum energy variance of this width image, | | represent and take absolute value.
Preferably, Image selection module 63 is more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, selects a width to focus on best image and include in several urine sediment images described:
When spectrum energy computing module 62 calculates spectrum energy average and the variance of all urine sediment images for same Urine sediments analyzer sample, Image selection module 63 judges that whether the maximum spectrum energy average difference with secondary big spectrum energy average is more than preset value, if so, then using image corresponding for maximum spectrum energy average as focus on optimized image；Otherwise, relatively spectrum energy variance maximum, that secondary big spectrum energy average is corresponding, using image corresponding for wherein less spectrum energy variance as focusing on optimized image.
The foregoing is only the preferred embodiment of the application, not in order to limit the application, all within spirit herein and principle, any amendment of making, equivalent replacements, improvement etc., should be included within the scope that the application protects.
Claims (10)
1. a method for the Urine sediments analyzer MIcrosope image that selective focus is best, including:
Several urine sediment images of a Urine sediments analyzer sample are gathered by finely tuning microscopical camera lens；
For each width urine sediment image, detect all urine thin born of the same parents in this width image, detected each urine thin born of the same parents are carried out high-pass filtering, obtain the spectrum energy of this urine thin born of the same parents, and calculate spectrum energy average and the variance of this width image；
More big according to spectrum energy average and variance is more little, the principle that image focusing is more good, select a width to focus on best image in several urine sediment images described.
2. method according to claim 1, it is characterised in that
Farther include before all urine thin born of the same parents in this width image of described detection: gathering prospect sample set and background sample set in advance, wherein, prospect sample is the image only comprising urine thin born of the same parents, and background sample is do not comprise urine thin born of the same parents only to comprise the image of background；Extract from each prospect sample and each background sample respectively and preset feature；Adopt default training algorithm that foreground features collection and background characteristics collection are trained, obtain distinguishing the grader of foreground image and background image；
All urine thin born of the same parents in this width image of described detection include: slided on this width urine sediment image to preset the first step-length by the window with foreground and background sample formed objects, for the window after sliding every time, adopt the feature extracting method identical with training process to gather from this window region and preset feature, adopting described grader to distinguish this window region is prospect or background, if prospect, then this window region is labeled as urine thin born of the same parents.
3. method according to claim 2, it is characterised in that described respectively from each prospect sample and each background sample extract preset feature include:
For every width prospect or background sample image, the window of default a*b size is slided on this sample image to preset the second step-length, wherein, a=A/m1, the height of b=B/m2, A, B respectively prospect or background sample image and width, m1, m2 are the integer more than 1；
For the window region after sliding every time, extract following four category features:
Feature one, this window region is divided into left and right two parts, feature one=fabs (sum (C1)-sum (D1))/512,
Wherein, C1 represents the left-half of this window region, and sum (C1) represents that D1 represents the right half part of this window region to all pixels summation in C1, sum (D1) represents all pixels in D1 is sued for peace, and fabs represents and takes absolute value；
Feature two, this window region is divided into upper and lower two parts, feature two=fabs (sum (C2)-sum (D2))/512,
Wherein, C2 represents the first half of this window region, and sum (C2) represents that D2 represents the latter half of this window region to all pixels summation in C2, and sum (D2) represents all pixels summation in D2；
Feature three, this window region is from left to right divided into three parts, if respectively C3, D3, E3, wherein, / 4th of the width of C3=this window width, the half of the width of D3=this window width, / 4th of the width of E3=this window width, feature three=fabs (sum (C3)+sum (E3)-sum (D3))/512
Wherein, sum (C3) represents all pixels summation in C3, and sum (D3) represents that all pixels in E3 are sued for peace by sum (E3) expression to all pixels summation in D3；
Feature four, this window region is divided into three parts from top to bottom, if respectively C4, D4, E4, wherein, / 4th of the height of C4=this window height, the half of the height of D4=this window height, / 4th of the height of E4=this window height, feature four=fabs (sum (C4)+sum (E4)-sum (D4))/512
Wherein, sum (C4) represents all pixels summation in C4, and sum (D4) represents that all pixels in E4 are sued for peace by sum (E4) expression to all pixels summation in D4.
4. method according to claim 1, it is characterised in that described detected each urine thin born of the same parents are carried out high-pass filtering include:
For each urine thin born of the same parents, the window of 8*8 size is slided in this urine thin born of the same parents region to preset the 3rd step-length, for the window after sliding every time, the convolution kernel of the pixel value in this window region Yu default 8*8 size is carried out convolution algorithm, obtains the spectrum energy of this window region；
When window slides end in this urine thin born of the same parents region, the spectrum energy of the window region after slip every time is added, obtains the spectrum energy of this urine thin born of the same parents,
Wherein, default convolution kernel is: the coefficient in the region of central authorities 4*4 is all+3, and the coefficient in remaining area is all-1.
5., according to the arbitrary described method of Claims 1-4, it is characterised in that described more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, select a width to focus on best image in several urine sediment images described and include:
When calculating spectrum energy average and the variance of all urine sediment images, judge that whether the maximum spectrum energy average difference with secondary big spectrum energy average is more than preset value, if so, then using image corresponding for maximum spectrum energy average as focus on optimized image；Otherwise, relatively spectrum energy variance maximum, that secondary big spectrum energy average is corresponding, using image corresponding for wherein less spectrum energy variance as focusing on optimized image.
6. a device for the Urine sediments analyzer MIcrosope image that selective focus is best, including:
One urine thin born of the same parents' detection module, for detecting all urine thin born of the same parents in each width urine sediment image；Described urine sediment image is for same Urine sediments analyzer sample, by finely tuning the arbitrary width in several urine sediment images of this sample that microscopical camera lens collects；
One spectrum energy computing module, for each urine thin born of the same parents in each width urine sediment image are carried out high-pass filtering, obtains the spectrum energy of this urine thin born of the same parents, and calculates spectrum energy average and the variance of this width urine sediment image；
One Image selection module, is used for more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, selects a width to focus on best image in several urine sediment images described.
7. device according to claim 6, it is characterised in that described urine thin born of the same parents' detection module includes:
One training module, for gathering prospect sample set and background sample set in advance, wherein, prospect sample is the image only comprising urine thin born of the same parents, and background sample is do not comprise urine thin born of the same parents only to comprise the image of background；Extract from each prospect sample and each background sample respectively and preset feature；Adopt default training algorithm that foreground features collection and background characteristics collection are trained, obtain distinguishing the grader of foreground image and background image；
One detection module, for for each width urine sediment image, the window of the foreground and background sample formed objects gathered with training module is slided on this width urine sediment image to preset the first step-length, for the window after sliding every time, adopt the feature extracting method identical with training module to gather from this window region and preset feature, adopting described grader to distinguish this window region is prospect or background, if prospect, then this window region is labeled as urine thin born of the same parents.
8. device according to claim 7, it is characterised in that described training module extracts default feature respectively from each prospect sample and each background sample and includes:
For the window region after sliding every time, extract following four category features:
Feature one, this window region is divided into left and right two parts, feature one=fabs (sum (C1)-sum (D1))/512,
Wherein, C1 represents the left-half of this window region, and sum (C1) represents that D1 represents the right half part of this window region to all pixels summation in C1, sum (D1) represents all pixels in D1 is sued for peace, and fabs represents and takes absolute value；
Feature two, this window region is divided into upper and lower two parts, feature two=fabs (sum (C2)-sum (D2))/512,
Wherein, C2 represents the first half of this window region, and sum (C2) represents that D2 represents the latter half of this window region to all pixels summation in C2, and sum (D2) represents all pixels summation in D2；
Feature three, this window region is from left to right divided into three parts, if respectively C3, D3, E3, wherein, / 4th of the width of C3=this window width, the half of the width of D3=this window width, / 4th of the width of E3=this window width, feature three=fabs (sum (C3)+sum (E3)-sum (D3))/512
Wherein, sum (C3) represents all pixels summation in C3, and sum (D3) represents that all pixels in E3 are sued for peace by sum (E3) expression to all pixels summation in D3；
Feature four, this window region is divided into three parts from top to bottom, if respectively C4, D4, E4, wherein, / 4th of the height of C4=this window height, the half of the height of D4=this window height, / 4th of the height of E4=this window height, feature four=fabs (sum (C4)+sum (E4)-sum (D4))/512
Wherein, sum (C4) represents all pixels summation in C4, and sum (D4) represents that all pixels in E4 are sued for peace by sum (E4) expression to all pixels summation in D4.
9. device according to claim 6, it is characterised in that each urine thin born of the same parents in each width urine sediment image are carried out high-pass filtering and include by described spectrum energy computing module:
For each urine thin born of the same parents, the window of 8*8 size is slided in this urine thin born of the same parents region to preset the 3rd step-length, for the window after sliding every time, the convolution kernel of the pixel value in this window region Yu default 8*8 size is carried out convolution algorithm, obtains the spectrum energy of this window region；
When window slides end in this urine thin born of the same parents region, the spectrum energy of the window region after slip every time is added, obtains the spectrum energy of this urine thin born of the same parents,
Wherein, default convolution kernel is: the coefficient in the region of central authorities 4*4 is all+3, and the coefficient in remaining area is all-1.
10. according to the arbitrary described device of claim 6 to 9, it is characterized in that, described Image selection module is more big according to spectrum energy average and variance is more little, the principle that image focusing is more good, selects a width to focus on best image and include in several urine sediment images described:
The spectrum energy average of all urine sediment images for same Urine sediments analyzer sample calculated for spectrum energy computing module and variance, judge that whether the maximum spectrum energy average difference with secondary big spectrum energy average is more than preset value, if so, then using image corresponding for maximum spectrum energy average as focus on optimized image；Otherwise, relatively spectrum energy variance maximum, that secondary big spectrum energy average is corresponding, using image corresponding for wherein less spectrum energy variance as focusing on optimized image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410805362.8A CN105760878A (en) | 2014-12-19 | 2014-12-19 | Method and device for selecting urinary sediment microscope image with optimal focusing performance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410805362.8A CN105760878A (en) | 2014-12-19 | 2014-12-19 | Method and device for selecting urinary sediment microscope image with optimal focusing performance |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105760878A true CN105760878A (en) | 2016-07-13 |
Family
ID=56341206
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410805362.8A Pending CN105760878A (en) | 2014-12-19 | 2014-12-19 | Method and device for selecting urinary sediment microscope image with optimal focusing performance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105760878A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644425A (en) * | 2017-09-30 | 2018-01-30 | 湖南友哲科技有限公司 | Target image choosing method, device, computer equipment and storage medium |
| CN112330586A (en) * | 2019-08-02 | 2021-02-05 | 苏州迈瑞科技有限公司 | Urine analyzer diagnosis speed-increasing method, urine analyzer and computer readable storage medium |
| CN113538435A (en) * | 2021-09-17 | 2021-10-22 | 北京航空航天大学 | Pancreatic cancer pathological image classification method and system based on deep learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6463438B1 (en) * | 1994-06-03 | 2002-10-08 | Urocor, Inc. | Neural network for cell image analysis for identification of abnormal cells |
| CN101702053A (en) * | 2009-11-13 | 2010-05-05 | 长春迪瑞实业有限公司 | Method for automatically focusing microscope system in urinary sediment examination equipment |
| CN101713776A (en) * | 2009-11-13 | 2010-05-26 | 长春迪瑞实业有限公司 | Neural network-based method for identifying and classifying visible components in urine |
| CN101900737A (en) * | 2010-06-10 | 2010-12-01 | 上海理工大学 | Automatic identification system for urinary sediment visible components based on support vector machine |
| CN103793709A (en) * | 2012-10-26 | 2014-05-14 | 西门子医疗保健诊断公司 | Cell recognition method and device, and urine analyzer |
| CN103793902A (en) * | 2012-10-26 | 2014-05-14 | 西门子医疗保健诊断公司 | Casts identification method and device, and urine analyzer |
-
2014
- 2014-12-19 CN CN201410805362.8A patent/CN105760878A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6463438B1 (en) * | 1994-06-03 | 2002-10-08 | Urocor, Inc. | Neural network for cell image analysis for identification of abnormal cells |
| CN101702053A (en) * | 2009-11-13 | 2010-05-05 | 长春迪瑞实业有限公司 | Method for automatically focusing microscope system in urinary sediment examination equipment |
| CN101713776A (en) * | 2009-11-13 | 2010-05-26 | 长春迪瑞实业有限公司 | Neural network-based method for identifying and classifying visible components in urine |
| CN101900737A (en) * | 2010-06-10 | 2010-12-01 | 上海理工大学 | Automatic identification system for urinary sediment visible components based on support vector machine |
| CN103793709A (en) * | 2012-10-26 | 2014-05-14 | 西门子医疗保健诊断公司 | Cell recognition method and device, and urine analyzer |
| CN103793902A (en) * | 2012-10-26 | 2014-05-14 | 西门子医疗保健诊断公司 | Casts identification method and device, and urine analyzer |
Non-Patent Citations (3)
| Title |
|---|
| 周晓谋: "尿液显微颗粒自动检测与图像处理研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 李亚妹: "基于多尺度计算的尿沉渣图像识别方法研究", 《中国硕士学位论文全文数据库 信息科技辑》 * |
| 苏传朋: "尿沉渣显微图像有形成分自动分割算法研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644425A (en) * | 2017-09-30 | 2018-01-30 | 湖南友哲科技有限公司 | Target image choosing method, device, computer equipment and storage medium |
| CN112330586A (en) * | 2019-08-02 | 2021-02-05 | 苏州迈瑞科技有限公司 | Urine analyzer diagnosis speed-increasing method, urine analyzer and computer readable storage medium |
| CN113538435A (en) * | 2021-09-17 | 2021-10-22 | 北京航空航天大学 | Pancreatic cancer pathological image classification method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huang et al. | A multidirectional and multiscale morphological index for automatic building extraction from multispectral GeoEye-1 imagery | |
| CN104392432A (en) | Histogram of oriented gradient-based display panel defect detection method | |
| CN112258496A (en) | Underground drainage pipeline disease segmentation method based on full convolution neural network | |
| CN104732500A (en) | Traditional Chinese medicinal material microscopic image noise filtering system and method adopting pulse coupling neural network | |
| CN102254159B (en) | Interpretation method for digital readout instrument | |
| CN104200217B (en) | Hyperspectrum classification method based on composite kernel function | |
| CN106650737A (en) | Image automatic cutting method | |
| EP3819859A1 (en) | Sky filter method for panoramic images and portable terminal | |
| CN103258206A (en) | Silicon solar cell surface defect detection and identification method | |
| CN106548160A (en) | A kind of face smile detection method | |
| CN102087652A (en) | Method for screening images and system thereof | |
| CN106548169A (en) | Fuzzy literal Enhancement Method and device based on deep neural network | |
| CN104574296B (en) | A kind of method for polarizing the m ultiwavelet fusion treatment picture for removing haze | |
| CN102132323A (en) | Automatic image straightening | |
| CN109344851B (en) | Image classification display method and device, analysis instrument and storage medium | |
| CN102521600A (en) | Method and system for identifying white-leg shrimp disease on basis of machine vision | |
| JPH10508709A (en) | Device for detecting air bubbles in cover slip adhesive | |
| CN104574381B (en) | A kind of full reference image quality appraisement method based on local binary patterns | |
| CN107066959A (en) | A kind of hyperspectral image classification method based on Steerable filter and linear space correlation information | |
| CN108492291A (en) | A kind of photovoltaic silicon chip Defect Detection system and method based on CNN segmentations | |
| CN102830404A (en) | Method for identifying laser imaging radar ground target based on range profile | |
| CN102663410B (en) | Method and system for detecting microcalcifications in mammogram | |
| CN107657220A (en) | A kind of leukorrhea mould automatic testing method based on HOG features and SVM | |
| CN105760878A (en) | Method and device for selecting urinary sediment microscope image with optimal focusing performance | |
| CN106651792A (en) | Method and device for removing strip noise of satellite image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160713 |
|
| WD01 | Invention patent application deemed withdrawn after publication |