CN103903235B - Cell image ultra-resolution ratio reconstructing method - Google Patents
Cell image ultra-resolution ratio reconstructing method Download PDFInfo
- Publication number
- CN103903235B CN103903235B CN201410040294.0A CN201410040294A CN103903235B CN 103903235 B CN103903235 B CN 103903235B CN 201410040294 A CN201410040294 A CN 201410040294A CN 103903235 B CN103903235 B CN 103903235B
- Authority
- CN
- China
- Prior art keywords
- cell
- pixel
- interpolation
- resolution
- image
- 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.)
- Active
Links
Landscapes
- Image Processing (AREA)
Abstract
Cell image ultra-resolution ratio reconstructing method, the cell image sequence of 1: 1 low resolution is obtained by microfluid microscopic system, then the N number of same cells image needed for being therefrom partitioned into super-resolution reconstruction, and high-resolution cell image information is reconstructed using the cell image information of these low resolution.The present invention only needs to find multiple same cells in a two field picture carry out super-resolution reconstruction, without reading many two field pictures, reduce memory space, improve processing speed, and can equally gather more relevant informations, the cell after amplification is embodied more minutias, reach the requirement that can differentiate diagnosis cell.
Description
Technical field
The present invention is belonging to microfluid reconstructing method technical field, is related to a kind of cell image ultra-resolution ratio reconstructing method.
Background technology
In recent years, a kind of emerging chip technology --- microfluidic chip technology with its exclusive quick analysis, low-power consumption,
It is miniaturized and automates and develop very fast.It has that flow of fluid is controllable, consumption sample and reagent are few, analyze speed into
The advantages of hundred times of raisings, analyzed while up to a hundred samples can be carried out within a few minutes or even shorter time.This causes it
Its exclusive charming glamour and development prospect are presented gradually in medical research field.
It is well known that in current biomedical research, the observational study of biological cell be always in one very
Consequence.By the observational study to cell movement image sequence, we can observe the character mutation of cell and move
State behavior, and cell is classified and counted, provide good foundation to medical diagnosis.But existing observation of cell is aobvious
Micro-mirror device is excessively huge and costliness, and observation of cell can only be observed on slide or on glassware, and this is caused
We have not observed one of most important life characteristics of cell --- mobility.These a variety of defects, in micro-fluidic core
After the technology of piece occurs, all a new hope, and imminent major transformation are brought to medical field.
The general width in microchannel in micro-fluidic chip is 10~50 rans, its sizableness with biological cell.
And can be by methods such as controllable pressure so that biological cell is very easy to be manipulated, observe and examine in microchannel
Survey, and he is overcome in existing equipment that detecting system is excessively huge, detection process is complicated, detection cycle is long, must be corresponding
Laboratory in detect etc. numerous shortcomings, it fully achieves low cost, low-power consumption, and can it is individual other or do not have
The standby outlying district for setting up laboratory condition carries out preliminary quick diagnosis.Based on these tempting advantages, a new research
Arise at the historic moment and develop more and more rapider in direction --- Optofluidic Microscope.
Earliest by California Institute of Technology, OFM systems are have developed, and obtained image of the worm in microchannel, he is sharp
With optical diffraction principle, one layer of light non-transmittable layers with dot matrix are done between microchannel and sensor.But the OFM systems
System complex process, it is necessary to need the vertical irradiation of light, and limited resolution system.So be born on this basis later again
SROFM systems, that layer of complicated technology that it need not be between microchannel and sensor, and can just be worked under natural light,
Directly with 1 in sensor acquisition trajectory:1 image sequence, then using super-resolution method, obtains high-resolution
Image so that resolution ratio can reach smaller than OFM.
The main thought of super-resolution reconstruction is intended to be generated by signal estimation theory using sequence of low resolution pictures
High-definition picture.At present, in the Super-Resolution Image Restoration used in SROFM systems, it is broadly divided into based on multiframe
Super-resolution reconstruction algorithm and the super-resolution reconstruction algorithm based on single frames.Super-resolution reconstruction algorithm based on multiframe be by
Several adjacent low-resolution image information aggregates get up, and useful information integrate obtain the single super-resolution of a width
Image.Super-resolution reconstruction based on single frames is to pass through phase using the relevant information of the different directions of neighbor pixel in original image
The difference answered realizes super-resolution reconstruction.Super-resolution reconstruction algorithm based on multiframe needs very big memory space come to phase
The Pixel Information of adjacent frame is stored, such as the image that 50 frames have just been used in testing has carried out super-resolution reconstruction, and this is much
The requirement of real-time of image procossing can not be met.Super-resolution reconstruction algorithm based on single frames disclosure satisfy that requirement of real-time, but
It is when being amplified to cell image, after the information only with the adjacent one part of pixel of cell peripheral, therefore amplification
The phenotypic characteristic of cell can not be embodied well.
Either multiframe super-resolution restructing algorithm, or single frames super-resolution reconstruction algorithm is all to reconstruct one kind to fill
Divide the method for showing cell characteristic information, real cell is restored rather than real.
The content of the invention
It is an object of the invention to provide a kind of cell image ultra-resolution ratio reconstructing method, solving prior art needs greatly
Memory space, complicated tracing process, and the problem of the requirement of real-time of image procossing can not be met, and it is traditional based on
The phenotypic characteristic of cell can not obtain the problem for embodying well after the super-resolution reconstruction algorithm amplification of single frames.
The technical scheme is that, cell image ultra-resolution ratio reconstructing method is obtained by microfluid microscopic system
1:The cell image sequence of 1 low resolution, the N number of same cells image needed for being then therefrom partitioned into super-resolution reconstruction,
And high-resolution cell image information is reconstructed using the cell image information of these low resolution.
The features of the present invention is also resided in：
Specifically include following steps：
Step 1, the cell image sequence of low resolution is obtained by cmos image sensor；
Step 2, n sample cell required for super-resolution reconstruction is searched in the current frame, cell from background
Split and extract；
Step 3, after searching n cell, information according to this n cell carries out Super-resolution reconstruction to target cell
Structure.
Above-mentioned steps 3 include：
(1) one is taken out from the cell for splitting as the reference cell A of interpolation and one n-1 times of net is made
Lattice, the pixel value of the reference cell A of interpolation is placed on the left upper end of each grid cell；
(2) registration is carried out to the cell that remaining is partitioned into, the cell picture element matrix being partitioned into is thin with the reference of interpolation
Born of the same parents' picture element matrix carries out respective pixel location and subtracts each other one by one, and the difference for obtaining is squared, and finally the matrix for obtaining, each is poor
Value is added summation, obtains a cost value sad；Then the cell for splitting is rotated by 90 ° continuation and interpolation cell pixel
Matrix carries out square summation after respective pixel location is subtracted each other one by one；The rest may be inferred, and the cell that each is partitioned into can obtain 4 sides
Upward sad, then compares this 4 sad, and the direction taken representated by the sad of minimum is that follow-up interpolation is prepared；
(3) the reference cell A of interpolation is amplified 4 times using the method for image element interpolation, obtains matrix B；
(4) by the method for 1/4 pixel value difference Motion Estimate, the cell of segmentation is inserted into step 1 and is set up
Super-resolution grid in, specific method is the pixel of the correspondence position that A is individually subtracted with the cell corresponding pixel points being partitioned into
Then square value, is finally added summation, obtains a cost value sad1；N-1 cost value is always obtained, compares size, obtain
Minimum value corresponding to position be exactly the cell the being partitioned into optimum position to be inserted into grid.
(5) remaining pixel value after interpolation is drawn with the method for weighted sum；
(6) picture element matrix for amplifying n-1 times after interpolation is filtered, it is caused in removal Interpolation Process to obscure,
And suppress noise；
(7) result after (6) are processed is sharpened treatment to strengthen detailed information.
(8) with the method for image element interpolation, the result that (7) draw is amplified, finally amplifies 2 (n-1) times；Use
Process is with (3) above.
(9) and then to the image information of target area strengthen to protrude the information characteristics of target.
Obtain a cell by amplifying 2 (n-1) times after super-resolution reconstruction.
In above-mentioned (3), the method for obtaining matrix B is：
First the reference cell A of interpolation is put and is twice, obtained by equation below：
Bk=(Ak+[k/5]+Ak+1+[k/5])/2
Ck=(Ak+Ak+6)/2
Dk=(Bk+Bk+5+Ck+Ck+1)/4 (k=1,2 ... 30) (1)
Wherein AkIt is the original pixel value in the reference cell A of interpolation, Bk、Ck、DkN-1 times for cell to be referred to surpasses
It is to be evaluated in resolution grid.
The E of boundary1And H1And F1And G1Obtained by the method for spillover position edge replication processes, formula is as follows：
Ek=(A6k+A6k)/2
Hk=(Ek+Ek+1+C6k+C6k)/4
Fk=(A30+k+A30+k)/2
Gk=(B25+k+B25+k+Fk+Fk+1)/4 (k=1,2 ... 5) (2)
Wherein AkIt is the original pixel value in the reference cell A of interpolation, Ck、BkAsked Deng waiting of having been obtained for formula (1)
Value, Ek、Hk、Fk、GkIt is the value to be tried to achieve of boundary；Wherein k takes 1,2 ... 5, and each alphabetical variable represents difference in cell image
Pixel position in a matrix.
The reference cell A that twice interpolation is put according to the method in formula above (1) and formula (2) is carried out one again
Secondary amplification.Specific method is above on the basis of the result for obtaining, and recycles a quarter pixel shown in formula below (3)
The end product for being drawn by formula (2) above is put and is twice by interpolation method, using the original pixels of the reference cell A of interpolation
Data and being averaging above with 1/2nd precision pixels that formula (1) and formula (2) are calculated get, and specific formula is such as
Under：
A0=(B3+A1)/2 (3)
Wherein A1、B3It is a0 surrounding pixel values.
When 1/4 interpolated data is calculated, the data for being used are influenceed by half-pixel data position；Half-pix is odd number
During position, using " pawl " type of falling；During for even number, using just " pawl " type.
In above-mentioned (5), the algorithm of specific pixel value is as follows：
Wherein the coefficient of each be qA1, qA2, qA7, qA7 (i, j), qA (i, j) if value defer to pending pixel
It is then value 1 that the Pixel Information of point is present, and is value 0 if the Pixel Information of pending pixel does not exist；A (i, j) is needs
The pixel value tried to achieve, its computation rule is the result if the original pixel value of correspondence mesh point in pending pixel is for 0
It is calculated by above formula, if the pixel value of correspondence mesh point is not zero in pending pixel, proves to treat
Correspondence mesh point inserts new pixel value in (3) in step 3 in treatment pixel, then the pixel of pending pixel
Value still takes the pixel value of insertion in (3) in step 3；And the pixel variable-value in the expression formula of the superincumbent formula right side is former
It is also then according to this standard.
The beneficial effects of the invention are as follows：
1st, the present invention obtains the cell image sequence of low resolution from cmos image sensor, amplifies according to target cell
Multiple n find n same cells in the current frame, target cell is amplified n times by the method for interpolation of the present invention.
Of the present invention is the cell image ultra-resolution ratio reconstructing method based on single frames multiple target, relative to the oversubscription based on multiframe
Resolution reconstructing method, the present invention only needs to just to collect in the current frame n cell of needs, can greatly reduce storage
Space, improves processing speed；The ultra-resolution ratio reconstructing method based on single frames traditional relative to some for having existed, the present invention
It is that target cell is amplified using the information of multiple same cells, the information of collection is more, improves super-resolution reconstruction
Effect, can more accurately show the minutia of some cells.
2nd, the present invention only needs to find multiple same cells in a two field picture carry out super-resolution reconstruction, without
Many two field pictures are read, memory space is reduced, processing speed is improve, and can equally gather more relevant informations,
The cell after amplification is embodied more minutias, reach the requirement that can differentiate diagnosis cell.
Brief description of the drawings
Fig. 1 is a part of Optofluidic Microscope for not containing cell image by a frame of any treatment
Sectional drawing；
Fig. 2 is the picture element matrix schematic diagram of the reference cell of interpolation；
Fig. 3 is n-1 times of super-resolution of the reference cell of interpolation in cell image ultra-resolution ratio reconstructing method of the present invention
Grid schematic diagram；
Fig. 4 be cell image ultra-resolution ratio reconstructing method of the present invention in cell to be referred to put with 1/2nd picture element interpolation methods
Big 2 times of picture element matrix schematic diagram；
Fig. 5 is cell 1/4 pixel interpolation method amplification to be referred in cell image ultra-resolution ratio reconstructing method of the present invention
4 times of picture element matrix schematic diagram；
Fig. 6 is the matrix of the cell insertion super-resolution grid of segmentation in cell image ultra-resolution ratio reconstructing method of the present invention
Schematic diagram；
Fig. 7 is to calculate remaining picture after interpolation with the method for weighted sum in cell image ultra-resolution ratio reconstructing method of the present invention
The matrix schematic diagram of prime information；
Fig. 8 is similar 51 being partitioned into Optofluidic Microscope:The bone marrow cell of 1 size；
Fig. 9 is the cell to be referred to selected from 5 cells being partitioned into；
Figure 10 is the design sketch of super-resolution grid in cell image ultra-resolution ratio reconstructing method of the present invention；
Figure 11 is to be exaggerated 4 times by image element interpolation method in cell image ultra-resolution ratio reconstructing method of the present invention to treat
With reference to cell effect figure；
Figure 12 is that remaining cell of segmentation is inserted super-resolution net in cell image ultra-resolution ratio reconstructing method of the present invention
The design sketch of lattice；
After Figure 13 is for interpolation in cell image ultra-resolution ratio reconstructing method of the present invention, rest of pixels point is by after weighted sum
Design sketch；
Figure 14 is the effect of the super-resolution image after medium filtering in cell image ultra-resolution ratio reconstructing method of the present invention
Figure；
Figure 15 is the effect after being sharpened to filtered image in cell image ultra-resolution ratio reconstructing method of the present invention
Figure；
Figure 16 is image after sharpening in cell image ultra-resolution ratio reconstructing method of the present invention plus the image before sharpening
Design sketch；
Figure 17 is after enhancing contrast with the method for grey level window in cell image ultra-resolution ratio reconstructing method of the present invention
Design sketch；
Figure 18 is after being exaggerated 2 times with the method for image element interpolation in cell image ultra-resolution ratio reconstructing method of the present invention
Design sketch；
Figure 19 is that 10000 times of cell image design sketch is exaggerated under magnifying glass.
Specific embodiment
The present invention is described in detail with reference to the accompanying drawings and detailed description.
The implementation of cell image ultra-resolution ratio reconstructing method technical scheme proposed by the invention is based on a hypothesis：
Cell in the image collected in Optofluidic Microscope systems is all cell lineage.
The present invention is specifically implemented according to following steps：
Step 1：The cell image sequence of low resolution is obtained by cmos image sensor；
Step 2：N sample cell required for searching super-resolution reconstruction in the current frame.Using corresponding segmentation
Method from background splits cell and extract.
Step 3：After searching n cell, it is possible to which information according to this n cell carries out super-resolution to target cell
Rate is reconstructed.
In step 3, the ultra-resolution ratio reconstructing method based on interpolation is used, i.e., with n similar low resolution cell
Information reconstruct a high-resolution cell for amplifying 2 (n-1) times by the method for interpolation, specifically include following steps：
(1) a reference cell as interpolation is taken out from the cell for splitting as shown in Fig. 2 and making one
N-1 times of grid, is placed on the left upper end of each grid cell, as shown in Figure 3 with reference to the pixel value of cell A.
(2) registration and then to the cell that remaining is partitioned into is carried out, and the cell picture element matrix being partitioned into, with interpolation
Respective pixel location is carried out with reference to cell picture element matrix to subtract each other one by one, the difference for obtaining is squared, finally that the matrix for obtaining is every
One difference value summation, obtains a cost value sad.The cell for splitting is being rotated by 90 ° continuation and interpolation cell
Picture element matrix carries out square summation after respective pixel location is subtracted each other one by one.The rest may be inferred, and the cell that each is partitioned into can obtain 4
Sad on individual direction, then compares this 4 values, and the direction taken representated by the value of minimum is that follow-up difference is prepared.
(3) the reference cell A of interpolation is amplified 4 times using the method for point pixel value difference, obtains matrix B.
The reference cell A of interpolation was put before this was twice, such as Fig. 4, and there is equation below to obtain：
B1=(A1+A2)/2
C1=(A1+A7)/2
D1=(B1+B2+C1+C2)/4 (1)
Wherein A1~A36If Fig. 4 is the original pixel value in the reference cell A of interpolation, B1、C1、D1If Fig. 4 is to wait to refer to
It is to be evaluated in n-1 times of super-resolution grid of cell.
The E of boundary1And H1And F1And G1Obtained by the method for spillover position edge replication processes, formula is as follows:
E1=(A6+A6)/2
H1=(E1+E2+C6+C6)/4
F1=(A31+A31)/2
G1=(B26+B26+F1+F2)/4 (2)
Wherein A1~A36If Fig. 4 is the original pixel value in the reference cell A of interpolation, C6、B26Deng for formula (1) in the hope of
To be evaluated, the E for obtaining1、H1、F1、G1It is boundary value to be asked.
Then put using using the 1/4 pixel interpolation method shown in equation 3 below on the basis of amplifying 2 times
It is twice, 1/2nd precision pixels mainly inserted by whole pixel data and above with formula (1) and formula 2 are calculated
Get, as shown in Figure 5.Half-pixel data A11/4 accuracy data a0, a1 ... a7, a8 of surrounding is according to linear interpolation formula
(3) it is calculated：
A0=(B3+A1)/2 (3)
Wherein A1、B3It is illustrated in figure 5 surrounding pixel values.
When 1/4 interpolated data is calculated, the data for being used receive the influence of half-pixel data position.Half-pix is odd number
During position, using " pawl " type of falling；During for even number, using just " pawl " type；For example, in such as Fig. 5,8 half pictures around whole pixel X4
Vegetarian refreshments is respectively C1、B4、C2、A1、A2、C3、B7And C4, wherein A1It is half-pix point that searching position is 3, so he uses
" pawl " type.
(4) and then by the method for 1/4 pixel value difference Motion Estimate, the cell of segmentation is inserted into step 1 and is built
In vertical super-resolution grid, specific method is the picture of the correspondence position that A is individually subtracted with the cell corresponding pixel points being partitioned into
Then square element value, is finally added summation, obtains a cost value sad1.N-1 cost value is always obtained, compares size, obtain
To minimum value corresponding to position be exactly the cell the being partitioned into optimum position to be inserted into grid.Such as Fig. 6, from the back of the body
The position that the cell J and K of 2 be partitioned into scape is inserted in this super-resolution grid is calculated most by as above method
Best placement.
(5) remaining gridding information, such as Fig. 7 after interpolation and then with the method for weighted sum are drawn.Specific pixel value is by such as
Shown in lower formula：
The value of the wherein coefficient (i.e. qa, qb, qc, qd, qe etc.) of each is deferred to if the Pixel Information in the shop is present
It is value 1, is value 0 if the Pixel Information in the shop does not exist.Such as Fig. 7 institutes such as alphabetical a, b, c, d, e, f, g, h, i, j, k, l, m
It is shown as calculative pixel value.Such as Fig. 7 such as k1, j1 is the pixel value inserted by step 4
(6) picture element matrix for amplifying n-1 times after interpolation is filtered, it is caused in removal Interpolation Process to obscure,
And suppress noise.
(7) and then to the result after (6) are processed it is sharpened treatment to strengthen detailed information.
(8) with the method for point pixel value difference, the result that (7) draw is amplified, finally amplifies 2 (n-1) times.Use
Process is with above step 3.
(9) and then to the image information of target area strengthen to protrude the information characteristics of target.
9 steps more than, just obtain a cell by amplifying 2 (n-1) times after super-resolution reconstruction.Theoretically,
The cell of acquirement is more, and information is more abundant, and the multiple of amplification is more, cytological map of the formed cell image under microscope
Picture.
Embodiment
5 cells are taken in experiment to be illustrated as a example by cell is amplified into 16 times.
Step 1：The cell image sequence of low resolution is obtained by cmos image sensor；
Step 2：Feature according to cell splits cell from background, and 5 for being partitioned into cell is as shown in Figure 8.
Step 3：5 information of cell being partitioned into using this carry out super-resolution reconstruction.Using super based on interpolation
Resolution reconstruction method, i.e., reconstructed with the image information of this 5 similar low resolution cells by certain difference approach
One amplify 16 times the high-resolution minutia that can embody the cell virtual cell image information, specifically include with
Lower step：
(1) cell being partitioned into is taken for interpolation refers to cell, such as Fig. 9.
(2) this is made with reference to cell quadruplication the grid of high-resolution interpolation, such as Figure 10.
(3) using the method for point pixel value difference, the image information for treating to be inserted into reference to cell 4 times of amplification, such as Figure 11.
(4) with remaining 4 cells being partitioned into, each corresponding unit in each pixel and Fig. 9 of each cell
Then square remaining 15 pixel pixel makes the difference,.The squared difference summation that 36 units are obtained altogether.Then size is compared,
Minimum is then the cell Pixel Information that this is partitioned into, and is inserted into the optimum position in super-resolution grid, such as Figure 12.
(5) empty pixel remaining for super-resolution grid is obtained using the mode of weighted sum above-mentioned, its
Result such as Figure 13.
(6) medium filtering is carried out to the cell for obtaining, as shown in figure 14, to be exaggerated 4 times in experiment and using medium filtering
Cell afterwards.
(7) to going confusion after cell image Sobel operators be filtered, such as Figure 15 for experiment in carry out chaotic intermediate value
After filtering, the image after being sharpened with Sobel.The method being added with Figure 14 respective pixels by Figure 15 is obtained as shown in figure 16
Superimposed image before and after sharpening.
(8) using the method for grey level window, contrast enhancing is carried out to the volume image after amplification, such as Figure 17 is with grey level window side
Method carries out the enhanced image of contrast.
(9) image after sharpening is being amplified 2 times, after such as Figure 18 is to amplify with the above-mentioned point of method of pixel value difference
Cell image.
9 steps more than, just obtain a cell image by 8 times of amplification after super-resolution reconstruction.Due to this hair
What the cell image ultra-resolution ratio reconstructing method of bright proposition was utilized is not only the Pixel Information with reference to cell, also uses other same
The Pixel Information of class cell, takes full advantage of the relevant information for collecting, and improves definition of the interpolation with reference to cell, and this
Method provides memory space without the data message for a large amount of two field pictures, and due to need not can just reconstruct high-resolution after multiframe
Rate image, it is possible to meet the requirement of real-time.
Figure 19 is the cell image of 10000 times of amplification under the microscope.From experimental result as can be seen that oversubscription of the invention
Resolution reconstructing method can reflect the requirement of cell minutia, and effect is extraordinary.And of the present invention be
The method of Digital Image Processing, by super on the basis of the cell image of the low resolution that cmos image sensor is collected
Resolution reconstruction algorithm is amplified to cell, it is not necessary to microscope, it is not required that high-resolution sampling array, and interpolation
Method calculate simple, speed is fast, and only with single-frame images, it is only necessary to less electronic processing hardware, not only cheap,
And it is compact convenient, there is good practicality.
Claims (3)
1. cell image ultra-resolution ratio reconstructing method, it is characterised in that obtain 1 by microfluid microscopic system:1 low resolution
The cell image sequence of rate, the N number of same cells image needed for being then therefrom partitioned into super-resolution reconstruction, and it is low using these
The cell image information of resolution ratio reconstructs high-resolution cell image information；
Specifically include following steps：
Step 1, the cell image sequence of low resolution is obtained by cmos image sensor；
Step 2, n sample cell required for super-resolution reconstruction is searched in the current frame, splits cell from background
And extract；
Step 3, after searching n cell, information according to this n cell carries out super-resolution reconstruction to target cell；
The step 3 includes：
(1) one is taken out from the cell for splitting as the reference cell A of interpolation and one n-1 times of grid is made,
The pixel value of the reference cell A of interpolation is placed on the left upper end of each grid cell；
(2) registration is carried out to the cell that remaining is partitioned into, the cell picture element matrix and the reference cell picture of interpolation that are partitioned into
Prime matrix carries out respective pixel location and subtracts each other one by one, and the difference for obtaining is squared, finally each difference phase of the matrix for obtaining
Plus summation, obtain a cost value sad；Then the cell for splitting is rotated by 90 ° continuation and interpolation cell picture element matrix
Carry out square summation after respective pixel location is subtracted each other one by one；The rest may be inferred, and the cell that each is partitioned into can be obtained on 4 directions
Sad, then compare this 4 sad, take minimum sad representated by direction be that follow-up interpolation is prepared；
(3) the reference cell A of interpolation is amplified 4 times using the method for image element interpolation, obtains matrix B；
(4) by the method for 1/4 pixel value difference Motion Estimate, the cell of segmentation be inserted into that step 1 set up it is super
In resolution grid, specific method is the pixel value of the correspondence position that A is individually subtracted with the cell corresponding pixel points being partitioned into, so
Afterwards square, summation is finally added, a cost value sad1 is obtained；N-1 cost value is always obtained, compares size, obtain most
Position corresponding to small value is exactly the cell the being partitioned into optimum position to be inserted into grid；
(5) remaining pixel value after interpolation is drawn with the method for weighted sum；
(6) picture element matrix for amplifying n-1 times after interpolation is filtered, it is caused in removal Interpolation Process to obscure, and
Suppress noise；
(7) result after (6) are processed is sharpened treatment to strengthen detailed information；
(8) with the method for image element interpolation, the result that (7) draw is amplified, finally amplifies 2 (n-1) times；The process of use
With (3) above；
(9) and then to the image information of target area strengthen to protrude the information characteristics of target；
Obtain a cell by amplifying 2 (n-1) times after super-resolution reconstruction.
2. cell image ultra-resolution ratio reconstructing method as claimed in claim 1, it is characterised in that in (3), obtain matrix
The method of B is：
First the reference cell A of interpolation is put and is twice, obtained by equation below：
Bk=(Ak+[k/5]+Ak+1+[k/5])/2
Ck=(Ak+Ak+6)/2
Dk=(Bk+Bk+5+Ck+Ck+1)/4 (k=1,2 ... 30) (1)
Wherein AkIt is the original pixel value in the reference cell A of interpolation, Bk、Ck、DkIt is n-1 times of super-resolution of cell to be referred to
It is to be evaluated in grid；
The E of boundary1And H1And F1And G1Obtained by the method for spillover position edge replication processes, formula is as follows：
Ek=(A6k+A6k)/2
Hk=(Ek+Ek+1+C6k+C6k)/4
Fk=(A30+k+A30+k)/2
Gk=(B25+k+B25+k+Fk+Fk+1)/4 (k=1,2 ... 5) (2)
Wherein AkIt is the original pixel value in the reference cell A of interpolation, Ck、BkIt is to be evaluated Deng what is obtained for formula (1),
Ek、Hk、Fk、GkIt is the value to be tried to achieve of boundary；Wherein k takes 1,2 ... 5, and each alphabetical variable represents different pictures in cell image
Vegetarian refreshments position in a matrix；
The reference cell A that twice interpolation is put according to the method in formula above (1) and formula (2) is once put again
Greatly；Specific method is above on the basis of the result for obtaining, and recycles the 1/4 pixel interpolation shown in formula below (3)
The end product for being drawn by formula (2) above is put and is twice by method, using the raw pixel data of the reference cell A of interpolation
It is averaging with 1/2nd precision pixels calculated above with formula (1) and formula (2) and is got, specific formula is as follows：
A0=(B3+A1)/2 (3)
Wherein A1、B3It is a0 surrounding pixel values；
When 1/4 interpolated data is calculated, the data for being used are influenceed by half-pixel data position；Half-pix is odd positions
When, using " pawl " type of falling；During for even number, using just " pawl " type.
3. cell image ultra-resolution ratio reconstructing method as claimed in claim 2, it is characterised in that in (5), specific pixel
Value-based algorithm is as follows：
Wherein the coefficient of each be qA1, qA2, qA7, qA7 (i, j), qA (i, j) if value defer to pending pixel
It is then value 1 that Pixel Information is present, and is value 0 if the Pixel Information of pending pixel does not exist；A (i, j) is tried to achieve for needs
Pixel value, its computation rule be if correspondence mesh point in pending pixel original pixel value for if 0 result be upper
What the formula in face was calculated, if the pixel value of correspondence mesh point is not zero in pending pixel, prove pending
Correspondence mesh point inserts new pixel value in (3) in the step 3 in pixel, then the pixel of pending pixel
Value still takes in the step 3 pixel value of insertion in (3)；And the pixel change in the expression formula of the superincumbent formula right side is measured
Value principle is also according to this standard.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410040294.0A CN103903235B (en) | 2014-01-27 | 2014-01-27 | Cell image ultra-resolution ratio reconstructing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410040294.0A CN103903235B (en) | 2014-01-27 | 2014-01-27 | Cell image ultra-resolution ratio reconstructing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103903235A CN103903235A (en) | 2014-07-02 |
| CN103903235B true CN103903235B (en) | 2017-06-06 |
Family
ID=50994542
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410040294.0A Active CN103903235B (en) | 2014-01-27 | 2014-01-27 | Cell image ultra-resolution ratio reconstructing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103903235B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106483276A (en) * | 2016-09-29 | 2017-03-08 | 大连理工大学 | A kind of optofluidic blood cell micro imaging system based on smart mobile phone |
| CN107764779B (en) * | 2017-09-27 | 2020-09-15 | 上海交通大学 | Super-resolution imaging method and system of confocal system |
| CN109146790B (en) * | 2018-08-27 | 2022-10-11 | 深圳大学 | Image reconstruction method and device, electronic equipment and storage medium |
| CN113269672B (en) * | 2021-04-14 | 2023-02-24 | 佛山科学技术学院 | Super-resolution cell image construction method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101198063A (en) * | 2006-12-05 | 2008-06-11 | 华为技术有限公司 | Encoding and decoding method and device, image element interpolation processing method and device |
| CN101227614A (en) * | 2008-01-22 | 2008-07-23 | 炬力集成电路设计有限公司 | Motion estimation device and method of video coding system |
| US7505636B2 (en) * | 2004-03-04 | 2009-03-17 | Broadcom Corporation | System and method for two-pass interpolation for quarter-pel motion compensation |
| CN103473751A (en) * | 2013-08-14 | 2013-12-25 | 西安理工大学 | CMOS sensor cell image super-resolution reconstruction method based on multiple objects |
-
2014
- 2014-01-27 CN CN201410040294.0A patent/CN103903235B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7505636B2 (en) * | 2004-03-04 | 2009-03-17 | Broadcom Corporation | System and method for two-pass interpolation for quarter-pel motion compensation |
| CN101198063A (en) * | 2006-12-05 | 2008-06-11 | 华为技术有限公司 | Encoding and decoding method and device, image element interpolation processing method and device |
| CN101227614A (en) * | 2008-01-22 | 2008-07-23 | 炬力集成电路设计有限公司 | Motion estimation device and method of video coding system |
| CN103473751A (en) * | 2013-08-14 | 2013-12-25 | 西安理工大学 | CMOS sensor cell image super-resolution reconstruction method based on multiple objects |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103903235A (en) | 2014-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334847B (en) | A kind of face identification method based on deep learning under real scene | |
| CN111429474B (en) | Mammary gland DCE-MRI image focus segmentation model establishment and segmentation method based on mixed convolution | |
| CN110544205B (en) | Image super-resolution reconstruction method based on visible light and infrared cross input | |
| CN106339998A (en) | Multi-focus image fusion method based on contrast pyramid transformation | |
| CN103871058B (en) | Compressed sampling matrix decomposition-based infrared small target detection method | |
| CN103903235B (en) | Cell image ultra-resolution ratio reconstructing method | |
| CN109389585B (en) | Brain tissue extraction method based on full convolution neural network | |
| CN109523470B (en) | Depth image super-resolution reconstruction method and system | |
| CN103473751B (en) | Based on the cmos sensor cell image ultra-resolution ratio reconstructing method of multiple goal | |
| CN103413286A (en) | United reestablishing method of high dynamic range and high-definition pictures based on learning | |
| CN103279935A (en) | Method and system of thermal infrared remote sensing image super-resolution reconstruction based on MAP algorithm | |
| CN104463785A (en) | Method and device for amplifying ultrasound image | |
| CN112949636A (en) | License plate super-resolution identification method and system and computer readable medium | |
| He et al. | Remote sensing image super-resolution using deep–shallow cascaded convolutional neural networks | |
| CN115471448A (en) | Artificial intelligence-based thymus tumor histopathology typing method and device | |
| CN114332283A (en) | Training method based on double-domain neural network and photoacoustic image reconstruction method | |
| CN104463822A (en) | Multi-focus image fusing method and device based on multi-scale overall filtering | |
| CN103106412B (en) | Flaky medium recognition methods and recognition device | |
| Messer et al. | A multiscale wavelet algorithm for atom tracking in STM movies | |
| Mazurek et al. | From the slit-island method to the ising model: Analysis of irregular grayscale objects | |
| CN105654070A (en) | Low-resolution face recognition method | |
| CN107818567B (en) | Brain local morphological feature description method based on cortical top point cloud | |
| Li et al. | SPA-RESUNET: strip pooling attention resunet for multi-class segmentation of vertebrae and intervertebral discs | |
| CN113191949A (en) | Multi-scale super-resolution pathological image digitization method and system and storage medium | |
| Ma et al. | Research on single-frame super-resolution reconstruction algorithm for low resolution cell images based on convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C53 | Correction of patent for invention or patent application | ||
| CB03 | Change of inventor or designer information |
Inventor after: Yu Ningmei Inventor after: Zhang Yang Inventor after: Shi Xiaoyu Inventor after: Ren Ru Inventor before: Yu Ningmei Inventor before: Shi Xiaoyu Inventor before: Ren Ru |
|
| COR | Change of bibliographic data |
Free format text: CORRECT: INVENTOR; FROM: YU NINGMEI SHI XIAOYU REN RU TO: YU NINGMEI ZHANG YANG SHI XIAOYU REN RU |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |