CN106203384A - A kind of cell division identification method of multiresolution - Google Patents
A kind of cell division identification method of multiresolution Download PDFInfo
- Publication number
- CN106203384A CN106203384A CN201610578288.XA CN201610578288A CN106203384A CN 106203384 A CN106203384 A CN 106203384A CN 201610578288 A CN201610578288 A CN 201610578288A CN 106203384 A CN106203384 A CN 106203384A
- Authority
- CN
- China
- Prior art keywords
- cell division
- cell
- sequence
- multiresolution
- model
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/695—Preprocessing, e.g. image segmentation
Abstract
The invention discloses the cell division identification method of a kind of multiresolution, described cell division identification method comprises the following steps: multiresolution features cell division collection and acellular being divided collection by hidden conditional random fields builds model respectively, obtains the model parameter under different resolution；Arbitrarily choose test cell division candidate sequence as cycle tests, extract visual signature sequence, will visual signature sequence inputting model be tested, obtain sample label corresponding under different model parameters be 1 and be 0 probability；If the probability that sample label is 1 more than be 0 probability, then cell division candidate sequence comprises cell division event, does not the most comprise cell division event.This method obtains the characteristic sequence under multiresolution by the grouping process of view-based access control model feature, make the study of state transition in model learning more efficient, and capture the high-level semantics information in cell division candidate sequence, significantly improve fissional discrimination.
Description
Technical field
The present invention relates to image procossing and area of pattern recognition, particularly relate to the cell division identification side of a kind of multiresolution
Method.
Background technology
Detecting cell division in timing phase contrast microscope image is a major issue in biological study, many
Field suffers from being widely applied value (such as list of references [1]).There is a lot of cell, and shape in micro-image the most simultaneously
Shape is different, and background is complicated.Along with the progress of biomedical level, the importance of cellular activity highlights day by day.In Basic of Biology
In the middle of research field, and the senior application such as stem cell manipulation, drug research and organizational project, the research of cell proliferation makes all
As neomorph medical science imagination is possibly realized.Cell behavior is of a great variety, including: cell division, cell migration, metamorphosis,
Apoptosis etc. (such as list of references [2]), wherein cell division event is particularly important.Cell passes through mitosis, by nucleus
Interior hereditary material is averagely allocated to offspring, thus carries out propagation and the self renewal (such as list of references [3]) of cell.Due to carefully
In born of the same parents' fission process, cellular morphology and cosmetic variation are the biggest so that fissional identification has certain difficulty.
Micro-image is carried out, in the detection of cell division event, need to relate to the field such as image procossing, pattern recognition.
For the research of small-scale short time, can by the contraction in observation of cell fission process, become round, the mode such as brightness flop is entered
Pedestrian's work mark (such as list of references [4]).But for extensive cell image for a long time, it is unrealistic for manually marking
, the most fissional automatic identification just has great significance.In order to obtain fissional position, shape, cell it
Between the information such as contact, recognizer is the research topic of a lot of scholar quickly and accurately.
Existing can be largely classified into three kinds based on cell division method: method based on track, the method for feature based,
Method based on graph model (such as list of references [5]).Method based on track is the most directly perceived, and it passes through cell in movement locus
The change of visual characteristic carries out the judgement divided, but this method depends on cell tracker, this inherently one the most tired
Difficult research topic；The method of feature based does not relies on cell tracker, can be judged thin by certain feature of image sequence
Born of the same parents divide, but this method substantial amounts of data of needs are as training grader, and cannot obtain fissional particular location；Base
Different from both approaches in the method for graph model, generally include: candidate sequence detection, sequence signature are extracted and sequence classification is several
Individual step.
Following challenge mainly it is faced with in the identification of cell division event:
1, cellular morphology is different.In a width micro-image, usually containing multiple cell, the shape of these cells is respectively arranged with not
With, add difficulty to fissional identification.Meanwhile, different equipment, different cell categories, all can cause cellular morphology
Image presents variation.A kind of detection mode to cellular morphology with robustness is proposed, to fissional identification extremely
Close important.
2, cellular context is complicated.Micro-image is many from culture dish, makes an uproar owing to cell quantity is numerous and there is a lot of image
Point, causes single cell background in the picture complicated, and even up to human eye is difficult to the degree distinguished.
3, existing temporal model has the drawback that and cannot solve the problem that cell division has multiresolution information.Carefully
In the image sequence of born of the same parents' division, comprise the information (such as list of references [6]) of different time resolution, i.e. continuing of four cycles
Time is the most different.And in the middle of the study of graph model, the study redirecting rule between state extremely relies on data.If one
The persistent period of state is longer, the probability of state transition will be made to strain mutually greatly, be unfavorable for the study of parameter.
Summary of the invention
The invention provides the cell division identification method of a kind of multiresolution, the present invention solves in cell division event
The problem that inconsistent the caused temporal model learning difficulty of each phase duration is big so that fissional identification is permissible
The information of the different semantic hierarchies of capture, significantly improves fissional discrimination, described below:
A kind of cell division identification method of multiresolution, described cell division identification method comprises the following steps:
Multiresolution features cell division collection and acellular being divided collection by hidden conditional random fields builds model respectively,
Obtain the model parameter under different resolution；
Arbitrarily choose test cell division candidate sequence as cycle tests, extract visual signature sequence, by visual signature
Sequence inputting model is tested, obtain under different model parameters corresponding sample label be 1 and be 0 probability；
If the probability that sample label is 1 more than be 0 probability, then cell division candidate sequence comprises cell division thing
Part, does not the most comprise cell division event.
Described multiresolution features particularly as follows:
By the method for view-based access control model characteristic similarity, cell data set is grouped, obtain cell division collection and non-carefully
Born of the same parents divide collection, and loop iteration, obtain multiresolution features respectively.
The composition of described cell data set particularly as follows:
Each frame to all cells division candidate sequence carries out the extraction of visual signature, constitutes cell data set.
Described cell division identification method also includes:
The image sequence containing cell is selected as cell division candidate sequence, by choose from cell image sequence
Cell division candidate sequence is divided into the positive sample comprising cell division event and does not comprise the negative sample of cell division event.
The technical scheme that the present invention provides provides the benefit that: this method is obtained by the grouping process of view-based access control model feature
Characteristic sequence under multiresolution so that in model learning, the study of state transition is more efficient, and capture cell division time
Select the high-level semantics information in sequence, significantly improve fissional discrimination；By experimental verification, this method achieves relatively
High recall ratio and precision ratio, meet the multiple needs in actual application.
Accompanying drawing explanation
Fig. 1 is the flow chart of the cell division identification method of a kind of multiresolution.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below embodiment of the present invention is made further
Ground describes in detail.
Embodiment 1
In order to solve cell division image contains the problem of multiresolution information, promote state transition in temporal model
The learning efficiency, embodiments provides the cell division identification method of a kind of multiresolution, sees Fig. 1, this cell division
Recognition methods comprises the following steps:
101: multiresolution features cell division collection and acellular being divided collection by hidden conditional random fields builds respectively
Model, obtains the model parameter under different resolution；
102: arbitrarily choose test cell division candidate sequence as cycle tests, extract visual signature sequence, by vision
Characteristic sequence input model is tested, obtain under different model parameters corresponding sample label be 1 and be 0 probability；
103: if the probability that sample label is 1 more than be 0 probability, then cell division candidate sequence comprises cell division
Event, does not the most comprise cell division event.
Wherein, the multiresolution features in step 101 particularly as follows:
By the method for view-based access control model characteristic similarity, cell data set is grouped, obtain cell division collection and non-carefully
Born of the same parents divide collection, and loop iteration, obtain multiresolution features respectively.
Further, above-mentioned cell data set composition particularly as follows:
Each frame to all cells division candidate sequence carries out the extraction of visual signature, constitutes cell data set.
Further, this cell division identification method also includes:
The image sequence containing cell is selected as cell division candidate sequence, by choose from cell image sequence
Cell division candidate sequence is divided into the positive sample comprising cell division event and does not comprise the negative sample of cell division event.
In sum, embodiments providing the cell division identification method of a kind of multiresolution, this method is passed through
Choosing the image sequence in micro-image data base as data base, in application process, input cell divides candidate sequence, to it
Extract visual signature and set up multi-resolution models and carry out fissional detection, it is thus achieved that preferably recognition result, significantly carrying
High fissional discrimination.
Embodiment 2
Below in conjunction with Fig. 1, the scheme in embodiment 1 is described in detail by concrete Computing Principle, as detailed below retouches
State:
201: from cell image sequence, select the image sequence containing cell as cell division candidate sequence, will choose
To cell division candidate sequence be divided into the positive sample comprising cell division event and do not comprise the negative sample of cell division event；
Wherein, choosing of cell division candidate sequence can pass through the methods such as hand picking, luminance threshold or regional connectivity, this
The choosing method of cell candidate sequence is not limited by inventive embodiments.
202: each frame of all cells division candidate sequence is carried out Visual Feature Retrieval Process, constitutes cell data set
Wherein, N is the total quantity of cell division candidate sequence, and i is the sequence number of candidate sequence sample, xiFor i-th cell
The content of division candidate sequence,Represent that value is the D*T of real numberiDimension space, D is the dimension of visual signature, TiFor i-th
The frame number of cell division candidate sequence.
Wherein, the dimension of D is relevant with selected visual signature, and the embodiment of the present invention is without limitation.yiFor sample
Label, its value is Y={0,1}, and value is 0 to represent and do not comprise cell division event in this sample sequence, and value is 1 to represent this sample
In comprise cell division event；Without loss of generality, the embodiment of the present invention carries out universal search to all cells division candidate sequence
Tree feature (Generalized Search Trees, GIST) (such as list of references [7]) is extracted.
203: by the method for view-based access control model characteristic similarity, cell data set is grouped, obtain cell division collection and
Acellular division collection, and loop iteration, it is thus achieved that multiresolution features；
Wherein, the problem that this step is solved is the problem including multiresolution information in cell division event sequence,
It is embodied as the fissional four-stage persistent period different.If the persistent period in a stage is the most long, in model
During habit, redirect probability between state and will strain mutually greatly, affect model performance.
For a certain sample x, x=[x1,x2,...xT], wherein xi, i ∈ [1, T] is the visual signature of a certain frame, for D
Dimensional vector.The purpose using vision similarity is to weigh the similarity between different frame feature, and obtains difference on this basis
The visual signature of resolution.
Conventional measuring similarity has Euclidean distance, manhatton distance, Chebyshev's distance, COS distance and is correlated with
The modes such as coefficient, the selection of measuring similarity is not limited by the embodiment of the present invention.
204: multiresolution features cell division collection and acellular being divided collection by hidden conditional random fields builds respectively
Model, obtains model parameter w under different resolution*1,w*2,...w*l；
Wherein, the characteristic sequence under different resolution can obtain different model parameters.Specifically,Training process can obtain model parameter w*1,Training process meeting
Obtain model parameter w*2,Training process can obtain model parameter w*l, whereini∈{1,2,…,Tl, for the cell characteristic sequence under different resolution.
205: arbitrarily choose test cell division candidate sequence as cycle tests, it is extracted visual signature sequence x, will
Visual signature sequence inputting model is tested, obtains the Probability p of this sequence sample label corresponding under different model parameters
(y|x；w*)；
Wherein, the extraction of visual signature sequence x can use GIST feature, and this is not restricted by the embodiment of the present invention；w*Bag
Include model parameter w under all resolving powers*1,w*2,...w*l, respectively from step 204 to the feature under different resolving powers
The training process of sequence.
206: if p (y=1 | x；w*) > p (y=0 | x；w*), then cell division candidate sequence comprises cell division thing
Part, does not otherwise comprise cell division event in cell division candidate sequence.
In sum, the embodiment of the present invention utilizes the visual signature of cell division candidate sequence, has preferable robustness,
Eliminate the interference factor impacts on cell division sequence signature such as cellular morphology, color, brightness and background noise.Exist simultaneously
On the basis of temporal model, the complexity that lifting feature does not extracts, further increase the accuracy of cell division event identification,
Yield good result.
Embodiment 3
Below in conjunction with concrete computing formula, the scheme in embodiment 1 and 2 is discussed further, as detailed below retouch
State:
One, without loss of generality, the embodiment of the present invention chooses the COS distance tolerance as visible sensation distance similarity.
The COS distance of vector x and y is defined asX y is inner product of vectors.Should be noted that
It is that the biggest similarity of COS distance is the highest.Concrete grouping process is as follows:
(1) the l layer of cell data set is divided into TlGroupAs l=1,xtFor the t frame feature of cell data set, will each frame feature as one group；For cell data set
T group cell data characteristics in l layer.
(2) in each group, using the frame minimum with remaining frame average distance as representing frame WhereinIt it is the representative frame of l layer t group；dist(xi,xj)
It is the distance of the i-th frame and jth frame, selects COS distance as tolerance
(3) if the distance between the representative frame of two adjacent groups is more than given threshold tau, then two groups are merged, lay equal stress on
Frame is represented, even in newly choosing groupThen
(4) loop iteration, until the most given threshold tau of the distance between all consecutive frames, stops iteration.I.e. to all
T,Set up.
Through above-mentioned steps, characteristic sequence x is divided into the characteristic sequence under l resolving power, is respectivelyWherein,
For t group cell data characteristics under the l resolution of the cell data set after above-mentioned calculating.
Two, concrete model training process is as follows:
(1) conditional probability distribution of acquisition hidden conditional random fields (such as list of references [8]):
Wherein, y is sequence label, and value is 0 or 1；X is the visual signature sequence of candidate sequence, and its dimension is that D*T, D are
Visual signature dimension, T is the frame number of sample；W=[wg,h；wg,d；wy,h；wy,h,h] it is model parameter vector, concrete meaning sees below
Describe in detail；H={h1,h2...,hT, wherein hi∈ H, i ∈ 1,2 ..., T}, H represent all hidden state sets being likely to occur；F
(y,h,x；W) it is characteristic function,For normalization coefficient, y' is sequence label, and value is 0
Or 1.
Characteristic function is:
Section 1 is observational characteristic functionWherein G is gate function structure
The set become；wg,hIt it is the weight on the limit connecting hidden state h and gate function g in graph model；L [] is indicator function；htFor a certain
Grouped data c (xt) hidden state；H' is a certain hidden state in H, and H represents all hidden state sets being likely to occur of t,It is that gate function is to a certain packet c (xtIn), all observations is flat
All export, c (xt) it is the set of multiframe visual signature, c (x time initialt)=xt, t ∈ [1, T], wg,dIt is graph model to connect regard
Feel feature xdWeight with the limit of gate function g；X ' is c (xtEach frame visual signature in).
Section 2 is label characteristics function f2(y,h,t；W)=wy,hL [y=y'] l [ht=h'], wherein wy,hIt it is graph model
The weight on the limit of hidden state h of middle link and sequence label y；L [] is indicator function；
Section 3 is converting characteristic function f3(y,h,t,t+1；W)=wy,h,hL [y=y'] l [ht=h'] l [ht+1=h "],
Wherein wy,h,hIt is that graph model connects adjacent hidden state ht、ht+1Weight with the limit of the sequence label y of t cell data；l
[] is indicator function；ht+1Hidden state for the t+1 moment；H " be a certain hidden state in H', H' represent the t+1 moment all can
The hidden state set that can occur.
(2) L-BFGS Quasi-Newton algorithm (such as list of references [9]) is used to solve optimization problem:
Obtain parameter model w when optimized parameter L (w) takes minima*,For regular terms, parameter w
Obedience variance is σ2Gauss distribution,p(yi|xi；W) it is parameter when being w, cellular sequences xiRight
The sequence label y answerediProbability；wlFor cell the l resolution drag parameter.
That three, tests in characteristic sequence input model concretely comprises the following steps:
(1) test feature sequence x and the most optimized parameter w*As parameter, and by p (y | x；w*) zero setting；
(2) list entries is layered according to the method for view-based access control model characteristic similarity, obtains the spy of different resolution
Levy, then by under different resolution data input corresponding resolution hidden conditional random fields in obtain probability logp (y | xl；w*l),
And logp that probability is added up (y | x；w*)=logp (y | x；w*)+logp(y|xl；w*l)。
When implementing, it is also possible to use other algorithm to solve the problems such as above-mentioned measuring similarity, model solution, this
Inventive embodiments only provides a concrete example and illustrates, and concrete algorithm is realized the step embodiment of the present invention and does not limits
System.
Embodiment 4
Below in conjunction with concrete experiment, the scheme in embodiment 1 and 2 is carried out feasibility checking, described below:
The cell database that experiment is used is from mouse stem cell, optical microscope (Zeiss Axiovert T135V)
Stem cell developmental process is spaced and within five minutes, shoots.Image sequence one has 1013 frames, and the resolution of each image is
1392*1040.In this image sequence, after Image semantic classification, select cell division candidate sequence, use artificial notation methods
It is divided into training sample (295 positive samples comprising cell division event, 204 negative samples not comprising cell division event
This) and test sample (295 positive samples comprising cell division event, 205 negative samples not comprising cell division event),
Each candidate sequence regional resolution is 25*25.About cell type, cell culture environment and data acquisition equipment information and
Parameter arranges and refers to document (such as list of references [10]), and this is not repeated by the embodiment of the present invention.
Through literature query, use scheme of the prior art to carry out recall ratio that cell division identification can reach and
Precision ratio is respectively 75%, and 77%.Carry out the model learning under Multi-resolution and test by the embodiment of the present invention, completed
Cell division identification has reached the recall ratio of 83% and the precision ratio of 86%, its result is better than existing method, it was demonstrated that method
Feasibility and effectiveness.
In sum, the embodiment of the present invention proposes a kind of recognizer that can be used for detecting cell division event, we
Method selects candidate sequence in micro-image, obtains the region being likely to occur cell division event, as cell division data base；
On this basis, extract the visual signature of image sequence, set up multi-resolution models, obtain accurate testing result；And
And the testing result of multiresolution makes Model Abstraction go out the semantic information of candidate sequence, recognition result is finally made more to manage
Think.
List of references:
[1]Oliva A,Torralba A.Modeling the Shape of the Scene:A Holistic
Representation of the Spatial Envelope[J].International Journal of Computer
Vision,2001,42(3):145-175.
[2]Nocedal B J,Wright S J.Numerical Optimization,Springer series in
operations research[J].Siam J Optimization,2012.
[3]Felzenszwalb P F,Huttenlocher D P.Efficient Graph-Based Image
Segmentation[J].International Journal of Computer Vision,2004,59(2):167-181.
[4]Tversky B,Hemenway K.Categories of environmental scenes☆[J]
.Cognitive Psychology,1983,15(1):121-149.
[5]Gou X,Han H C,Hu S,et al.Applying Combined Optical Tweezers and
Fluorescence Microscopy Technologies to Manipulate Cell Adhesions for Cell-
to-Cell Interaction Study[J].IEEE transactions on bio-medical engineering,
2013,60(8):2308-2315.
[6]Chai J,Song Q.Multiple-protein detections of single-cells reveal
cell-cell heterogeneity in human cells.[J].IEEE transactions on bio-medical
engineering,2015,62(1):30-38.
[7]Liu J,Siragam V,Gong Z,et al.Robotic adherent cell injection for
characterizing cell-cell communication.[J].IEEE Transactions on Biomedical
Engineering,2015,62(1):119-125.
[8]Wang Z,Tun L W,Yih M T S,et al.Visual Servoed Three-Dimensional
Cell Rotation System.[J].Biomedical Engineering IEEE Transactions on,2015,62
(10):2498-2507.
[9]Yuan G,Wei Z,Wang Z.Gradient trust region algorithm with limited
memory BFGS update for nonsmooth convex minimization[J].Computational
Optimization&Applications,2013,54(1):45-64.
[10]Nakashima Y,Tsusu K,Hikichi Y,et al.Evaluation of cell-cell or
cell-substrate adhesion effect on cellular differentiation using a microwell
array having convertible culture surface[C]//International Symposium on
Micro-Nanomechatronics and Human Science.IEEE,2014.
It will be appreciated by those skilled in the art that accompanying drawing is the schematic diagram of a preferred embodiment, the invention described above embodiment
Sequence number, just to describing, does not represent the quality of embodiment.
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all spirit in the present invention and
Within principle, any modification, equivalent substitution and improvement etc. made, should be included within the scope of the present invention.
Claims (4)
1. the cell division identification method of a multiresolution, it is characterised in that described cell division identification method includes following
Step:
Multiresolution features cell division collection and acellular being divided collection by hidden conditional random fields builds model respectively, obtains
Model parameter under different resolution；
Arbitrarily choose test cell division candidate sequence as cycle tests, extract visual signature sequence, by visual signature sequence
Input model is tested, obtain under different model parameters corresponding sample label be 1 and be 0 probability；
If the probability that sample label is 1 more than be 0 probability, then cell division candidate sequence comprises cell division event, no
The most do not comprise cell division event.
The cell division identification method of a kind of multiresolution the most according to claim 1, it is characterised in that described many resolutions
Rate feature particularly as follows:
By the method for view-based access control model characteristic similarity, cell data set is grouped, obtains cell division collection and acellular divides
Split collection, and loop iteration, obtain multiresolution features respectively.
The cell division identification method of a kind of multiresolution the most according to claim 2, it is characterised in that described cell number
According to collection composition particularly as follows:
Each frame to all cells division candidate sequence carries out the extraction of visual signature, constitutes cell data set.
4., according to the cell division identification method of a kind of multiresolution described in any claim in claim 1-3, it is special
Levying and be, described cell division identification method also includes:
The image sequence containing cell is selected as cell division candidate sequence, the cell that will choose from cell image sequence
Division candidate sequence is divided into the positive sample comprising cell division event and does not comprise the negative sample of cell division event.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610578288.XA CN106203384B (en) | 2016-07-19 | 2016-07-19 | multi-resolution cell division recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610578288.XA CN106203384B (en) | 2016-07-19 | 2016-07-19 | multi-resolution cell division recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106203384A true CN106203384A (en) | 2016-12-07 |
| CN106203384B CN106203384B (en) | 2020-01-31 |
Family
ID=57492011
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610578288.XA Active CN106203384B (en) | 2016-07-19 | 2016-07-19 | multi-resolution cell division recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106203384B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194319A (en) * | 2017-04-24 | 2017-09-22 | 天津大学 | The mitotic mapping sorted based on SVMs and knowledge method for distinguishing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101980250A (en) * | 2010-10-15 | 2011-02-23 | 北京航空航天大学 | Method for identifying target based on dimension reduction local feature descriptor and hidden conditional random field |
| CN102156988A (en) * | 2011-05-27 | 2011-08-17 | 天津大学 | Cell division sequence detection method |
| CN102722717A (en) * | 2012-05-22 | 2012-10-10 | 天津大学 | Cell division identification method |
| CN105303169A (en) * | 2015-10-16 | 2016-02-03 | 天津大学 | Slow characteristic based cell division recognition method and recognition device thereof |
-
2016
- 2016-07-19 CN CN201610578288.XA patent/CN106203384B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101980250A (en) * | 2010-10-15 | 2011-02-23 | 北京航空航天大学 | Method for identifying target based on dimension reduction local feature descriptor and hidden conditional random field |
| CN102156988A (en) * | 2011-05-27 | 2011-08-17 | 天津大学 | Cell division sequence detection method |
| CN102722717A (en) * | 2012-05-22 | 2012-10-10 | 天津大学 | Cell division identification method |
| CN105303169A (en) * | 2015-10-16 | 2016-02-03 | 天津大学 | Slow characteristic based cell division recognition method and recognition device thereof |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194319A (en) * | 2017-04-24 | 2017-09-22 | 天津大学 | The mitotic mapping sorted based on SVMs and knowledge method for distinguishing |
| CN107194319B (en) * | 2017-04-24 | 2020-09-22 | 天津大学 | Mitosis positioning and identifying method based on support vector machine sequencing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106203384B (en) | 2020-01-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | A multilevel point-cluster-based discriminative feature for ALS point cloud classification | |
| Srivastava et al. | Joint height estimation and semantic labeling of monocular aerial images with CNNs | |
| CN111639679B (en) | Small sample learning method based on multi-scale metric learning | |
| CN104217225B (en) | A kind of sensation target detection and mask method | |
| CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
| CN105844292B (en) | A kind of image scene mask method based on condition random field and secondary dictionary learning | |
| CN108319957A (en) | A kind of large-scale point cloud semantic segmentation method based on overtrick figure | |
| CN106682598A (en) | Multi-pose facial feature point detection method based on cascade regression | |
| Kim et al. | Color–texture segmentation using unsupervised graph cuts | |
| CN104361313B (en) | A kind of gesture identification method merged based on Multiple Kernel Learning heterogeneous characteristic | |
| CN107066559A (en) | A kind of method for searching three-dimension model based on deep learning | |
| CN103390278B (en) | A kind of video unusual checking system | |
| CN107220277A (en) | Image retrieval algorithm based on cartographical sketching | |
| CN106095893A (en) | A kind of cross-media retrieval method | |
| Zhang et al. | Unsupervised difference representation learning for detecting multiple types of changes in multitemporal remote sensing images | |
| CN103942749B (en) | A kind of based on revising cluster hypothesis and the EO-1 hyperion terrain classification method of semi-supervised very fast learning machine | |
| CN109582782A (en) | A kind of Text Clustering Method based on Weakly supervised deep learning | |
| CN105320764B (en) | A kind of 3D model retrieval method and its retrieval device based on the slow feature of increment | |
| CN110135459A (en) | A kind of zero sample classification method based on double triple depth measure learning networks | |
| CN108846404A (en) | A kind of image significance detection method and device based on the sequence of related constraint figure | |
| CN104751463B (en) | A kind of threedimensional model optimal viewing angle choosing method based on sketch outline feature | |
| CN105303169A (en) | Slow characteristic based cell division recognition method and recognition device thereof | |
| Weinmann et al. | A hybrid semantic point cloud classification-segmentation framework based on geometric features and semantic rules | |
| CN103440651B (en) | A kind of multi-tag image labeling result fusion method minimized based on order | |
| Zhang et al. | Abnormal region detection in cervical smear images based on fully convolutional 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |