CN108921133A - Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features - Google Patents
Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features Download PDFInfo
- Publication number
- CN108921133A CN108921133A CN201810842154.3A CN201810842154A CN108921133A CN 108921133 A CN108921133 A CN 108921133A CN 201810842154 A CN201810842154 A CN 201810842154A CN 108921133 A CN108921133 A CN 108921133A
- Authority
- CN
- China
- Prior art keywords
- feature
- multimode
- module
- fusion
- pixel
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/14—Vascular patterns
Abstract
The present invention relates to a kind of multimode unsupervised learning retinal vessel segmenting system based on Fusion Features, including：Image noise reduction and enhancement module enhances colored eye fundus image contrast for denoising to colored eye fundus image；Feature extraction and Fusion Module for extracting the invariant moment features, Hessian matrix character, Gabor wavelet feature, phase equalization feature, Candy boundary operator feature of colored eye fundus image pixel, and are fused to feature vector；Multimode study module for the feature vector of colored eye fundus image pixel to be divided into multiple modules, and clusters respectively；And synthesize with interpretation of result module, it is used to synthesize cluster result, and compare analysis.The Segmentation Method of Retinal Blood Vessels that system proposed by the present invention compensates for supervision must obtain the deficiency of length of difficult and training time etc. using expert's hand labeled blood vessel as goldstandard, there are training sample.
Description
Technical field
The present invention relates to eye fundus image analysis technical field, unsupervised of especially a kind of multimode based on Fusion Features
Practise retinal vessel segmenting system.
Background technique
The systemic diseases such as retinal vasculopathy is diabetes, hypertension, cardiovascular disease provide important information, this
A little diseases frequently can lead to retinal vessel and the variation such as branch bifurcation, bending occur.Such as in hypertensive patient, arteria retina
It will lead to intermittent contractions；Diabetic retinopathy can cause retinal blood enlargement of pipe and locking, macular edema, go out
Blood, exudation and ischaemic episodes.Retinal vessel segmentation is used for early screening from colored eye fundus image and adjuvant clinical is examined
It is disconnected to be of great significance.
Currently, Segmentation Method of Retinal Blood Vessels is broadly divided into two classes：One kind is the supervised learning method based on feature, such as
Using k neighbour (kNearest Neighbor, KNN), support vector machines (Support Vector Machine, SVM), nerve
Network (Neural Network, NN), gauss hybrid models (Gaussian mixture model, GMM), extreme learning machine
(Extreme Learning Machine, ELM), AdaBoost or convolutional neural networks (Convolutional Neural
Networks, CNN) etc. carry out retinal vessel segmentation.The retina that supervised learning method has usually been marked with ophthalmologist, oculist
Blood vessel is standard, extracts the pixel characteristic of eye fundus image, obtains classifier by training, this classifier is used for retinal vessel
Segmentation.Although above-described Segmentation Method of Retinal Blood Vessels segmentation effect is pretty good, there must be expert's manual segmentation blood vessel for gold
The deficiencies of standard carries out supervised learning, obtains difficulty and long training time there are training sample.
Another kind of is unsupervised learning method, is based on blood based on multiple dimensioned method including the method based on matched filtering
The method of pipe tracking, the method based on mathematical morphology, the method based on threshold transition, the method etc. based on model.It is unsupervised
Learning method then without expert's hand labeled retinal vessel, according to the tree of retinal vessel, blood vessel width threshold value, is divided
The characteristics such as branch angle directly carry out blood vessel segmentation.Unsupervised learning method is quick and easy, but to the accurate of retinal vessel segmentation
Degree usually less than has measure of supervision.
Summary of the invention
The multimode unsupervised learning retinal vessel segmentation based on Fusion Features that the purpose of the present invention is to provide a kind of
System, to overcome defect existing in the prior art.
To achieve the above object, the technical scheme is that：A kind of multimode unsupervised learning based on Fusion Features
Retinal vessel segmenting system, including：
Image noise reduction and enhancement module enhances colored eye fundus image contrast for denoising to colored eye fundus image；
Feature extraction and Fusion Module, invariant moment features, the Hessian matrix for extracting colored eye fundus image pixel are special
Sign, Gabor wavelet feature, phase equalization feature, Candy boundary operator feature, and it is fused to feature vector；
Multimode study module, for the feature vector of colored eye fundus image pixel to be divided into multiple modules, and respectively
Cluster；
And synthesize with interpretation of result module, it is used to synthesize cluster result, and compare analysis, exports result.
In an embodiment of the present invention, it is denoised by described image and enhances module and extract the green of the colored eye fundus image
Chrominance channel carries out gaussian filtering smoothing denoising, using the bottom cap transformation enhancing colored eye fundus image medium vessels and background contrast
Degree.
In an embodiment of the present invention, colored eye fundus image pixel is extracted by the feature extraction and Fusion Module
2 not bending moment it is several for embodying that blood vessel line segment is different with width, angle and direction is variant as the invariant moment features
What feature.
In an embodiment of the present invention, each pixel is chosen in default scale by the feature extraction and Fusion Module
Under maximum confidence as Hessian matrix character, for embodying blood vessel line feature.
In an embodiment of the present invention, by the feature extraction and Fusion Module choose each pixel default scale with
The response maximum value result of preset direction is overlapped as Gabor wavelet feature.
In an embodiment of the present invention, consistent with the phase that Fusion Module extracts each pixel by the feature extraction
Property is as phase equalization feature.
In an embodiment of the present invention, it is mentioned by the feature extraction and Fusion Module using default Candy boundary operator
Take vessel boundary feature as Candy boundary operator feature.
In an embodiment of the present invention, the feature extraction and Fusion Module are by the invariant moment features, the Hessian
Matrix character, the Gabor wavelet feature, the phase equalization feature, the Candy boundary operator feature and gray value melt
It is combined into 18 dimensional feature vectors.
In an embodiment of the present invention, piecemeal study is carried out to colored eye fundus image by the multimode study module,
Feature vector is divided into 4 submodules, k-means clustering algorithm is respectively adopted, calculates each pixel characteristic to cluster centre
Distance, obtain average value；Blood vessel pixel and background pixel are divided by updating iteration cluster centre, the time is reduced and space is opened
Pin.
In an embodiment of the present invention, each submodule cluster result is carried out by the synthesis and interpretation of result module
Synthesis obtains retinal vessel segmentation result, and is that standard compares with the blood-vessel image of pre-stored expert's hand labeled
To analysis.
Compared to the prior art, the invention has the advantages that：Multimode proposed by the present invention based on Fusion Features
Block unsupervised learning retinal vessel segmenting system realizes the segmentation of retinal vessel, including image denoising enhancing, feature mention
It takes and merges, multimode learns, and synthesis and interpretation of result facilitate area of computer aided clinical diagnosis fundus oculi disease.The present invention is more
Mended supervision Segmentation Method of Retinal Blood Vessels must using expert's hand labeled blood vessel as goldstandard, there are training sample obtain it is tired
The deficiency of length of difficult and training time etc..The present invention is of far-reaching significance in ophthalmology, and multimode unsupervised learning extracts view
Retinal vasculature can reduce the burden of eye doctor's duplication of labour, improve accuracy and work that oculist judges fundus oculi disease
Efficiency provides more accurate foundation for fundus oculi disease early screening and clinical diagnosis.
Detailed description of the invention
Fig. 1 is feature extraction and Fusion Module schematic diagram in one embodiment of the invention.
Fig. 2 is multimode study module schematic diagram in one embodiment of the invention.
Specific embodiment
With reference to the accompanying drawing, technical solution of the present invention is specifically described.
A kind of multimode unsupervised learning retinal vessel segmenting system based on Fusion Features of the present invention, including：Image
Denoising enhancing module, feature extraction and Fusion Module, multimode study module, synthesis and interpretation of result module.
Image noise reduction and enhancement module is used to enhance picture contrast processing to image denoising；
Feature extraction and Fusion Module are used to extract invariant moment features, the Hessian matrix character, Gabor of image pixel
Wavelet character, phase equalization feature, Candy boundary operator feature are simultaneously fused to feature vector；
Multimode study module is for being divided into multiple modules for the feature vector of image pixel and clustering respectively；
Synthesis and interpretation of result module are for synthesizing cluster result and comparing analysis.
Further, in the present embodiment, since colored eye fundus image is RGB image, also known as true color image.On image
Each pixel color be to be determined by the brightness value of three color of red, green, blue.Picture contrast is most ideal on green channel, blood vessel wheel
Wide the most obvious with background difference, noise jamming is also few, is conducive to blood vessel segmentation, therefore in the present embodiment, from colored eyeground figure
As carrying out denoising and enhancing processing in accordance with the following steps：
1, the green channel component for choosing colored eyeground figure carries out gaussian filtering pretreatment, it is therefore an objective to reduce noise；
2, cap transformation in bottom is carried out to the image after gaussian filtering process, enhances the contrast of blood vessel and background pixel.Simultaneously
Also using treated gray value of image as one of feature.
Further, in the present embodiment, different feature extracting methods can describe the different structure characteristic of blood vessel, such as
Width, direction, threadiness and edge, vessel information can more fully be embodied by merging these features single features being relatively used only.This reality
It is special by the invariant moment features of extraction eye fundus image, Hessian matrix character, Gabor wavelet feature, phase equalization to apply example
Sign, Candy boundary operator feature, and feature vector is fused to realize that retinal vessel is divided.
Further, in the present embodiment, the acquisition modes of above-mentioned each feature are as follows：
1, the blood vessel in retina is piecewise linearity, and blood vessel line width is different, each not phase of direction, angle and shape
Together, in order to preferably be fitted these blood vessel line segments, blood vessel is described using having the not bending moment of translation, rotation and scale invariability
Feature.Centered on each pixel (i, j), the subgraph of building 3 × 3, by 2 dimension (p+q) rank geometric moments of each subgraph in
Heart square is defined as：
Wherein, p, q=1,2,3 ..., f (i, j) are 3 × 3 size subgraphs centered on each pixel (i, j),It is influenced to eliminate image scaled variation bring, definition normalization central moment is as follows：
Wherein, p, q=1,2,3 ...；Bending moment does not have 7 to Hu, selects 2 invariant moment features Φ1=η20
+η02With
2, retinal vessel is local linear structure and width is different, using the Hessian matrix sensitive to linear structure
Extract blood vessel pixel characteristic.Hessian matrix can be expressed as
Wherein, Lxx,Lxy,Lyx,LyyThe respectively second-order partial differential coefficient of L, L (x, y, σ)=G (x, y, σ) * I (x, y), G (x, y,
It is σ) standard deviation for σ Gaussian function, σ is the scale space factor, and I is original image, and (x, y) is location of pixels, and * indicates convolution behaviour
Make.For linear structure, Hessian matrix exgenvalue λ1And λ2For describing blood vessel and thread in retinal images, therefore
Blood vessel confidence level function is introduced to describe blood vessel intensity.Calculate blood vessel pixel confidence level formula be：
Wherein, | λ1|≤|λ2|, RB=λ1/λ2,β=0.5,And choose each pixel
Maximum confidence of the point under scale σ=0.5 and σ=2.5 is as final output.
3, phase equalization is a kind of method that Fourier space component and maximum point are characterized a little in detection image, right
The line feature of blood vessel has detection property well, and this method is applied to the processing of colored eye fundus image, can be well solved
The brings blood vessel segmentations such as picture contrast is not high, noise jamming is strong are difficult.Therefore, extract the phase equalization of each pixel
It is represented by
Wherein, o indicates that the index in direction, n indicate dimensional information, ToFor noise compensation, AnoFor amplitude,Represent the phase difference of x, WoAs weighting function, when frequency spectrum is narrow, it can change filter and exists
The frequency response characteristic of corresponding position.It is defined asWherein, the cutoff frequency that c is responded as filter,
Here c=0.4 is taken；R is the gain that gain parameter is used to control filter, takes r=10；The calculation method of S (x) is by filter
The sum of response amplitude divided by image all pixels point response amplitude maximum value；ε is constant, and when too small in amplitude, phase is consistent
Property problem will become unstable, need to avoid denominator plus a ε at this time to be 0, ε=0.0001 here.Finally by phase one
Cause property is as one of extraction blood vessel feature.
4, multiple dimensioned Gabor wavelet, which is converted, has good direction and scale selection characteristic when extracting direction character,
It can detecte the blood vessel of different directions and different scale, rotate angle such as with 15 ° for interval, 12 directions from 0 ° to 180 °,
Scale a=2 and a=5.And Gabor wavelet transformation is insensitive to illumination variation, can overcome eye ground image irradiation not
Equal influence selects Gabor wavelet transformation to carry out the extraction of retinal vessel feature.Gabor wavelet converts Tψ(b, θ a) are defined
For：
Wherein, CψIndicate normaliztion constant, ψ indicates analysis wavelet, and parameter b, a, θ respectively indicate expansion scale, translational movement
With rotation angle.rθIt is rotation operator, is represented by rθ(x)=(x cos θ-y sin θ, x sin θ+y cos θ), θ ∈ [0,2
π], f is original image,AndIndicate corresponding Fourier transformation.Formula 8) ink0For complex exponential
Frequency vector.Parameter ε=4, k are set0=[0,2.7] not only contributes to examine the direction character of retinal vessel in this way
When surveying, while ensuring that retinal images carry out Gabor wavelet transformation, the response of target blood pixel is higher than background pixel
Point.Different directions and different sizes for detection blood vessel, take respectively at scale a=2 and a=5, rotate angle, θ with 15 ° for step
Length is rotationally-varying from 0 °~180 °, and the response maximum value result in totally 12 directions is overlapped as last output.
5, Candy edge detection operator has good signal-to-noise ratio, has unique response to single edge, preferable to position
The advantages that performance, is capable of detecting when vessel boundary.Multi-scale edge extraction is carried out using Candy edge detection operator.It uses first
Gaussian filtering template (5 × 5) is filtered to eliminate noise eyeground figure.Image is found in the direction x and the direction y with derivative operator
Partial derivative (Gx,Gy), and find out gradient magnitude and direction.Again the gradient direction at edge be divided into 4 directions (0 °, 45 °, 90 °,
135°).The gradient value of pixel is compared with different neighbor pixels, to decide whether being marginal point.This process is known as
" non-maximum restraining ".Finally two T are calculated using accumulative histogramhigh, TlowThreshold value, it is all to be greater than ThighBe then marginal point,
It is all to be less than TlowBe not then marginal point.What if testing result had fallen between, would see that the neighbor pixel of this pixel is
It is no to have more than Thigh, it is marginal point if having, is not otherwise just marginal point.Pass through the processing of the above method, retinal vessel
Result of the image after Candy edge detection is as one of feature.
6, different feature extracting methods can describe the different structure characteristic of blood vessel, such as width, direction, threadiness and side
Edge, vessel information can more fully be embodied by merging these features single features being relatively used only.It is carried out using the method for Fusion Features
Blood vessel segmentation.As shown in Fig. 2, providing various characteristic patterns and segmentation result.As shown in Figure 1, using pretreated gray value as
The different characteristic vector of feature vector and extraction, including the dimension of Gabor wavelet feature 12, the dimension of invariant moment features 2, Hessian feature 1
Dimension, the dimension of phase property 1, Candy feature 1 are tieed up, and fusion constructs carry out blood vessel segmentation at 18 dimensional feature vector, segmentation result compared with
Better than single features.
Further, in the present embodiment, every width eyeground figure size is 584*565 in used DRIVE database, altogether
329960 pixel samples points, 18 dimensional feature vectors of each pixel samples point, feature samples number is more, using multimode k-
Means clustering method can efficiently reduce room and time expense.Multimode study module is provided in the present embodiment, according to as follows
Mode is realized：
Using all pixels feature of each image as sample set；Sample set is divided into 4 submodules and carries out k-means respectively
Cluster calculates each pixel characteristic to the distance of cluster centre, obtains average value；Blood vessel is divided by updating iteration cluster centre
Pixel and background pixel reduce time and space expense.After obtaining 4 segmentation results, pass through synthesis and interpretation of result module pair
Each submodule cluster result is synthesized, synthesis output retinal vessel segmentation result.As shown in Fig. 2, specific step is as follows：
Step 1：Feature samples collection is divided into 4 module { I according to size m × n of image1,...I4}；
Step 2：Each module obtains segmentation result { label using k-means clustering learning1,...label4}；
Step 3：Combination and segmentation result label.
Further, in the present embodiment, by synthesis and interpretation of result module, by retinal vessel segmentation result and in advance
The blood-vessel image of the expert's hand labeled first stored is compared for standard.
The above are preferred embodiments of the present invention, all any changes made according to the technical solution of the present invention, and generated function is made
When with range without departing from technical solution of the present invention, all belong to the scope of protection of the present invention.
Claims (10)
1. a kind of multimode unsupervised learning retinal vessel segmenting system based on Fusion Features, which is characterized in that including：
Image noise reduction and enhancement module enhances colored eye fundus image contrast for denoising to colored eye fundus image；
Feature extraction and Fusion Module, for extract the invariant moment features of colored eye fundus image pixel, Hessian matrix character,
Gabor wavelet feature, phase equalization feature, Candy boundary operator feature, and it is fused to feature vector；
Multimode study module for the feature vector of colored eye fundus image pixel to be divided into multiple modules, and clusters respectively；
And synthesize with interpretation of result module, it is used to synthesize cluster result, and compare analysis, exports result.
2. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, the green channel that module extracts the colored eye fundus image is denoised and enhanced by described image, carries out Gauss filter
Wave smoothing denoising, using the bottom cap transformation enhancing colored eye fundus image medium vessels and background contrasts.
3. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
Be characterized in that, extract 2 of colored eye fundus image pixel bending moment be used as by the feature extraction and Fusion Module described in
Invariant moment features have the geometrical characteristic that width is different, angle and direction is variant for embodying blood vessel line segment.
4. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, maximum confidence conduct of each pixel under default scale is chosen by the feature extraction and Fusion Module
Hessian matrix character, for embodying blood vessel line feature.
5. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, it is maximum in the response of default scale and preset direction to choose each pixel by the feature extraction and Fusion Module
Value result is overlapped as Gabor wavelet feature.
6. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, the phase equalization of each pixel is extracted as phase equalization spy by the feature extraction and Fusion Module
Sign.
7. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, the conduct of vessel boundary feature is extracted using default Candy boundary operator by the feature extraction and Fusion Module
Candy boundary operator feature.
8. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, the feature extraction and Fusion Module are by the invariant moment features, the Hessian matrix character, the Gabor
Wavelet character, the phase equalization feature, the Candy boundary operator feature and gray value are fused to 18 dimensional feature vectors.
9. the multimode unsupervised learning retinal vessel segmenting system according to claim 1 based on Fusion Features,
It is characterized in that, piecemeal study is carried out to colored eye fundus image by the multimode study module, feature vector is divided into 4
K-means clustering algorithm is respectively adopted in submodule, calculates each pixel characteristic to the distance of cluster centre, obtains average value；It is logical
It crosses update iteration cluster centre and divides blood vessel pixel and background pixel, reduce time and space expense.
10. the multimode unsupervised learning retinal vessel segmenting system according to claim 9 based on Fusion Features,
It is characterized in that, each submodule cluster result is synthesized by the synthesis and interpretation of result module, obtains retinal blood
Pipe segmentation result, and be compared with the blood-vessel image of pre-stored expert's hand labeled for standard.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810842154.3A CN108921133A (en) | 2018-07-27 | 2018-07-27 | Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810842154.3A CN108921133A (en) | 2018-07-27 | 2018-07-27 | Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108921133A true CN108921133A (en) | 2018-11-30 |
Family
ID=64417427
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810842154.3A Pending CN108921133A (en) | 2018-07-27 | 2018-07-27 | Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921133A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109998681A (en) * | 2019-03-16 | 2019-07-12 | 哈尔滨理工大学 | A kind of inner cavity image pre-processing method for distinguishing areas of specular reflection and blood vessel |
| CN110796086A (en) * | 2019-10-30 | 2020-02-14 | 中国科学院宁波工业技术研究院慈溪生物医学工程研究所 | Iris segmentation method of AS-OCT image based on local phase tensor algorithm |
| CN111950714A (en) * | 2020-08-24 | 2020-11-17 | 重庆市云迈科技有限公司 | Energy spectrum CT image domain material identification method based on 3D full convolution neural network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104881862A (en) * | 2015-04-03 | 2015-09-02 | 南通大学 | Retinal vascular tortuosity calculation method based on ophthalmoscope image and application thereof |
| CN107248161A (en) * | 2017-05-11 | 2017-10-13 | 江西理工大学 | Retinal vessel extracting method is supervised in a kind of having for multiple features fusion |
-
2018
- 2018-07-27 CN CN201810842154.3A patent/CN108921133A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104881862A (en) * | 2015-04-03 | 2015-09-02 | 南通大学 | Retinal vascular tortuosity calculation method based on ophthalmoscope image and application thereof |
| CN107248161A (en) * | 2017-05-11 | 2017-10-13 | 江西理工大学 | Retinal vessel extracting method is supervised in a kind of having for multiple features fusion |
Non-Patent Citations (5)
| Title |
|---|
| JOAO V. B. SOARES 等: "Retinal Vessel Segmentation Using the 2-D Gabor Wavelet and Supervised Classification", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
| 曹建农: "《图像分割方法研究》", 31 August 2006 * |
| 朱承璋 等: "基于多特征融合和随机森林的视网膜血管分割", 《计算机辅助设计与图形学学报》 * |
| 焦李成 等: "《雷达图像解译技术》", 31 December 2017 * |
| 谢剑斌 等: "《视觉感知与智能视频监控》", 31 March 2012 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109998681A (en) * | 2019-03-16 | 2019-07-12 | 哈尔滨理工大学 | A kind of inner cavity image pre-processing method for distinguishing areas of specular reflection and blood vessel |
| CN110796086A (en) * | 2019-10-30 | 2020-02-14 | 中国科学院宁波工业技术研究院慈溪生物医学工程研究所 | Iris segmentation method of AS-OCT image based on local phase tensor algorithm |
| CN111950714A (en) * | 2020-08-24 | 2020-11-17 | 重庆市云迈科技有限公司 | Energy spectrum CT image domain material identification method based on 3D full convolution neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mary et al. | Retinal fundus image analysis for diagnosis of glaucoma: a comprehensive survey | |
| CN110503649B (en) | Liver segmentation method based on spatial multi-scale U-net and superpixel correction | |
| CN106204555B (en) | A kind of optic disk localization method of combination Gbvs model and phase equalization | |
| Li et al. | Automated feature extraction in color retinal images by a model based approach | |
| Ramakanth et al. | Approximate nearest neighbour field based optic disk detection | |
| WO2022001571A1 (en) | Computing method based on super-pixel image similarity | |
| Dai et al. | Optic disc segmentation based on variational model with multiple energies | |
| Streeter et al. | Microaneurysm detection in colour fundus images | |
| CN108986106A (en) | Retinal vessel automatic division method towards glaucoma clinical diagnosis | |
| CN108921133A (en) | Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features | |
| Gao et al. | An effective retinal blood vessel segmentation by using automatic random walks based on centerline extraction | |
| Rodrigues et al. | Retinal vessel segmentation using parallel grayscale skeletonization algorithm and mathematical morphology | |
| Mendonça et al. | Segmentation of the vascular network of the retina | |
| Zhang et al. | TUnet-LBF: Retinal fundus image fine segmentation model based on transformer Unet network and LBF | |
| CN108665474B (en) | B-COSFIRE-based retinal vessel segmentation method for fundus image | |
| CN113947805A (en) | Eye shake type classification method based on video image | |
| Fang et al. | Retinal vessel landmark detection using deep learning and hessian matrix | |
| Zhou et al. | A novel approach for red lesions detection using superpixel multi-feature classification in color fundus images | |
| Ali et al. | Optic Disc Localization in Retinal Fundus Images Based on You Only Look Once Network (YOLO). | |
| Vázquez et al. | Using retinex image enhancement to improve the artery/vein classification in retinal images | |
| Abdullah et al. | Application of grow cut algorithm for localization and extraction of optic disc in retinal images | |
| Desiani et al. | A robust techniques of enhancement and segmentation blood vessels in retinal image using deep learning | |
| Honale et al. | A review of methods for blood vessel segmentation in retinal images | |
| Rodtook et al. | Optic disc localization using graph traversal algorithm along blood vessel in polar retinal image | |
| CN112017132A (en) | Vein image enhancement method based on maximum curvature method and multi-scale Hessian matrix |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181130 |