CN109003279A - Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model - Google Patents
Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model Download PDFInfo
- Publication number
- CN109003279A CN109003279A CN201810733535.8A CN201810733535A CN109003279A CN 109003279 A CN109003279 A CN 109003279A CN 201810733535 A CN201810733535 A CN 201810733535A CN 109003279 A CN109003279 A CN 109003279A
- Authority
- CN
- China
- Prior art keywords
- image
- filter
- blood vessel
- pixel
- response
- 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
- 238000000034 method Methods 0.000 title claims abstract description 72
- 210000004204 blood vessel Anatomy 0.000 title claims abstract description 63
- 230000011218 segmentation Effects 0.000 title claims abstract description 53
- 210000001525 retina Anatomy 0.000 title claims abstract description 11
- 238000003064 k means clustering Methods 0.000 title claims abstract description 10
- 238000002372 labelling Methods 0.000 title abstract 3
- 238000012549 training Methods 0.000 claims abstract description 51
- 238000012360 testing method Methods 0.000 claims abstract description 37
- 238000000605 extraction Methods 0.000 claims abstract description 31
- 230000008569 process Effects 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 16
- 239000013598 vector Substances 0.000 claims abstract description 15
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 11
- 238000010801 machine learning Methods 0.000 claims abstract description 7
- 230000004044 response Effects 0.000 claims description 64
- 238000001914 filtration Methods 0.000 claims description 35
- 230000006870 function Effects 0.000 claims description 18
- 230000004256 retinal image Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000001514 detection method Methods 0.000 claims description 11
- 230000002792 vascular Effects 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 241000669244 Unaspis euonymi Species 0.000 claims description 2
- 239000008280 blood Substances 0.000 claims description 2
- 210000004369 blood Anatomy 0.000 claims description 2
- 239000003086 colorant Substances 0.000 claims description 2
- 239000006227 byproduct Substances 0.000 claims 1
- 230000004069 differentiation Effects 0.000 claims 1
- 210000001210 retinal vessel Anatomy 0.000 abstract description 7
- 238000013145 classification model Methods 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 230000002207 retinal effect Effects 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 206010012689 Diabetic retinopathy Diseases 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 210000003462 vein Anatomy 0.000 description 2
- 206010003211 Arteriosclerosis coronary artery Diseases 0.000 description 1
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 208000034189 Sclerosis Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000036772 blood pressure Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 206010012601 diabetes mellitus Diseases 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000004089 microcirculation Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004233 retinal vasculature Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 235000015170 shellfish Nutrition 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009885 systemic effect Effects 0.000 description 1
- 210000005166 vasculature Anatomy 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Eye Examination Apparatus (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a fundus retina blood vessel segmentation method and system based on a K-Means clustering labeling and naive Bayes model. The method comprises the following steps: randomly extracting color fundus images in the data set, and constructing a training set and a testing set; extracting a gray scale image of a G channel of the color fundus image as a feature extraction object; extracting the characteristics of the gray level image, and expressing each pixel in the image by using a multi-dimensional characteristic vector; clustering and labeling the feature vectors of each image in the training set after feature extraction by using a K-Means clustering algorithm; training a naive Bayes model based on training set data of K-Means clustering labels; and (5) segmenting the blood vessels of each image in the test set by using a trained naive Bayes model. The invention takes the clustering result as the mark with supervision training, and trains the naive Bayes classification model by using the mark to segment the retinal vessel, so that the whole process does not need to be manually marked, time and labor are saved, and the processing efficiency of the machine learning model is greatly improved.
Description
Technical field
The present invention relates to field of medical image processing, specifically, more particularly to a kind of based on K-Means cluster mark
With the eye ground blood vessel segmentation method of model-naive Bayesian.
Background technique
Retinal vessel is the important component of systemic microcirculation system, the variation and diabetes, height of morphosis
The coincident with severity degree of condition of the cardiovascular diseases such as blood pressure, coronary sclerosis and cerebrovascular sclerosis is closely related.Therefore, retina
The research of image blood vessel segmentation technology has a very important significance clinical medicine application.
Current eye ground blood vessel method mainly has two general orientation: rule-based blood vessel segmentation method and based on learn
The blood vessel segmentation method of habit.What rule-based blood vessel segmentation method mainly utilized is image processing techniques, according to vessel properties
Some rules are designed to find blood vessel: different scale lower linear filter is devised according to the piecewise linearity of blood vessel and connectivity,
By combining filter result and the thresholding vessel segmentation that handles to the end under different scale;It is transversal due to blood vessel
Face at handstand Gaussian Profile, so dimensional Gaussian matched filtering method using characterization different directions 12 Gaussian templates as
Matched filter is filtered image, then carries out thresholding to response results, chooses and responds maximum matched filtering knot
Fruit exports as blood vessel, finally extracts retinal vascular images.Filtering method based on Hessian matrix utilizes Hessian
The characteristic value of matrix enhances all blood vessel structures including vascular bifurcation, and inhibits non-vascular structure.These image procossings
Method requires thresholding processing, and needs to select different threshold values for different retinal fundus images, this is not
The process of one automation.
Blood vessel segmentation problem is considered as to the classification problem of Pixel-level based on the dividing method of study, has label by one group
Training dataset training obtain classifier.Wherein, feature mainly includes that grey scale pixel value, boundary characteristic and Gabor wavelet become
It changes, classifier mainly includes multilayer neural network, Bayes classifier, random forest and support vector machines.Due to introducing
Differentiate the powerful mechanism of inquiry learning, the segmentation performance of these methods is often better than rule-based dividing method.Based on study
Method can divide automatically and accuracy rate is higher, and main flow is pre-processing image data, image characteristics extraction and model instruction
Practice.Difficult point is that retinal fundus images feature extraction and Pixel-level label data obtain.Having the machine learning techniques of supervision needs
The artificial labeled data of a large amount of expert is wanted, and the retinal vessel in eye fundus image is in the tree-shaped and the tiny blood of tip of diverging
The presence of pipe makes label task more difficult.
In computer-aided medical science image analysis, can from a large amount of medical image of infection from hospital as training set, but
If often unpractical it is required that medical expert is identified the blood vessel in these images.If only used a small amount of
Have label example, then the learning system trained using them is often difficult have strong generalization ability;On the other hand, such as
Fruit has label example without being then to data resource using a large amount of " inexpensively " unmarked examples using only a small amount of " expensive "
Greatly waste.
Summary of the invention
The a large amount of artificial labeled data of expert is needed according to the machine learning techniques for having supervision set forth above, it is time-consuming and laborious
The technical issues of, and a kind of eye ground blood vessel segmentation based on K-Means cluster mark and model-naive Bayesian is provided
Method.The present invention mainly using the result of K-Means cluster mark as the label of data, then goes to train classification using it again
Model is split, to improve work efficiency, segmentation effect is more preferable, saves manpower and material resources.The technology that the present invention uses
Means are as follows:
A kind of eye ground blood vessel segmentation method based on K-Means cluster mark and model-naive Bayesian, including
Following steps:
The Pixel-level label of training set image is constructed, specifically: each image in the training set after feature extraction is made
Feature vector is subjected to cluster mark with K-Means clustering algorithm;
Training set data training model-naive Bayesian based on K-Means cluster mark;
With the blood vessel of each image in trained model-naive Bayesian segmentation test set.
Further, also there are following steps before the Pixel-level label of the step building training set image:
The colored eye fundus image in data set is randomly selected, training set and test set are constructed;
Object of the grayscale image in the channel G of colored eye fundus image as feature extraction is extracted, and to retina region of interest
Domain carries out expansion process;
Feature extraction is carried out to the grayscale image after expansion process, by each pixel in training set and test set image
It is indicated with the feature vector of ten dimensions, specifically:
The contrast for enhancing the grayscale image by the adaptive histogram equalization method of contrast-limited, by this gray scale
It is worth as a kind of feature for distinguishing blood vessel and background, gray scale described in technical treatment is enhanced by the blood vessel designed based on vessel properties
Figure,
The grayscale image is filtered as follows respectively:
By Gabor wavelet conversion process, four-dimensional Gabor characteristic is obtained;
Retinal images are filtered by linear detector, obtain the response results of linearity test as one-dimensional
Linearity test feature;
Retinal images are filtered by the filter of gaussian shape, chooses and responds maximum filter result work
For one-dimensional Gauss matching characteristic;
Frangi filtering characteristics based on Hessian matrix are filtered retinal images, choose peak response
As one-dimensional Frangi filtering characteristics;
Retinal images are filtered based on B-COSFIRE filtering characteristics, by symmetrical B-COSFIRE filter
It sums to obtain final one-dimensional B-COSFIRE filtering characteristics with asymmetric B-COSFIRE filter response results;
The one-dimensional weber Expressive Features that operator solves any pixel point in retinal images are described by weber;
It is described as follows with also having after the blood vessel of each image in trained model-naive Bayesian segmentation test set
Step:
Judge whether blood vessel segmentation method is effective by preset evaluation index.
Further, the step randomly selects the colored eye fundus image in data set, constructs in training set and test set,
Data set uses disclosed DRIVE data set, and the training set is identical with the picture number of test set, and each image corresponding 2
The Mask image in the effective information region of the result of a expert's manual segmentation and 1 display eye fundus image.
Further, Gabor wavelet conversion process is carried out in the following way:
Gabor kernel function is as follows:
Wherein, x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ,
(u0,v0) indicate the temporal frequency of sinusoidal plane wave, σxAnd σyIndicate the variance of the oval Gauss on the direction x and y,
Which determine the size of the filter zone of action, i.e. scale, θ is rotation angle,
One group of direction Gabor wavelet different with scale, which is obtained, by change parameter σ and θ compares institute under fixed size
There is the response on direction, using peak response as the characteristic image under the scale, according to the scale of setting, under all scales
As a result it is used as Gabor characteristic.
Further, the linear detector is by under the window of W*W, length of rotation be the detection line of l obtain it is N number of not
Equidirectional filter is filtered grayscale image with filter, and when the direction of detection line is consistent with vessel directions, response is most
Greatly, compare it is angled under response retain maximum value as the testing result under the windowBy the length for changing detection line
Degree adapts to the blood vessel under different scale, and the response results summation under different scale is averagely obtained to the response results of linearity test again
As linearity test feature.
Further, the filter graph picture filtering processing of gaussian shape is carried out in the following way:
The Gaussian kernel function of handstand is as follows:
WhereinThe mean normalization of filter to 0, L is indicated along y-axis
The neighborhood length of smooth noise, t are constants, setting x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ, θ=(0:12) * π/
12 change to detect the blood vessel of different directions, and the peak response on N number of direction is chosen under a certain scale as under the scale
Filter result compares the response under different scale, chooses and responds maximum filter result as Gauss matching characteristic.
Further, the Frangi filtering processing based on Hessian matrix is carried out in the following way:
For two-dimensional input image, obtain constructing each pixel (x, y) with the second order local derviation convolution of gaussian kernel function
Hessian matrix,
The result of Frangi filtering is calculated by such as minor function:
Wherein, the characteristic value of Hessian matrix is λ1And λ2, and | λ1| < | λ2|, Rb=λ2/λ1,
β controls the parameter of whole smoothness for adjusting linear and block-like difference, c statement, and more each pixel exists
Response under each scale chooses peak response as Frangi filtering characteristics.
Further, the Difference of Gaussian filter that B-COSFIRE filter is collinearly arranged by one group, i.e. DoG filter structure
At B-COSFIRE filter response is obtained by calculating the geometric mean of all DoG filter responses, and the DoG filter response is logical
The convolution for crossing DoG filter and image obtains,
For extracting continuous blood vessel structure feature, DoG filter exists the COSFIRE model of symmetrical structure in pairs, they
Center be located at the two sides of B-COSFIRE filter center in the opposite direction;The COSFIRE model of unsymmetric structure is for detecting
Tail vein, the center of DoG filter are located at the side of B-COSFIRE filter center, obscure to DoG filter response
Operation, the fuzzy operation are the maximum value of DoG filter weight threshold response, and weight is DoG filter response and Gaussian function
Final B- is made and obtained to symmetrical B-COSFIRE filter and asymmetric B-COSFIRE filter response result by the product of coefficient
COSFIRE filtering characteristics.
Further, pass through the weber Expressive Features of any pixel point in following equations retinal images:
Wherein, p is any pixel point in image I, and N (p) indicates the neighborhood territory pixel set of pixel p, and I (z) indicates target picture
The gray value of any pixel of N (p) in the neighborhood of plain p.
It is further, described to judge whether blood vessel segmentation method is effective by preset evaluation index, specifically:
Wherein, the puncta vasculosa of each image is distinguished by four kinds of different colors, the puncta vasculosa true positives TP correctly divided,
The non-vascular point false positive FP of segmentation errors, the puncta vasculosa true negative TN of erroneous segmentation, the non-vascular point false negative correctly divided
FN, Acc indicate that the correct pixel of segmentation accounts for the ratio of whole image pixel summation, i.e. accuracy, and Se indicates that segmentation is correct
Puncta vasculosa account for the ratio of goldstandard puncta vasculosa summation, i.e. sensitivity, Sp indicates that dividing correct background dot accounts for goldstandard background
The ratio of point summation, i.e., specific, Ppv indicates to divide the ratio that correct puncta vasculosa accounts for the puncta vasculosa being partitioned into, i.e. precision is pre-
Survey ratio.
The present invention also provides a kind of eye ground blood vessels based on K-Means cluster mark and model-naive Bayesian
Segmenting system, comprising:
Sample sampling unit constructs training set and test for randomly selecting colored eye fundus image from DRIVE data set
Collection;
Pretreatment unit, for extract the grayscale image in the channel G in colored eye fundus image and to retina sense therein it is emerging
Interesting region carries out expansion process;
Feature extraction unit carries out feature extraction for the grayscale image to the channel G after expansion process, by training set and survey
Each pixel in examination collection image is indicated with the feature vector of ten dimensions comprising:
Module, Gauss matching characteristic extraction module, Frangi is filtered in Gabor wavelet processing module, linear detector
Filtering characteristics extraction module, B-COSFIRE filtering characteristics extraction module, weber Expressive Features extraction module;
Cluster mark unit: for feature vector to be carried out cluster mark by K-Means clustering algorithm;
Machine learning model training unit: for the simple pattra leaves of training set data training by K-Means cluster mark
This model;
Image processing unit: for being split by blood vessel of the parted pattern to each image in test set;
Segmentation effect judges unit: judging for the validity to blood vessel segmentation method.
Compared with the prior art, the invention has the following advantages that
The present invention, using the result of cluster as the label of Training, then utilizes these label instructions by clustering technique
To practice Naive Bayes Classification Model and carries out retinal vessel segmentation, whole process does not need artificially to participate in label, and it is time saving and energy saving, greatly
Amplitude improves the treatment effeciency of machine learning model, and relative to the segmentation result of cluster, the present invention is divided in retinal images
Accuracy rate is improved in treatment process.
The present invention can be widely popularized in field of medical image processing based on the above reasons.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to do simply to introduce, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without any creative labor, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the method for the present invention flow chart.
Fig. 2 is apparatus of the present invention module map.
Fig. 3 is the schematic diagram of original eye ground image and Mask image in the present invention.
Fig. 4 is the Analysis of Contrast schematic diagram of RGB triple channel grayscale image in the present invention, wherein (a) is red channel,
(b) it is green channel, (c) is blue channel.
Fig. 5 is retina area-of-interest expansion process schematic diagram in green channel grayscale image of the present invention.
Fig. 6 is that the present invention successively obtains obtaining characteristic pattern schematic diagram after each feature extraction, wherein (a) is original color
Image (b) is gray feature, is (c) Gabor transformation, be (d) linearity test (e) is Gauss matched filtering, (f) is Frangi
Filtering (g) filters for B-COSFIRE, (h) describes for weber.
Fig. 7 is the cluster result schematic diagram that the present invention uses K-Means clustering algorithm in training set image.
Fig. 8 is the segmentation result schematic diagram that the trained parted pattern of the present invention divides test set.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
The model that the present invention protects all should belong in member's every other embodiment obtained without making creative work
It encloses.
As shown in Figure 1, the present invention provides a kind of eyeground based on K-Means cluster mark and model-naive Bayesian to regard
Retinal vasculature dividing method, includes the following steps:
S1, colored eye fundus image in data set is randomly selected, the present embodiment uses disclosed DRIVE data set,
DRIVE is the colored eyeground figure that Niemeijer team established in 2004 according to the work of Dutch screening for diabetic retinopathy
Library, image are to shoot to obtain from 453 25~90 years old Different Individuals.The present embodiment extracts totally 40 width image in database,
Wherein 7 width have early diabetic retinopathy, and 33 width are the eyeground figure of not diabetic retinopathy, each image
Pixel be 584 × 565.
It is divided into training set and test set, each 20 width image of subset, the result of the corresponding 2 expert's manual segmentations of each image
The mask image in the effective information region of eye fundus image is shown with 1.As shown in figure 3, being partial colour retinal fundus images
With corresponding Mask image.The library is the most frequently used database for measuring Segmentation Method of Retinal Blood Vessels performance quality.
The grayscale image contrast of the RGB triple channel of S2, the colored eye fundus image of analysis, the grayscale image medium vessels and back in the channel G
The contrast highest of scape, as shown in figure 4, object of the grayscale image in the channel G of the colored eye fundus image of extraction as feature extraction, is
The influence for reducing region contour edge noise, carries out expansion process as shown in Figure 5 to retina area-of-interest.
S3, the contrast for enhancing the grayscale image by the adaptive histogram equalization method of contrast-limited, by this
Gray value is enhanced described in technical treatment as a kind of feature for distinguishing blood vessel and background by the blood vessel designed based on vessel properties
Grayscale image highlights the blood vessel structure in eye fundus image, while inhibiting insignificant part, reinforces image interpretation and identification.Point
For based on airspace and based on the big algorithm of frequency domain two, the former directly calculates image grayscale, the latter is based on image transform domain to change
Coefficient is changed to be modified.
As shown in fig. 6, the grayscale image is filtered as follows respectively:
Gabor wavelet transformation, the blood vessel curvature in retinal images is small, width gradual change and direction is arbitrary, and
Gabor wavelet transformation is sensitive to image border, the characteristic with good scale selection and direction selection, so using Gabor
Image is transformed into Frequency domain from time domain by wavelet transformation.
Gabor kernel function is the Gauss function modulated by multiple SIN function:
Wherein, x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ,
(u0,v0) indicate the temporal frequency of sinusoidal plane wave, σxAnd σyIndicate the variance of the oval Gauss on the direction x and y,
Which determine the sizes of the filter zone of action, it is seen that parallel and inhibition zone striped quantity is more, commonly referred to as scale, and θ is rotation
Gyration.
Change the available one group of direction the major parameter σ and θ Gabor wavelet different with scale.Under fixed size, than
Response on more all directions, using peak response as the characteristic image under the scale.In an experiment setting scale be [2,3,4,
5], the Gabor characteristic of 4 dimensions can be obtained in this way.
Linear detector is the filter of an engineer, be mainly based upon blood vessel piecewise linearity and connectivity this
What priori knowledge was suggested, by fixed size mode development to multi-scalability.Under the window of W*W, length is the detection of l
Line with 15 ° in 0:165 rotation obtain the filters of 12 different directions original image be filtered, when detection line direction and
When vessel directions are consistent, response is maximum, compare institute it is angled under response reservation maximum value as the testing result under the windowIn the length for changing detection line to adapt to the blood vessel under different scale, finally the response results under different scale are summed
The response results of linearity test are averagely obtained again as linearity test feature.
Gauss matched filtering feature.Retinal vessel cross section intensity profile is in the Gaussian Profile stood upside down, it is possible to benefit
It is used to detect blood vessel come ' matching ' with the filter of gaussian shape.The Gaussian kernel function of handstand is defined as:
WhereinBy the mean normalization of filter to 0.L is smooth along y-axis
The neighborhood length of noise, theoretically L is bigger, and retinal images smooth effect can be made more excellent, while can also reduce retina
Picture noise.T is constant, is usually arranged as 3, because 99% area is located in [- 3 σ, 3 σ] range under Gaussian curve.When setting
X'=xcos θ+ysin θ, y'=-xsin θ+ycos θ, θ=(0:12) * π/12 variation are set to detect the blood vessel of different directions.?
The peak response on 12 directions is chosen under a certain scale as the filter result under the scale.In experiment, L=9, ruler are set
Degree variation σ=1:0.3:2, compares the response under different scale, chooses and responds maximum filter result as Gauss matching characteristic.
Frangi filtering characteristics.The vasculature part on eyeground is the structure of a tubulose, and the response that Gauss second order is led compares
Greatly;The background parts on eyeground are evenly distributed, and the response that Gauss second order is led is smaller.And Hessian matrix is a Second Order Partial
Matrix is led, and for two dimensional image, the spy at each pixel there are two Hessian matrix exgenvalues, at puncta vasculosa
Value indicative small one and large one, characteristic value at intersecting blood vessels point is both very big, the both very little of the characteristic value at background dot.
For two-dimensional input image, obtain constructing each pixel (x, y) with the second order local derviation convolution of gaussian kernel function
Hessian matrix,
The result of Frangi filtering is calculated by such as minor function:
Wherein, the characteristic value of Hessian matrix is λ1And λ2, and | λ1| < | λ2|, Rb=λ2/λ1,
β controls the parameter of whole smoothness for adjusting linear and block-like difference, c statement, and the present embodiment chooses β and is
The range scale of 0.5, c 5, Gauss function is σ ∈ [1,3], and response of more each pixel under each scale is chosen
Peak response is as Frangi filtering characteristics.
B-COSFIRE filtering characteristics.
The Difference of Gaussian filter that B-COSFIRE filter is collinearly arranged by one group, i.e. DoG filter are constituted, B-
COSFIRE filter response is obtained by calculating the geometric mean of all DoG filter responses.DoG filter expression is as follows:
The DoG filter response of image obtains c by the convolution of DoG filter and imageσ(x, y)=| I*DoG |.Parameter
CollectionDescribe the parameter of n DoG filter, σiIt is the standard deviation of each DoG filter.(ρi,φi) be
Polar coordinate position indicates position of the center with respect to COSFIRE filter center of each DoG filter.
For extracting continuous blood vessel structure feature, DoG filter exists the COSFIRE model of symmetrical structure in pairs, they
Center be located at the two sides of B-COSFIRE filter center in the opposite direction;
For the COSFIRE model of unsymmetric structure for detecting tail vein, the center of DoG filter is located at B-COSFIRE
The side of filter center,
In order to improve the fault-tolerance of each position, fuzzy operation is carried out to DoG filter response, the fuzzy operation is
The maximum value of DoG filter weight threshold response, weight are the response of DoG filter and Gaussian function Gσ' (x', y') coefficient multiplies
Product, wherein standard deviation sigma ' be distance ρiLinear function:
σ '=σ0+αρi
σ0And ρiIt is all constant.Then to φiShift operation is carried out after the DoG filter blurs operation of opposite direction, is moved
Bit vector is (Δ xi,Δyi), Δ xi=-ρicosφi,Δyi=-ρisinφi。
Each tuple (σ in final parameter collection Si,ρi,φi) corresponding DoG filter fuzzy displacement response are as follows:
So the response of B-COSFIRE filter is that the corresponding all fuzzy weightings for shifting DoG filter responses of parameter set S are several
What mean value, calculation formula are as follows:
Wherein | |tIt indicates to maximum response threshold valueization processing,
When changing φ, the arrangement mode of DoG filter, which can change, is equivalent to rotation B-COSFIRE filter, calculates figure
Each of picture filter result of the pixel on 12 direction ψ, is maximized the B-COSFIRE feature as the pixel,
Correlation formula is as follows:
Finally symmetrical B-COSFIRE filter and asymmetric B-COSFIRE filter response results are made and obtained final
B-COSFIRE filtering characteristics.
Parameter setting about filter is as follows:
Weber description operator is as a kind of strong local texture description method, for any pixel point p in image I,
Differential excitation is gray scale difference value of the pixel relative to its neighborhood territory pixel, then asks the summation of difference and the ratio of central pixel point
Then value solves arc tangent and obtains weber Expressive Features of the gradient as pixel p of differential excition,
Wherein, p is any pixel point in image I, and N (p) indicates the neighborhood territory pixel set of pixel p, and I (z) indicates target picture
The gray value of any pixel of N (p) in the neighborhood of plain p.
S4,10 dimensions of each pixel as shown in fig. 7, after feature extraction, in training set and test set image
Feature vector indicates.For each image in training set, all feature vectors are polymerized to 2 classes-using K-Means clustering algorithm
Blood vessel cluster (being labeled as 1) and background cluster (being labeled as 0), such training set image is just provided with Pixel-level label.
S5, the training set data training model-naive Bayesian based on K-Means cluster mark, for there is a small amount of mistake mark
The data of label, model-naive Bayesian can approximation obtain the corresponding prior probability of true tag data and conditional probability distribution.
Based on 2016RaMATLAB platform, K-Means is clustered and model-naive Bayesian uses the statistics and machine of MATLAB
Function idx=kmeans (X, k) and Mdl=fitcnb (X, Y), X in device learning tool case are training datasets, and k is cluster
Number of clusters, idx are the labels of the X obtained by clustering algorithm, and Y is corresponding cluster labels, and Mdl is the simple shellfish that training obtains
This model of leaf.
S6, the blood vessel for dividing each image in test set with trained model-naive Bayesian.As shown in Fig. 8,
In, green indicates the puncta vasculosa correctly divided, and red indicates the non-vascular point of segmentation errors, and blue is the blood vessel of erroneous segmentation
Point;Black is the non-vascular point correctly divided.
S7, in order to judge whether blood vessel segmentation method effective, need an effective Performance Evaluating Indexes.By blood vessel segmentation
As a result compared with the goldstandard that expert demarcates manually, the segmentation result of pixel has 4 kinds of situations, true positives (true
Positive, TP) it indicates to divide correct puncta vasculosa;The blood vessel of false positive (false positive, FP) expression segmentation errors
Point;True negative (true negative, TN) indicates to divide correct background dot;False negative (false negative, FN) indicates
The background dot of segmentation errors.
Acc indicates to divide the ratio that correct pixel accounts for whole image pixel summation, i.e. accuracy, Se expression segmentation
Correct puncta vasculosa accounts for the ratio of goldstandard puncta vasculosa summation, i.e. sensitivity, and Sp indicates that dividing correct background dot accounts for goldstandard
The ratio of background dot summation, i.e., specific, Ppv indicates to divide the ratio that correct puncta vasculosa accounts for the puncta vasculosa being partitioned into, i.e., smart
Spend predicted ratio.
Following table gives the performance comparison using K-Mean clustering algorithm and context of methods, and the accuracy rate of two methods is about
96%, and in retinal images medium vessels ratio very little, so performance of the Se and PPv value more representative of a method.The side this paper
The Se value of method shows that more puncta vasculosas can be partitioned into more greatly, and 80% is all correct puncta vasculosa, so context of methods
Effect is promoted relative to K-Means cluster really.
Fig. 8
As shown in Fig. 2, the present invention also provides a kind of eyeground based on K-Means cluster mark and model-naive Bayesian
Retinal vessel segmenting system, comprising:
Sample sampling unit constructs training set and test for randomly selecting colored eye fundus image from DRIVE data set
Collection;
Pretreatment unit, for extract the grayscale image in the channel G in colored eye fundus image and to retina sense therein it is emerging
Interesting region carries out expansion process;
Feature extraction unit carries out feature extraction for the grayscale image to the channel G after expansion process, by training set and survey
Each pixel in examination collection image is indicated with the feature vector of ten dimensions comprising:
Module, Gauss matching characteristic extraction module, Frangi is filtered in Gabor wavelet processing module, linear detector
Filtering characteristics extraction module, B-COSFIRE filtering characteristics extraction module, weber Expressive Features extraction module;
Cluster mark unit: for feature vector to be carried out cluster mark by K-Means clustering algorithm;
Machine learning model training unit: for the simple pattra leaves of training set data training by K-Means cluster mark
This model;
Image processing unit: for being split by blood vessel of the parted pattern to each image in test set;
Segmentation effect judges unit: judging for the validity to blood vessel segmentation method.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement;And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (10)
1. a kind of eye ground blood vessel segmentation method based on K-Means cluster mark and model-naive Bayesian, feature
It is, includes the following steps:
The Pixel-level label of training set image is constructed, specifically: K- is used to each image in the training set after feature extraction
Feature vector is carried out cluster mark by Means clustering algorithm;
Training set data training model-naive Bayesian based on K-Means cluster mark;
With the blood vessel of each image in trained model-naive Bayesian segmentation test set.
2. the method according to claim 1, wherein before the Pixel-level label of the building training set image also
With following steps:
The colored eye fundus image in data set is randomly selected, training set and test set are constructed;
Extract object of the grayscale image in the channel G of colored eye fundus image as feature extraction, and to retina area-of-interest into
Row expansion process;
Feature extraction is carried out to the grayscale image after expansion process, by each pixel in training set and test set image with ten
The feature vector expression of dimension, specifically:
The contrast for enhancing the grayscale image by the adaptive histogram equalization method of contrast-limited makees this gray value
For a kind of feature for distinguishing blood vessel and background, grayscale image described in technical treatment is enhanced by the blood vessel designed based on vessel properties,
The grayscale image is filtered as follows respectively:
By Gabor wavelet conversion process, four-dimensional Gabor characteristic is obtained;
Retinal images are filtered by linear detector, obtain the response results of linearity test as one-dimensional line
Property detection feature;
Retinal images are filtered by the filter of gaussian shape, chooses and responds maximum filter result as one
The Gauss matching characteristic of dimension;
Frangi filtering characteristics based on Hessian matrix are filtered retinal images, choose peak response conduct
One-dimensional Frangi filtering characteristics;
Retinal images are filtered based on B-COSFIRE filtering characteristics, by symmetrical B-COSFIRE filter and non-right
B-COSFIRE filter response results are claimed to sum to obtain final one-dimensional B-COSFIRE filtering characteristics;
The one-dimensional weber Expressive Features that operator solves any pixel point in retinal images are described by weber;
The blood vessel with each image in trained model-naive Bayesian segmentation test set also has following steps later:
Judge whether blood vessel segmentation method is effective by preset evaluation index.
3. according to the method described in claim 2, it is characterized in that, the colored eye fundus image randomly selected in data set,
It constructs in training set and test set, data set uses disclosed DRIVE data set, the picture number of the training set and test set
It is identical, and the Mask in the effective information region of the result and 1 display eye fundus image of the corresponding 2 expert's manual segmentations of each image
Image.
4. according to the method described in claim 2, it is characterized in that, carrying out Gabor wavelet conversion process in the following way:
Gabor kernel function is as follows:
Wherein, x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ,
(u0,v0) indicate the temporal frequency of sinusoidal plane wave, σxAnd σyIt indicates the variance of the oval Gauss on the direction x and y, determines
The size of the filter zone of action, i.e. scale, θ are rotation angle,
One group of direction Gabor wavelet different with scale, under fixed size, more all sides are obtained by changing parameter σ and θ
Upward response, using peak response as the characteristic image under the scale, according to the scale of setting, the result under all scales
As Gabor characteristic.
5. according to the method described in claim 4, it is characterized in that, the linear detector is by rotating under the window of W*W
Length is that the detection line of l obtains the filter of N number of different directions, grayscale image is filtered with filter, when the side of detection line
To it is consistent with vessel directions when, response is maximum, compare institute it is angled under response reservation maximum value as the detection under the window
As a resultLength by changing detection line adapts to the blood vessel under different scale, again by the response results summation under different scale
The response results of linearity test are averagely obtained as linearity test feature;
The filter graph picture filtering processing of gaussian shape is carried out in the following way:
The Gaussian kernel function of handstand is as follows:
WhereinThe mean normalization of filter to 0, L is indicated smoothly to make an uproar along y-axis
The neighborhood length of sound, t are constants, setting x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ, θ=(0:N) * π/N variation with
The blood vessel for detecting different directions, chooses peak response on N number of direction as the filter result under the scale under a certain scale,
The response under different scale is compared, chooses and responds maximum filter result as Gauss matching characteristic.
6. according to the method described in claim 5, it is characterized in that, being carried out in the following way based on Hessian matrix
Frangi filtering processing:
For two-dimensional input image, obtain constructing each pixel (x, y) with the second order local derviation convolution of gaussian kernel function
Hessian matrix,
The result of Frangi filtering is calculated by such as minor function:
Wherein, the characteristic value of Hessian matrix is λ1And λ2, and | λ1| < | λ2|,
β controls the parameter of whole smoothness for adjusting linear and block-like difference, c statement, and more each pixel is each
Response under scale chooses peak response as Frangi filtering characteristics.
7. according to the method described in claim 6, it is characterized in that, the Gauss that B-COSFIRE filter is collinearly arranged by one group
Difference filter, i.e. DoG filter are constituted, and B-COSFIRE filter response is equal by the geometry for calculating all DoG filter responses
Value obtains, and the DoG filter response is obtained by the convolution of DoG filter and image,
For extracting continuous blood vessel structure feature, DoG filter exists the COSFIRE model of symmetrical structure in pairs, in them
The heart is located at the two sides of B-COSFIRE filter center in the opposite direction;The COSFIRE model of unsymmetric structure is for detecting end
Blood vessel, the center of DoG filter are located at the side of B-COSFIRE filter center, carry out fuzzy operation to DoG filter response,
The fuzzy operation is the maximum value of DoG filter weight threshold response, and weight is DoG filter response and Gaussian function coefficient
Final B- is made and obtained to symmetrical B-COSFIRE filter and asymmetric B-COSFIRE filter response result by product
COSFIRE filtering characteristics.
8. the method according to the description of claim 7 is characterized in that passing through any pixel in following equations retinal images
The weber Expressive Features of point:
Wherein, p is any pixel point in image I, and N (p) indicates the neighborhood territory pixel set of pixel p, and I (z) indicates object pixel p's
The gray value of any pixel of N (p) in neighborhood.
9. according to the method described in claim 2, it is characterized in that, described judge blood vessel segmentation side by preset evaluation index
Whether method is effective, specifically:
Wherein, the puncta vasculosa of each image passes through four kinds of different colors differentiations, the puncta vasculosa true positives TP correctly divided, segmentation
The non-vascular point false positive FP of mistake, the puncta vasculosa true negative TN of erroneous segmentation, the non-vascular point false negative FN correctly divided,
Acc indicates to divide the ratio that correct pixel accounts for whole image pixel summation, i.e. accuracy, the correct blood of Se expression segmentation
Pipe point accounts for the ratio of goldstandard puncta vasculosa summation, i.e. sensitivity, and it is total that Sp indicates that the correct background dot of segmentation accounts for goldstandard background dot
The ratio of sum, i.e., specific, Ppv indicates to divide the ratio that correct puncta vasculosa accounts for the puncta vasculosa being partitioned into, i.e. accuracy prediction ratio
Value.
10. a kind of eye ground blood vessel segmentation system based on K-Means cluster mark and model-naive Bayesian, feature
It is, comprising:
Sample sampling unit constructs training set and test set for randomly selecting colored eye fundus image from DRIVE data set;
Pretreatment unit, for extracting the grayscale image in the channel G in colored eye fundus image and to retina region of interest therein
Domain carries out expansion process;
Feature extraction unit carries out feature extraction for the grayscale image to the channel G after expansion process, by training set and test set
Each pixel in image is indicated with the feature vector of ten dimensions comprising:
Module, Gauss matching characteristic extraction module, Frangi filtering is filtered in Gabor wavelet processing module, linear detector
Characteristic extracting module, B-COSFIRE filtering characteristics extraction module, weber Expressive Features extraction module;
Cluster mark unit: for feature vector to be carried out cluster mark by K-Means clustering algorithm;
Machine learning model training unit: for the training set data training naive Bayesian mould by K-Means cluster mark
Type;
Image processing unit: for being split by blood vessel of the parted pattern to each image in test set;
Segmentation effect judges unit: judging for the validity to blood vessel segmentation method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810733535.8A CN109003279B (en) | 2018-07-06 | 2018-07-06 | Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810733535.8A CN109003279B (en) | 2018-07-06 | 2018-07-06 | Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109003279A true CN109003279A (en) | 2018-12-14 |
CN109003279B CN109003279B (en) | 2022-05-13 |
Family
ID=64599062
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810733535.8A Expired - Fee Related CN109003279B (en) | 2018-07-06 | 2018-07-06 | Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109003279B (en) |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109829931A (en) * | 2019-01-07 | 2019-05-31 | 三峡大学 | A kind of Segmentation Method of Retinal Blood Vessels increasing PCNN based on region |
CN109859146A (en) * | 2019-02-28 | 2019-06-07 | 电子科技大学 | A kind of colored eye fundus image blood vessel segmentation method based on U-net convolutional neural networks |
CN109978819A (en) * | 2019-01-22 | 2019-07-05 | 安徽海浪智能技术有限公司 | A method of segmentation retinal vessel is detected based on low scale blood vessel |
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 |
CN110135481A (en) * | 2019-04-30 | 2019-08-16 | 佛山科学技术学院 | A kind of crops lesion detection method and detection device |
CN110189309A (en) * | 2019-05-21 | 2019-08-30 | 上海商汤智能科技有限公司 | A kind of image processing method and device, electronic equipment and storage medium |
CN110310235A (en) * | 2019-05-21 | 2019-10-08 | 北京至真互联网技术有限公司 | Method for processing fundus images, device and equipment and storage medium |
CN110490040A (en) * | 2019-05-30 | 2019-11-22 | 浙江理工大学 | A method of local vascular stenosis in identification DSA coronary artery images |
CN110648312A (en) * | 2019-09-03 | 2020-01-03 | 上海工程技术大学 | Method for identifying wool and cashmere fibers based on scale morphological characteristic analysis |
CN111563890A (en) * | 2020-05-07 | 2020-08-21 | 浙江大学 | Fundus image blood vessel segmentation method and system based on deep forest |
CN112396565A (en) * | 2020-11-19 | 2021-02-23 | 同济大学 | Method and system for enhancing and segmenting blood vessels of images and videos of venipuncture robot |
CN112464010A (en) * | 2020-12-17 | 2021-03-09 | 中国矿业大学(北京) | Automatic image labeling method based on Bayesian network and classifier chain |
CN112529879A (en) * | 2020-12-16 | 2021-03-19 | 福建农林大学 | Multi-scale sampling retinal vessel image segmentation method based on B-COSFIRE filtering processing of support vector machine |
CN112862789A (en) * | 2021-02-10 | 2021-05-28 | 上海大学 | Interactive image segmentation method based on machine learning |
CN113409293A (en) * | 2021-06-30 | 2021-09-17 | 上海市第一人民医院 | Pathology image automatic segmentation system based on deep learning |
CN113450351A (en) * | 2021-08-06 | 2021-09-28 | 推想医疗科技股份有限公司 | Segmentation model training method, image segmentation method, device, equipment and medium |
CN113658207A (en) * | 2021-08-20 | 2021-11-16 | 华南师范大学 | Retinal vessel segmentation method and device based on guide filtering |
CN113762400A (en) * | 2021-09-13 | 2021-12-07 | 南昌大学 | Weld joint position autonomous extraction method based on naive Bayes classifier |
CN113989246A (en) * | 2021-10-29 | 2022-01-28 | 南开大学 | Transparent blood vessel image segmentation method based on blood flow characteristics |
CN113989170A (en) * | 2021-10-29 | 2022-01-28 | 南开大学 | Transparent blood vessel type identification method based on blood flow characteristics |
CN114627046A (en) * | 2021-12-30 | 2022-06-14 | 山东爱普电气设备有限公司 | Power line detection method and system based on visible light multi-filtering dimensionality |
CN114818839A (en) * | 2022-07-01 | 2022-07-29 | 之江实验室 | Deep learning-based optical fiber sensing underwater acoustic signal identification method and device |
CN114913986A (en) * | 2022-07-15 | 2022-08-16 | 四川大学华西医院 | Medical object clustering method and device, electronic equipment and storage medium |
CN116681715A (en) * | 2023-08-04 | 2023-09-01 | 杭州脉流科技有限公司 | Blood vessel segmentation method, device, equipment and storage medium based on pixel value change |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060034536A1 (en) * | 2004-06-23 | 2006-02-16 | Ogren Wayne E | Systems and methods relating to magnitude enhancement analysis suitable for high bit level displays on low bit level systems, determining the material thickness, and 3D visualization of color space dimensions |
US20110026789A1 (en) * | 2008-04-08 | 2011-02-03 | National University Of Singapore | Retinal image analysis systems and methods |
CN102346851A (en) * | 2011-11-04 | 2012-02-08 | 西安电子科技大学 | Image segmentation method based on NJW (Ng-Jordan-Weiss) spectral clustering mark |
CN103745473A (en) * | 2014-01-16 | 2014-04-23 | 南方医科大学 | Brain tissue extraction method |
CN104463215A (en) * | 2014-12-10 | 2015-03-25 | 东北大学 | Tiny aneurysm occurrence risk prediction system based on retina image processing |
CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
-
2018
- 2018-07-06 CN CN201810733535.8A patent/CN109003279B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060034536A1 (en) * | 2004-06-23 | 2006-02-16 | Ogren Wayne E | Systems and methods relating to magnitude enhancement analysis suitable for high bit level displays on low bit level systems, determining the material thickness, and 3D visualization of color space dimensions |
US20110026789A1 (en) * | 2008-04-08 | 2011-02-03 | National University Of Singapore | Retinal image analysis systems and methods |
CN102346851A (en) * | 2011-11-04 | 2012-02-08 | 西安电子科技大学 | Image segmentation method based on NJW (Ng-Jordan-Weiss) spectral clustering mark |
CN103745473A (en) * | 2014-01-16 | 2014-04-23 | 南方医科大学 | Brain tissue extraction method |
CN104463215A (en) * | 2014-12-10 | 2015-03-25 | 东北大学 | Tiny aneurysm occurrence risk prediction system based on retina image processing |
CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
Non-Patent Citations (4)
Title |
---|
R GEETHARAMANI 等: "Automatic segmentation of blood vessels from retinal fundus images through image processing and data mining techniques", 《SA¯DHANA¯ 》 * |
吴奎 等: "基于 2D Gabor 小波与组合线检测算子的视网膜血管分割", 《科学技术与工程》 * |
李媛媛 等: "基于融合相位特征的视网膜血管分割算法", 《计算机应用》 * |
蒋进: "基于图像识别的纸张计数***研究", 《中国优秀博硕士学位论文全文数据库(硕士) 工程科技Ⅰ辑》 * |
Cited By (36)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109829931A (en) * | 2019-01-07 | 2019-05-31 | 三峡大学 | A kind of Segmentation Method of Retinal Blood Vessels increasing PCNN based on region |
CN109978819A (en) * | 2019-01-22 | 2019-07-05 | 安徽海浪智能技术有限公司 | A method of segmentation retinal vessel is detected based on low scale blood vessel |
CN109859146A (en) * | 2019-02-28 | 2019-06-07 | 电子科技大学 | A kind of colored eye fundus image blood vessel segmentation method based on U-net convolutional neural networks |
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 |
CN110135481A (en) * | 2019-04-30 | 2019-08-16 | 佛山科学技术学院 | A kind of crops lesion detection method and detection device |
CN110189309B (en) * | 2019-05-21 | 2021-06-15 | 上海商汤智能科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN110189309A (en) * | 2019-05-21 | 2019-08-30 | 上海商汤智能科技有限公司 | A kind of image processing method and device, electronic equipment and storage medium |
CN110310235A (en) * | 2019-05-21 | 2019-10-08 | 北京至真互联网技术有限公司 | Method for processing fundus images, device and equipment and storage medium |
CN110490040A (en) * | 2019-05-30 | 2019-11-22 | 浙江理工大学 | A method of local vascular stenosis in identification DSA coronary artery images |
CN110490040B (en) * | 2019-05-30 | 2022-04-12 | 浙江理工大学 | Method for identifying local vascular stenosis degree in DSA coronary artery image |
CN110648312A (en) * | 2019-09-03 | 2020-01-03 | 上海工程技术大学 | Method for identifying wool and cashmere fibers based on scale morphological characteristic analysis |
CN111563890A (en) * | 2020-05-07 | 2020-08-21 | 浙江大学 | Fundus image blood vessel segmentation method and system based on deep forest |
CN112396565A (en) * | 2020-11-19 | 2021-02-23 | 同济大学 | Method and system for enhancing and segmenting blood vessels of images and videos of venipuncture robot |
CN112529879A (en) * | 2020-12-16 | 2021-03-19 | 福建农林大学 | Multi-scale sampling retinal vessel image segmentation method based on B-COSFIRE filtering processing of support vector machine |
CN112464010B (en) * | 2020-12-17 | 2021-08-27 | 中国矿业大学(北京) | Automatic image labeling method based on Bayesian network and classifier chain |
CN112464010A (en) * | 2020-12-17 | 2021-03-09 | 中国矿业大学(北京) | Automatic image labeling method based on Bayesian network and classifier chain |
CN112862789A (en) * | 2021-02-10 | 2021-05-28 | 上海大学 | Interactive image segmentation method based on machine learning |
CN112862789B (en) * | 2021-02-10 | 2023-06-13 | 上海大学 | Interactive image segmentation method based on machine learning |
CN113409293A (en) * | 2021-06-30 | 2021-09-17 | 上海市第一人民医院 | Pathology image automatic segmentation system based on deep learning |
CN113450351A (en) * | 2021-08-06 | 2021-09-28 | 推想医疗科技股份有限公司 | Segmentation model training method, image segmentation method, device, equipment and medium |
CN113450351B (en) * | 2021-08-06 | 2024-01-30 | 推想医疗科技股份有限公司 | Segmentation model training method, image segmentation method, device, equipment and medium |
CN113658207A (en) * | 2021-08-20 | 2021-11-16 | 华南师范大学 | Retinal vessel segmentation method and device based on guide filtering |
CN113658207B (en) * | 2021-08-20 | 2023-06-16 | 华南师范大学 | Retina blood vessel segmentation method and device based on guided filtering |
CN113762400A (en) * | 2021-09-13 | 2021-12-07 | 南昌大学 | Weld joint position autonomous extraction method based on naive Bayes classifier |
CN113762400B (en) * | 2021-09-13 | 2023-10-31 | 南昌大学 | Autonomous extraction method for weld joint position based on naive Bayes classifier |
CN113989170B (en) * | 2021-10-29 | 2023-01-24 | 南开大学 | Transparent blood vessel type identification method based on blood flow characteristics |
CN113989170A (en) * | 2021-10-29 | 2022-01-28 | 南开大学 | Transparent blood vessel type identification method based on blood flow characteristics |
CN113989246B (en) * | 2021-10-29 | 2023-01-24 | 南开大学 | Transparent blood vessel image segmentation method based on blood flow characteristics |
CN113989246A (en) * | 2021-10-29 | 2022-01-28 | 南开大学 | Transparent blood vessel image segmentation method based on blood flow characteristics |
CN114627046A (en) * | 2021-12-30 | 2022-06-14 | 山东爱普电气设备有限公司 | Power line detection method and system based on visible light multi-filtering dimensionality |
CN114818839B (en) * | 2022-07-01 | 2022-09-16 | 之江实验室 | Deep learning-based optical fiber sensing underwater acoustic signal identification method and device |
CN114818839A (en) * | 2022-07-01 | 2022-07-29 | 之江实验室 | Deep learning-based optical fiber sensing underwater acoustic signal identification method and device |
CN114913986B (en) * | 2022-07-15 | 2022-10-14 | 四川大学华西医院 | Medical object clustering method and device, electronic equipment and storage medium |
CN114913986A (en) * | 2022-07-15 | 2022-08-16 | 四川大学华西医院 | Medical object clustering method and device, electronic equipment and storage medium |
CN116681715A (en) * | 2023-08-04 | 2023-09-01 | 杭州脉流科技有限公司 | Blood vessel segmentation method, device, equipment and storage medium based on pixel value change |
CN116681715B (en) * | 2023-08-04 | 2023-10-10 | 杭州脉流科技有限公司 | Blood vessel segmentation method, device, equipment and storage medium based on pixel value change |
Also Published As
Publication number | Publication date |
---|---|
CN109003279B (en) | 2022-05-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109003279A (en) | Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model | |
Lupascu et al. | FABC: retinal vessel segmentation using AdaBoost | |
Amin et al. | A method for the detection and classification of diabetic retinopathy using structural predictors of bright lesions | |
Aslani et al. | A new supervised retinal vessel segmentation method based on robust hybrid features | |
Yin et al. | Accurate image analysis of the retina using hessian matrix and binarisation of thresholded entropy with application of texture mapping | |
Marín et al. | A new supervised method for blood vessel segmentation in retinal images by using gray-level and moment invariants-based features | |
Akram et al. | Automated detection of exudates and macula for grading of diabetic macular edema | |
Zhang et al. | Retinal vessel segmentation using multi-scale textons derived from keypoints | |
Kovács et al. | A self-calibrating approach for the segmentation of retinal vessels by template matching and contour reconstruction | |
Panda et al. | New binary Hausdorff symmetry measure based seeded region growing for retinal vessel segmentation | |
Lupaşcu et al. | Automatic unsupervised segmentation of retinal vessels using self-organizing maps and k-means clustering | |
Jordan et al. | A review of feature-based retinal image analysis | |
Pachade et al. | Retinal vasculature segmentation and measurement framework for color fundus and SLO images | |
Rodrigues et al. | Retinal vessel segmentation using parallel grayscale skeletonization algorithm and mathematical morphology | |
Jamil et al. | Melanocytic and nevus lesion detection from diseased dermoscopic images using fuzzy and wavelet techniques | |
Deng et al. | Age-related macular degeneration detection and stage classification using choroidal OCT images | |
Jadhav et al. | Classification of diabetes retina images using blood vessel area | |
Reddy et al. | Diabetic retinopathy through retinal image analysis: A review | |
Tamilarasi et al. | Automatic detection of microaneurysms using microstructure and wavelet methods | |
Bansal et al. | Retina vessels detection algorithm for biomedical symptoms diagnosis | |
Balakrishnan | NDC-IVM: An automatic segmentation of optic disc and cup region from medical images for glaucoma detection | |
Soares et al. | Microaneurysms detection using a novel neighborhood analysis | |
Peng et al. | Blood vessels segmentation by using cdnet | |
Barhoumi et al. | Pigment network detection in dermatoscopic images for melanoma diagnosis | |
CN116309633A (en) | Retina blood vessel segmentation method based on nuclear intuitionistic fuzzy C-means clustering |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220513 |