CN109671094A - A kind of eye fundus image blood vessel segmentation method based on frequency domain classification - Google Patents
A kind of eye fundus image blood vessel segmentation method based on frequency domain classification Download PDFInfo
- Publication number
- CN109671094A CN109671094A CN201811331874.XA CN201811331874A CN109671094A CN 109671094 A CN109671094 A CN 109671094A CN 201811331874 A CN201811331874 A CN 201811331874A CN 109671094 A CN109671094 A CN 109671094A
- Authority
- CN
- China
- Prior art keywords
- block
- fundus image
- eye fundus
- low
- sampling
- 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
- 210000004204 blood vessel Anatomy 0.000 title claims abstract description 31
- 230000011218 segmentation Effects 0.000 title claims abstract description 30
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000005070 sampling Methods 0.000 claims abstract description 43
- 239000000284 extract Substances 0.000 claims abstract description 20
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 238000009826 distribution Methods 0.000 claims abstract description 5
- 238000002156 mixing Methods 0.000 claims abstract description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 30
- 230000037361 pathway Effects 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 7
- 238000003384 imaging method Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 5
- 230000002792 vascular Effects 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 9
- 210000001210 retinal vessel Anatomy 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 206010020718 hyperplasia Diseases 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 230000002207 retinal effect Effects 0.000 description 1
- 230000004256 retinal image Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000005728 strengthening Methods 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
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
- G06T5/92—Dynamic range modification of images or parts thereof based on global image properties
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Eye Examination Apparatus (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of eye fundus image blood vessel segmentation methods based on frequency domain classification.For the inaccurate problem of blood vessel segmentation of the original eye fundus image full range stage treatment method of tradition, the low high-frequency information that eye fundus image is obtained using frequency domain pretreatment is proposed, the low-dimensional and high dimensional feature for then pointedly constructing multipath extract convolutional network.Wherein low-dimensional feature extraction convolutional network includes two symmetric paths in left and right, the main extraction and accurate positioning realized to eye fundus image blood vessel overall situation profile information.It includes two symmetric paths in left and right that high dimensional feature, which extracts convolutional network not only, and on the right during the up-sampling of path, the vessel borders information lost by merging port number realization mixing operation with the characteristic pattern of left side symmetric path with completion, will the further minutia for sharpening the distribution of eye fundus image vessel boundary.Height dimensional feature figure finally is merged using convolution kernel, to obtain more accurate eye fundus image blood vessel segmentation figure.
Description
Technical field
The present invention relates to machine learning and field of medical image processing, and in particular to a kind of eyeground figure based on frequency domain classification
As blood vessel segmentation method.
Background technique
Clinical research shows that the morphosis of retinal vessel when eyeground pathological changes is easy to happen pathology and sexually revises, specific table
It is now the hyperplasia of length of vessel, width, the variation of angle and blood vessel.Carrying out blood vessel segmentation to eye fundus image will be helpful to disease
Screening, diagnosis and analysis, however at present blood vessel segmentation mostly use manual type greatly, not only need clinical experience abundant, also
The a large amount of time and efforts of doctor will be consumed.
The network structure of retinal vessel is tree-shaped type, and branch is more, and minute blood vessel and background contrast in apparatus derivatorius
Very little is spent, profile and border is very fuzzy, and the segmentation that this allows for minute blood vessel is extremely difficult.In recent years, scholars propose perhaps
Multi-method, including retinal images are handled using Gauss matched filtering method combination threshold value；Tracking side based on probability
Method combination image local gray-scale information and blood vessel connection characteristic detect retinal vascular images.Also research and propose
A kind of Segmentation Method of Retinal Blood Vessels combined based on deep learning with conventional method, by full convolutional network end and shallow-layer
The blood vessel probability graph of information is merged, and then obtains desired retinal vessel segmentation figure, but since cutting procedure does not have
There is the relevance fully considered between image airspace and frequency domain, it is intended to which full range is handled by the Depth Expansion of convolutional neural networks
Segment data will not only significantly reduce the computing capability of network, but also the result that will lead to after segmentation is not fine enough, Space Consistency
Hardly result in holding.Therefore above method can extract most of retinal vessel, however eyeground figure lower for contrast
The segmentation task of picture, especially minute blood vessel, is generally unable to reach satisfactory segmentation effect.
Summary of the invention
Above-mentioned to solve the problems, such as, the invention proposes a kind of eye fundus image blood vessel segmentation sides based on frequency domain classification
Method.This method considers significant difference of the retinal vessel in contrast, changes original eye fundus image being directly inputted to volume
The traditional mode of product neural network, but frequency domain classification is carried out to eye fundus image first, the global profile of image is extracted respectively
Information and local detailed information, then pointedly construct independent multipath convolutional neural networks, realize retinal vessel
It extracts and merges, obtain segmentation figure.The present invention carries out the conversion of null tone domain to eye fundus image, extracts eyeground figure respectively using filter
The low frequency and high-frequency information of picture, then low frequency and high-frequency information contravariant are changed into airspace, it is separately input to constructed multipath volume
Feature extraction is carried out in product neural network, last fusion feature figure obtains segmentation figure.Comprising the following steps:
Step 1: the frequency domain classification processing of eye fundus image
Original eye fundus image f (x, y) is transformed into frequency domain F (u, v) by Fourier transformation first, then passes through height respectively
This low-pass filter and Gauss high-pass filter obtain the low frequency and radio-frequency component of eye fundus image, finally utilize Fourier contravariant
It changes and the low frequency and high-frequency information of eye fundus image is handled, obtain the frequency domain classification results of eye fundus image, respectively low frequency is believed
The corresponding f of breath1(x, y), f corresponding to high-frequency information2(x,y)；
Step 2: building low-dimensional feature extraction convolutional network, the global profile for describing eye fundus image using low-frequency component are special
Property, facilitate the separation of background and target blood, promotes the accuracy of prediction vessel borders information；The network is divided into left and right two
Path, whole network include 8 residual blocks, 2 down-samplings, 2 up-samplings and 4 convolutional layers；Wherein, each residual block packet
Containing two 3 × 3 empty convolutional layers, every two residual block and a down-sampling or a up-sampling form 1 block, and totally 4
block；The result of each block output is activated by ReLu function, followed by standardization processing；The left side
The global feature of eye fundus image vascular distribution is extracted by 2 block in path；2, the right block is used to be accurately positioned, wherein
Each block that left pathways pass through is the combination of two residual blocks and a down-sampling, and wherein right pathways each of pass through
Block is the combination of two residual blocks and a up-sampling；The low-frequency image f that step 1 is obtained1(x, y) is input to constructed
Low-dimensional feature extraction convolutional network in, obtain characteristic pattern FL；
Step 3: building high dimensional feature extracts convolutional network, and the details characteristic of eye fundus image, example are described using radio-frequency component
Such as the region that brightness change is violent；The network is equally divided into two paths in left and right, and whole network includes 14 residual blocks, under 4 times
Sampling, 4 up-samplings and 3 convolutional layers；Wherein, each residual block includes two 3 × 3 empty convolutional layers, and every two is residual
Poor block and a down-sampling or up-sampling form a block, totally 6 block, 1 block in left and right path junction
Include two residual blocks, a down-sampling and a up-sampling；For each block output result by ReLu function into
Line activating, followed by standardization processing；Left pathways capture eye fundus image vessel profile information by 3 block；The right
Path carries out mixing operation by 3 block and left side symmetric path, and mixing operation merges characteristic pattern port number, with completion
The further minutia for sharpening eye fundus image vascular distribution is obtained relatively sharp blood vessel by the vessel borders information of loss
Boundary；Each block that wherein left pathways pass through is the combination of two residual blocks and a down-sampling, and wherein right pathways are logical
The each block crossed is the combination of two residual blocks and a up-sampling；The high frequency imaging f that step 1 is obtained2(x, y) input
It is extracted in convolutional network to constructed high dimensional feature, obtains characteristic pattern FH；
Step 4: being added using port number by two groups of characteristic pattern FLAnd FHMerge, obtains high low-dimensional by 1 × 1-32 convolution kernel
Characteristic pattern F ' is then changed into single channel characteristic pattern F " using the convolution kernel of 1 × 1-1 by the characteristic pattern F ' of fusion, using
ReLu function activation after obtain original image f (x, y) corresponding to vessel segmentation output pixel value, and with it is corresponding known to
Blood vessel segmentation label carries out loss operation with the difference of two squares, adds up to the loss operation result of all training images, as a result remembers
For loss, and high and low dimensional feature is respectively trained using gradient descent method and extracts convolutional network；When penalty values loss meets convergence item
After part, terminate training；
Step 5: after high and low dimensional feature extracts convolutional network training, by the eye fundus image of Unknown Label by step 1~
4 are handled, and the single channel characteristic pattern F " corresponding to it is obtained, after the activation of ReLu function, the as blood vessel of eye fundus image
Segmentation result.
The invention has the following advantages that
(1) present invention changes conventional method for the full frequency band tupe of eye fundus image, avoids convolutional Neural net
Influence of the network Depth Expansion to network query function ability.Using the thought of scaling down processing, obtain respectively eye fundus image low frequency and
Then high-frequency information targetedly designs low-dimensional and high dimensional feature extracts convolutional network.
(2) it constructs low-dimensional and high dimensional feature extracts convolutional network, pass through convolution, up-sampling and the down-sampling of multipath respectively
Operation, is extracted the entirety and minutia of eye fundus image, is finally merged using convolution kernel to height dimensional feature figure, obtain essence
True vessel segmentation.
Detailed description of the invention
In order to keep the purpose of the present invention, technical scheme and beneficial effects clearer, the present invention provides following attached drawing and carries out
Illustrate:
Fig. 1 is eye fundus image blood vessel segmentation schematic network structure of the invention；
Fig. 2 is eye fundus image blood vessel segmentation flow chart of the invention.
Specific embodiment
The specific implementation process that the present invention will now be explained with reference to the accompanying drawings, attached drawing 1 are eye fundus image blood vessel of the invention point
Schematic network structure is cut, attached drawing 2 is eye fundus image blood vessel segmentation flow chart of the invention.
Step 1: the frequency domain classification processing of eye fundus image.Original image is pre-processed first, is carried out Fourier
Transformation is transformed into frequency domain, as shown in formula (1):
M, N respectively indicate the row and column of image size；F (x, y) indicates input picture；X, y respectively indicate airspace cross, vertical seat
Mark；F (u, v) indicates the frequency domain value after Fourier transformation；U, v respectively indicate frequency domain cross, ordinate.Then f (x, y) is led to respectively
Cross gauss low frequency filter (GLPF) and Gauss high-pass filter (GHPF), the transmission function of GLPF and GHPF are respectively H1(u,
And H v)2(u, v), as shown in formula (2) and (3):
Wherein D (u, v) indicates the distance away from Fourier transformation center origin, D0It is off frequency.By formula (2) and (3) point
The low-frequency component G of eye fundus image is not obtained multiplied by formula (1)1(u, v) and radio-frequency component G2(u,v).To G1(u, v) and G2(u, v) point
Not carry out Fourier inversion, obtain filtered low-frequency image f1(x, y) and high frequency imaging f2(x, y), as shown in formula (4):
Step 2: building low-dimensional feature extraction convolutional network.In view of low-frequency image mainly reflects the global wheel of eye fundus image
Wide information, the attribute with low dimensional feature, and low-frequency image can promote inspection to the assurance of Global Information to a certain extent
Survey precision.Therefore by low-frequency image f obtained in step 11(x, y) is input to constructed low-dimensional feature extraction convolutional network
In, as shown in 1 solid box of attached drawing, which is made of 4 convolutional layers and 4 block.2 block of left pathways, each
Block includes two residual blocks and a down-sampling, is mainly used for extracting the global feature of eye fundus image vascular distribution；The right road
Diameter 2 block, each block include two residual blocks and a up-sampling, are mainly used for the accurate fixed of eye fundus image blood vessel
Position.Wherein down-sampling uses bilinear interpolation value method using 2 × 2 maximum pondization operations, up-sampling.Specific step is as follows:
1.: image f1(x, y) after 1 × 1-32 convolution kernel by first left block when, wherein each residual block packet
Containing 64 convolution kernels, 64 dimensional feature figure F are converted images into1, as shown in formula (5):
F1=Res2 (pool (conv (f1(x,y)))) (5)
Wherein conv indicates convolution operation；Pool indicates pondization operation, using 2 × 2 maximum pond methods, similarly hereinafter；
Res2 indicates 2 residual block operations, similarly hereinafter.
2.: by F1It is activated by ReLu function, standardization processing is then carried out, by the 2nd, left side block, wherein wrapping
Containing 128 convolution kernels, characteristic pattern F is obtained2, as shown in formula (6):
F2=Res2 (pool (Norm (Re Lu (F1)))) (6)
Wherein ReLu indicates activation primitive, similarly hereinafter；Norm indicates standardization processing, similarly hereinafter.
3.: by F2Characteristic pattern after 1 × 1 × 256 convolution kernel obtains feature by the 1st block of right pathways
Scheme F3, as shown in formula (7):
F3=unsampling (Res2 (conv (F2))) (7)
Wherein unsampling indicates up-sampling operation, using bilinear interpolation value method, similarly hereinafter.
4.: by F3The final characteristic pattern F of low-frequency image is obtained by the 2nd block of right pathwaysL, as shown in formula (8):
FL=unsampling (Res2 (Norm (Re Lu (F3)))) (8)
Step 3: building high dimensional feature extracts convolutional network.In view of high frequency imaging mainly reflects the profile details of image,
It is the further reinforcing in low-frequency information to picture material.The network of the characteristics of for high frequency imaging, building are more concerned about details
Problem, the main minutia for capturing different layers.Therefore by high frequency imaging f obtained in step 12(x, y) is input to constructed
High-frequency characteristic extract convolutional network in, as shown in 1 dotted line frame of attached drawing, the high-frequency characteristic extract convolutional network include 3 volume
Lamination, 7 block.Its right and left each 3 block, each block include two residual blocks and a down-sampling or on adopt
Sample；Having 1 block in left and right path junction includes two residual blocks, a down-sampling and a up-sampling.The network institute
The convolutional layer weight in convolutional layer and low-dimensional feature extraction convolutional network for including is shared.Specific step is as follows:
1.: image is two residual blocks by 3 block of left pathways, each block after 1 × 1-32 convolution kernel
It is combined with a down-sampling, obtains characteristic pattern F3', as shown in formula (9):
Fi'=Res2 (pool (Norm (Re Lu (F 'i-1)))) (9)
Wherein i=1,2,3 indicate the number of left pathways block, F0' as shown in formula (10):
F0'=conv (f2(x,y)) (10)
2.: by F3' by the block of left and right path junction, which is a down-sampling, two residual blocks and one
A up-sampling combination, obtains characteristic pattern F '4, as shown in formula (11):
F′4=unsampling (Res2 (pool (Norm (Re Lu (F3′))))) (11)
3.: by F4' passing through 3 block of right pathways, each block is that two residual blocks and a up-sampling combine, together
When each up-sampling during by merging port number with the characteristic pattern of left side symmetric path, then again to merging after
Characteristic pattern is operated, and lost spatial information during pond is compensated for, and the details for strengthening eye fundus image blood vessel is special
Sign, will obtain relatively sharp characteristic pattern FH, as shown in formula (12):
Wherein j=4,5,6 and Fj' indicate the characteristic pattern for passing through j-th of block；copy(F′7-j)+Fj' indicate characteristic pattern
Fj' with the characteristic pattern of left side symmetric path merge port number.
Step 4: by characteristic pattern F obtained in step 2 and step 3LAnd FHIt merges, i.e., corresponding channel number is added, and is adopted
With 1 × 1-32 convolution kernel determine image corresponding to characteristic pattern F ', using 1 × 1-1 convolution kernel by F ' be changed into single channel spy
Sign figure F " obtains the output pixel value of vessel segmentation corresponding to original image f (x, y) after the activation of ReLu function, and
Loss operation is carried out with the difference of two squares with corresponding known blood vessel segmentation label, the loss operation result of all training images is carried out
It is cumulative, it is denoted as loss, as shown in formula (13):
Wherein n is the sample number of training image, and M, N are the length and width dimensions of training image,For i-th of training image pair
Answer vessel segmentation in the output pixel value of the position (j, k),For corresponding known blood vessel segmentation label value.It is finally right
Loss value carries out backpropagation, updates high and low dimensional feature respectively using gradient descent method and extracts weight in convolutional network and partially
It sets, when loss value is less than threshold epsilon, training terminates, and ε may be configured as the 1~3% of training image sampled pixel sum, that is, obtains
Network model after training.
Step 5: after low, high dimensional feature extracts convolutional network training, by the eye fundus image of Unknown Label by step 1~
4 are handled, and the single channel characteristic pattern F " corresponding to it is obtained, after the activation of ReLu function, the as blood vessel of eye fundus image
Segmentation result.
Claims (1)
1. a kind of eye fundus image blood vessel segmentation method based on frequency domain classification, which is characterized in that this method specifically includes following step
It is rapid:
Step 1: the frequency domain classification processing of eye fundus image
Original eye fundus image f (x, y) is transformed into frequency domain F (u, v) by Fourier transformation first, it is then low by Gauss respectively
Bandpass filter and Gauss high-pass filter obtain the low frequency and radio-frequency component of eye fundus image, finally utilize inverse fourier transform pair
The low frequency and high-frequency information of eye fundus image are handled, and the frequency domain classification results of eye fundus image, respectively low-frequency information institute are obtained
Corresponding f1(x, y), f corresponding to high-frequency information2(x,y)；
Step 2: building low-dimensional feature extraction convolutional network describes the global profile characteristic of eye fundus image using low-frequency component；It should
Network is divided into two paths in left and right, and whole network includes 8 residual blocks, 2 down-samplings, 2 up-samplings and 4 convolutional layers；Its
In, each residual block includes two 3 × 3 empty convolutional layers, every two residual block and a down-sampling or a up-sampling
1 block is formed, totally 4 block；The result of each block output is activated by ReLu function, then again
Carry out standardization processing；Left pathways extract the global feature of eye fundus image vascular distribution by 2 block；2, the right
For block for being accurately positioned, each block that wherein left pathways pass through is the combination of two residual blocks and a down-sampling,
Each block that right pathways pass through is the combination of two residual blocks and a up-sampling；The low-frequency image f that step 1 is obtained1
(x, y) is input in constructed low-dimensional feature extraction convolutional network, obtains characteristic pattern FL；
Step 3: building high dimensional feature extracts convolutional network, and the details characteristic of eye fundus image is described using radio-frequency component；The network
Equally it is divided into two paths in left and right, whole network includes 14 residual blocks, 4 down-samplings, 4 up-samplings and 3 convolutional layers；Its
In, each residual block includes that two 3 × 3 empty convolutional layers, every two residual block and a down-sampling or up-sampling form
One block, totally 6 block, 1 block in left and right path junction include two residual blocks, a down-sampling and one
A up-sampling；The result of each block output is activated by ReLu function, followed by standardization processing；
Left pathways capture eye fundus image vessel profile information by 3 block；Right pathways by 3 block simultaneously with the left side pair
Path is claimed to carry out mixing operation, mixing operation merges characteristic pattern port number；Each block that wherein left pathways pass through is two
The combination of a residual block and a down-sampling, each block that wherein right pathways pass through are to adopt on two residual blocks and one
The combination of sample；The high frequency imaging f that step 1 is obtained2(x, y) is input to constructed high dimensional feature and extracts in convolutional network, obtains
To characteristic pattern FH；
Step 4: being added using port number by two groups of characteristic pattern FLAnd FHMerge, obtains high low-dimensional by 1 × 1-32 convolution kernel and merge
Characteristic pattern F ', characteristic pattern F ' is then changed into single channel characteristic pattern F " using the convolution kernel of 1 × 1-1, using ReLu letter
The output pixel value of vessel segmentation corresponding to original image f (x, y) is obtained after number activation, and is divided with corresponding known blood vessel
It cuts label and loss operation is carried out with the difference of two squares, add up to the loss operation result of all training images, be as a result denoted as loss,
And high and low dimensional feature is respectively trained using gradient descent method and extracts convolutional network；After penalty values loss meets the condition of convergence, knot
Shu Xunlian；
Step 5: after high and low dimensional feature extracts convolutional network training, by the eye fundus image of Unknown Label by step 1~4 into
Row processing, obtains the single channel characteristic pattern F " corresponding to it, after the activation of ReLu function, the as blood vessel segmentation of eye fundus image
As a result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811331874.XA CN109671094B (en) | 2018-11-09 | 2018-11-09 | Fundus image blood vessel segmentation method based on frequency domain classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811331874.XA CN109671094B (en) | 2018-11-09 | 2018-11-09 | Fundus image blood vessel segmentation method based on frequency domain classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109671094A true CN109671094A (en) | 2019-04-23 |
| CN109671094B CN109671094B (en) | 2023-04-18 |
Family
ID=66142084
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811331874.XA Active CN109671094B (en) | 2018-11-09 | 2018-11-09 | Fundus image blood vessel segmentation method based on frequency domain classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109671094B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490843A (en) * | 2019-07-23 | 2019-11-22 | 苏州国科视清医疗科技有限公司 | A kind of eye fundus image blood vessel segmentation method |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN110807762A (en) * | 2019-09-19 | 2020-02-18 | 温州大学 | Intelligent retinal blood vessel image segmentation method based on GAN |
| CN110991611A (en) * | 2019-11-29 | 2020-04-10 | 北京市眼科研究所 | Full convolution neural network based on image segmentation |
| CN111489328A (en) * | 2020-03-06 | 2020-08-04 | 浙江工业大学 | Fundus image quality evaluation method based on blood vessel segmentation and background separation |
| CN111681252A (en) * | 2020-05-30 | 2020-09-18 | 重庆邮电大学 | Medical image automatic segmentation method based on multipath attention fusion |
| CN112365493A (en) * | 2020-11-30 | 2021-02-12 | 上海鹰瞳医疗科技有限公司 | Training data generation method and device for fundus image recognition model |
| CN112766178A (en) * | 2021-01-22 | 2021-05-07 | 广州大学 | Method, device, equipment and medium for positioning pests based on intelligent pest control system |
| CN113592790A (en) * | 2021-07-16 | 2021-11-02 | 大连民族大学 | Panoramic image segmentation method based on high-frequency and low-frequency enhancement, computer system and medium |
| CN113627285A (en) * | 2021-07-26 | 2021-11-09 | 长沙理工大学 | Video forensics method, system, and medium |
| CN116681980A (en) * | 2023-07-31 | 2023-09-01 | 北京建筑大学 | Deep learning-based large-deletion-rate image restoration method, device and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105551010A (en) * | 2016-01-20 | 2016-05-04 | 中国矿业大学 | Multi-focus image fusion method based on NSCT (Non-Subsampled Contourlet Transform) and depth information incentive PCNN (Pulse Coupled Neural Network) |
| US20160217586A1 (en) * | 2015-01-28 | 2016-07-28 | University Of Florida Research Foundation, Inc. | Method for the autonomous image segmentation of flow systems |
| US20170256033A1 (en) * | 2016-03-03 | 2017-09-07 | Mitsubishi Electric Research Laboratories, Inc. | Image Upsampling using Global and Local Constraints |
| CN107657612A (en) * | 2017-10-16 | 2018-02-02 | 西安交通大学 | Suitable for full-automatic the retinal vessel analysis method and system of intelligent and portable equipment |
| CN108520225A (en) * | 2018-03-30 | 2018-09-11 | 南京信息工程大学 | A kind of fingerprint detection sorting technique based on spatial alternation convolutional neural networks |
-
2018
- 2018-11-09 CN CN201811331874.XA patent/CN109671094B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160217586A1 (en) * | 2015-01-28 | 2016-07-28 | University Of Florida Research Foundation, Inc. | Method for the autonomous image segmentation of flow systems |
| CN105551010A (en) * | 2016-01-20 | 2016-05-04 | 中国矿业大学 | Multi-focus image fusion method based on NSCT (Non-Subsampled Contourlet Transform) and depth information incentive PCNN (Pulse Coupled Neural Network) |
| US20170256033A1 (en) * | 2016-03-03 | 2017-09-07 | Mitsubishi Electric Research Laboratories, Inc. | Image Upsampling using Global and Local Constraints |
| CN107657612A (en) * | 2017-10-16 | 2018-02-02 | 西安交通大学 | Suitable for full-automatic the retinal vessel analysis method and system of intelligent and portable equipment |
| CN108520225A (en) * | 2018-03-30 | 2018-09-11 | 南京信息工程大学 | A kind of fingerprint detection sorting technique based on spatial alternation convolutional neural networks |
Non-Patent Citations (3)
| Title |
|---|
| KIEN NGUYEN；ET.AL: "Fusing shrinking and expanding active contour models for robust iris segementation" * |
| 李媛媛;蔡轶珩;高旭蓉;: "基于融合相位特征的视网膜血管分割算法" * |
| 王强;范影乐;武薇;朱亚萍;: "正立和倒立面孔的混合识别" * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490843A (en) * | 2019-07-23 | 2019-11-22 | 苏州国科视清医疗科技有限公司 | A kind of eye fundus image blood vessel segmentation method |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN110619633B (en) * | 2019-09-10 | 2023-06-23 | 武汉科技大学 | Liver image segmentation method based on multipath filtering strategy |
| CN110807762A (en) * | 2019-09-19 | 2020-02-18 | 温州大学 | Intelligent retinal blood vessel image segmentation method based on GAN |
| CN110991611A (en) * | 2019-11-29 | 2020-04-10 | 北京市眼科研究所 | Full convolution neural network based on image segmentation |
| CN111489328A (en) * | 2020-03-06 | 2020-08-04 | 浙江工业大学 | Fundus image quality evaluation method based on blood vessel segmentation and background separation |
| CN111489328B (en) * | 2020-03-06 | 2023-06-30 | 浙江工业大学 | Fundus image quality evaluation method based on blood vessel segmentation and background separation |
| CN111681252B (en) * | 2020-05-30 | 2022-05-03 | 重庆邮电大学 | Medical image automatic segmentation method based on multipath attention fusion |
| CN111681252A (en) * | 2020-05-30 | 2020-09-18 | 重庆邮电大学 | Medical image automatic segmentation method based on multipath attention fusion |
| CN112365493A (en) * | 2020-11-30 | 2021-02-12 | 上海鹰瞳医疗科技有限公司 | Training data generation method and device for fundus image recognition model |
| CN112766178B (en) * | 2021-01-22 | 2023-06-23 | 广州大学 | Disease and pest positioning method, device, equipment and medium based on intelligent deinsectization system |
| CN112766178A (en) * | 2021-01-22 | 2021-05-07 | 广州大学 | Method, device, equipment and medium for positioning pests based on intelligent pest control system |
| CN113592790A (en) * | 2021-07-16 | 2021-11-02 | 大连民族大学 | Panoramic image segmentation method based on high-frequency and low-frequency enhancement, computer system and medium |
| CN113592790B (en) * | 2021-07-16 | 2023-08-29 | 大连民族大学 | Panoramic image segmentation method based on high-low frequency reinforcement, computer system and medium |
| CN113627285A (en) * | 2021-07-26 | 2021-11-09 | 长沙理工大学 | Video forensics method, system, and medium |
| CN116681980A (en) * | 2023-07-31 | 2023-09-01 | 北京建筑大学 | Deep learning-based large-deletion-rate image restoration method, device and storage medium |
| CN116681980B (en) * | 2023-07-31 | 2023-10-20 | 北京建筑大学 | Deep learning-based large-deletion-rate image restoration method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109671094B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109671094A (en) | A kind of eye fundus image blood vessel segmentation method based on frequency domain classification | |
| CN107220980B (en) | A kind of MRI image brain tumor automatic division method based on full convolutional network | |
| CN106920227B (en) | The Segmentation Method of Retinal Blood Vessels combined based on deep learning with conventional method | |
| CN111145170B (en) | Medical image segmentation method based on deep learning | |
| CN108596884B (en) | Esophagus cancer segmentation method in chest CT image | |
| CN112508864B (en) | Retinal vessel image segmentation method based on improved UNet + | |
| CN110930416B (en) | MRI image prostate segmentation method based on U-shaped network | |
| CN111161273B (en) | Medical ultrasonic image segmentation method based on deep learning | |
| CN109767440A (en) | A kind of imaged image data extending method towards deep learning model training and study | |
| CN110298262A (en) | Object identification method and device | |
| CN107492071A (en) | Medical image processing method and equipment | |
| CN109685813A (en) | A kind of U-shaped Segmentation Method of Retinal Blood Vessels of adaptive scale information | |
| CN109829855A (en) | A kind of super resolution ratio reconstruction method based on fusion multi-level features figure | |
| CN110232653A (en) | The quick light-duty intensive residual error network of super-resolution rebuilding | |
| CN103699904B (en) | The image computer auxiliary judgment method of multisequencing nuclear magnetic resonance image | |
| CN107437092A (en) | The sorting algorithm of retina OCT image based on Three dimensional convolution neutral net | |
| CN108764286A (en) | The classifying identification method of characteristic point in a kind of blood-vessel image based on transfer learning | |
| CN110097554A (en) | The Segmentation Method of Retinal Blood Vessels of convolution is separated based on intensive convolution sum depth | |
| Zhang et al. | LU-NET: An improved U-Net for ventricular segmentation | |
| CN110675411A (en) | Cervical squamous intraepithelial lesion recognition algorithm based on deep learning | |
| CN107644420A (en) | Blood-vessel image dividing method, MRI system based on central line pick-up | |
| CN108764342B (en) | Semantic segmentation method for optic discs and optic cups in fundus image | |
| CN112001928B (en) | Retina blood vessel segmentation method and system | |
| CN110084823A (en) | Three-dimensional brain tumor image partition method based on cascade anisotropy FCNN | |
| CN108416821A (en) | A kind of CT Image Super-resolution Reconstruction methods of deep neural network |
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 |