CN106504199B - A kind of eye fundus image Enhancement Method and system - Google Patents
A kind of eye fundus image Enhancement Method and system Download PDFInfo
- Publication number
- CN106504199B CN106504199B CN201610822306.4A CN201610822306A CN106504199B CN 106504199 B CN106504199 B CN 106504199B CN 201610822306 A CN201610822306 A CN 201610822306A CN 106504199 B CN106504199 B CN 106504199B
- Authority
- CN
- China
- Prior art keywords
- subimage block
- blood vessel
- eye fundus
- fundus image
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 210000004204 blood vessel Anatomy 0.000 claims abstract description 136
- 238000001914 filtration Methods 0.000 claims abstract description 30
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 238000010276 construction Methods 0.000 claims abstract description 8
- 230000002792 vascular Effects 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 11
- 230000003287 optical effect Effects 0.000 claims description 11
- 239000008280 blood Substances 0.000 claims description 10
- 210000004369 blood Anatomy 0.000 claims description 10
- 238000003709 image segmentation Methods 0.000 claims description 7
- 210000001508 eye Anatomy 0.000 description 64
- 230000002708 enhancing effect Effects 0.000 description 10
- 230000033115 angiogenesis Effects 0.000 description 3
- 210000005252 bulbus oculi Anatomy 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 239000002671 adjuvant Substances 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 208000030533 eye disease Diseases 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G06T5/77—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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
- G06T2207/20172—Image enhancement details
- G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
-
- 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
Abstract
The present invention relates to eye fundus image Enhancement Method and systems.Learn picture construction blood vessel dictionary using eyeground；Frangi filtering is carried out to eye fundus image to be reinforced, the blood vessel in the second subimage block is divided into thick blood vessel and thin and delicate blood vessel by utilization orientation filtering, and sets the residual error weight and threshold residual value of angiosomes accordingly；By each first subimage block inner product in the second subimage block and dictionary, maximum first subimage block of inner product is chosen, and calculate its corresponding sparse coefficient；Residual image is calculated using the first subimage block of selection and the second subimage block, and the residual error in residual error weight calculation the second subimage block medium vessels region using angiosomes, if residual error is greater than threshold residual value, the second subimage block then is set by residual image, final election of laying equal stress on takes the first subimage block and calculates residual error process；The second subimage block is reconstructed using sparse coefficient, and the second subimage block of each reconstruct is recombinated, the eye fundus image enhanced.The present invention effectively inhibits ambient noise, remains thin and delicate blood vessel.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of eye fundus image Enhancement Methods and system.
Background technique
Ocular imaging is the important means of medicine assisting in diagnosis and treatment, can be directly or indirectly by analysis eyeball blood-vessel image
Judge many eye diseases.In eyes image, there are the ocular angiogenesis of various different thicknesses degree, enhance these blood vessels,
Available apparent accurate ocular angiogenesis image is conducive to adjuvant clinical diagnosis.
Eye fundus image Enhancement Method has very much, and general common method has:
Field exponential smoothing.I.e. using the average value of pixel a certain in image and its neighborhood territory pixel gray scale as the pixel
Gray value.The advantages of this method is simply, the disadvantage is that ocular angiogenesis image can be made to thicken, to greatly reduce blood vessel
Clarity.
Save edge smoothing method.Different templates are designed, neighborhood territory pixel gray scale locating for a certain pixel in image is calculated
Variance, using the average gray of pixel contained by the smallest template of variance as the gray value of the pixel.The advantages of this method
It is that can preferably save boundary, the disadvantage is that target is cable architecture in eye fundus image, it is difficult to which noise and target are distinguished by variance.
Averaging of multiple image.Such method is that several eyeball blood-vessel images of same people is taken to be averaging processing.This method
The advantages of be centainly to inhibit noise the disadvantage is that needing multiple eyeball blood-vessel images not be suitable for individual eye fundus image at degree.
The enhancing of Frangi filtering image.Feature vector direction and spy of this method using the Hessian matrix of cable architecture
Value indicative enhances cable architecture, but such methods can be such that thin and delicate weak blood vessel loses.
Image de-noising method based on rarefaction representation obtains redundant dictionary by training, reconstructs original image further according to sparse coefficient
Picture, since the dictionary atom of selection does not have noise, so as to obtain noise suppression image.This method is imitated with preferable noise suppression
Fruit, but there are still lose thin and delicate weak blood vessel in being applied to eye fundus image enhancing problem.
It can be seen that existing eye fundus image Enhancement Method all can not preferably protect while carrying out optical fundus blood vessel enhancing
Stay thin and delicate blood vessel.
Summary of the invention
The technical problem to be solved by the invention is to provide a kind of eye fundus image Enhancement Method and systems, it is intended to solve existing
The problem of having the eye fundus image Enhancement Method of technology preferably can not retain thin and delicate blood vessel while carrying out optical fundus blood vessel enhancing.
The present invention is implemented as follows:
A kind of eye fundus image Enhancement Method, includes the following steps:
Step A: learning picture construction blood vessel dictionary using eyeground, includes the first son of setting quantity in the blood vessel dictionary
Image block；
Step B: Frangi filtering is carried out to eye fundus image to be reinforced, and the image that Frangi is filtered is divided into
Several the second overlapped subimage blocks；
Step C: utilization orientation filter carries out trend pass filtering to second subimage block, and according to trend pass filtering result
Judge that the optical fundus blood vessel for including in second subimage block is thick blood vessel or thin and delicate blood vessel；
Step D: the angiosomes in second subimage block are determined, and include according in second subimage block
The residual error weight and threshold residual value of the angiosomes in second subimage block is arranged in the type of optical fundus blood vessel；
Step E: by each first subimage block inner product in second subimage block and the blood vessel dictionary, it is determined
Middle maximum first subimage block of inner product, and calculate the corresponding sparse coefficient of maximum first subimage block of the inner product；
Step F: calculating residual image using maximum first subimage block of the inner product and second subimage block, and
The residual error in the second subimage block medium vessels region described in residual error weight calculation using the angiosomes；
Step G: when the norm of the residual error is greater than the threshold residual value, the second subgraph is set by residual image
Block, and the E that gos to step, otherwise, go to step H；
Step H: second subimage block is reconstructed using the sparse coefficient；
Step I: the eye fundus image is reconstructed using the second subimage block of each reconstruct, thus the eyeground figure enhanced
Picture.
Further, the step A includes:
Step A1: by eyeground study image segmentation at identical first subimage block of several sizes；First son
The quantity of image block is greater than the setting quantity；
Step A2: each first subimage block is subjected to inner product two-by-two；
Step A3: it chooses the first subimage block of the smallest setting quantity of inner product and constructs the blood vessel dictionary.
Further, the step B includes:
Step B1: eye fundus image to be reinforced is set as I (x, y), the two-dimensional Gaussian function that scale is σ is G (x, y；σ), sharp
The eye fundus image I (x, y) to be reinforced is smoothed with the two-dimensional Gaussian function, obtains smoothed image Iσ(x,
Y):
Wherein, For convolution operation；
Step B2: at scale σ, smoothed image I is calculatedσHessian matrix H at midpoint (x, y) (x, y)σ(x, y):
Step B3: to the Hessian matrix Hσ(x, y) does Eigenvalues analysis, obtains eigenvalue λ1、λ2, | λ1| < | λ2|；
Blood vessel feature under scale s are as follows:
Wherein,β and C is preset constant；
Step B4: under multiple dimensioned, v under each scale is taken0(s) maximum value as the eye fundus image I to be reinforced (x,
Y) Frangi filter result v:
Wherein, sminAnd smaxIt is smallest dimension and out to out respectively；
Step B5: the Frangi filter result v is divided into several the second overlapped subimage blocks.
Further, the step C includes:
Step C1: setting direction is respectively θ1=0, 8 anisotropic filters；
Step C2: assuming that direction is θiAnisotropic filter medium vessels region be Ω1, non-vascular region is Ω2, calculate two
A respective energy in regionWith
Wherein v (x, y) is value of the Frangi filter result v at (x, y), N1It is Ω1Middle number of pixels, N2It is Ω2Middle pixel
Number；
Step C3: it calculatesWithEnergy difference:
Step C4: maximum energy difference in above-mentioned 8 directions is determined:
Step C5: according to the EmaxVascular group is judged, if Emax>=T then includes in second subimage block
Eye fundus image is thick blood vessel, is otherwise thin and delicate blood vessel.
Further, the step D includes:
Step D1: using the angiosomes in anisotropic filter corresponding to maximum energy difference in 8 directions as
The actual region Ω of blood vessel1, by the non-vascular region in anisotropic filter corresponding to maximum energy difference in 8 directions
As the actual region Ω of non-vascular2；
Step D2: for comprising eye fundus image be thick blood vessel the second subimage block, by its angiosomes Ω1Residual error
Weight is set as 1, threshold residual value TR=T1；For comprising eye fundus image be thin and delicate blood vessel the second subimage block, by its area vasculosa
Domain Ω1Residual error weight be set as 1/vmax, threshold residual value TR=T2, wherein vmaxIt is the second subimage block Frangi filter result
Maximum value.
Further, the step E includes:
Step E1: the second subimage block vector is turned into x, i-th of first subimage blocks are in the blood vessel dictionary
di；
Step E2: the first subimage block each in the blood vessel dictionary and described second subimage block x inner product the maximum are made
For first the first subimage block d chosenr0:
Wherein, k is the number of the first subimage block in the blood vessel dictionary, r0It is the call number of dictionary, < x, di> is x
With diInner product operation；
Step E3: the first subimage block d is calculatedr0Corresponding sparse coefficient αr0:
αr0=< x, dr0>。
Further, the step F includes:
Step F1: the residual image R in the second subimage block medium vessels region is calculated:
R=x- < x, dr0>dr0；
Step F2: the residual error R is summed multiplied by the residual error Weight in the second subimage block medium vessels region, as
The final residual error in the second subimage block medium vessels region.
Further, second subimage block of reconstruct are as follows:
Wherein, S is the set that multiple sparse coefficients that step E is determined are performed a plurality of times, dr0It is that execution step E is true each time
Maximum first subimage block of the inner product made, αr0It is dr0Corresponding sparse coefficient.
Further, the step I includes:
Disjoint part of second subimage block of all reconstruct is merged, the eye fundus image completely enhanced.
A kind of eye fundus image enhancing system, comprising:
Blood vessel dictionary constructs module, learns picture construction blood vessel dictionary using eyeground, includes setting in the blood vessel dictionary
First subimage block of fixed number amount；
Image filtering and division module carry out Frangi filtering to eye fundus image to be reinforced, and Frangi are filtered
Obtained image is divided into several the second overlapped subimage blocks；
Vascular group judgment module, utilization orientation filter carry out trend pass filtering, and root to second subimage block
Judge that the optical fundus blood vessel for including in second subimage block is thick blood vessel or thin and delicate blood vessel according to trend pass filtering result；
Angiosomes and its residual error weight and threshold residual value determining module, determine the blood vessel in second subimage block
Region, and the blood vessel in second subimage block is arranged according to the type for the optical fundus blood vessel for including in second subimage block
The residual error weight and threshold residual value in region；
Sparse coefficient computing module, by each first subimage block in second subimage block and the blood vessel dictionary
Inner product determines wherein maximum first subimage block of inner product, and it is corresponding to calculate maximum first subimage block of the inner product
Sparse coefficient；
Angiosomes residual computations module utilizes maximum first subimage block of the inner product and second subgraph
Block calculates residual image, and utilizes the residual of the second subimage block medium vessels region described in the residual error weight calculation of the angiosomes
Difference；
Jump module sets second for residual image when the norm of the residual error is greater than the threshold residual value
Subimage block, and the sparse coefficient computing module is jumped to, otherwise, jump to the second subimage block reconstructed module；
Second subimage block reconstructed module reconstructs second subimage block using the sparse coefficient；
Eye fundus image reconstructed module reconstructs the eye fundus image using the second subimage block of each reconstruct, to obtain
The eye fundus image of enhancing.
Further, the blood vessel dictionary building module includes:
Eyeground learns image division module, by eyeground study image segmentation at identical first subgraph of several sizes
As block；The quantity of first subimage block is greater than the setting quantity；
First subimage block inner product module, carries out inner product for each first subimage block two-by-two；
Blood vessel dictionary constructs submodule, chooses described in the smallest setting quantity of inner product the first subimage block building
Blood vessel dictionary.
Further, described image filtering and division module include:
Smothing filtering module, sets eye fundus image to be reinforced as I (x, y), the two-dimensional Gaussian function that scale is σ be G (x,
y；σ), the eye fundus image I (x, y) to be reinforced is smoothed using the two-dimensional Gaussian function, is smoothly schemed
As Iσ(x, y):
Wherein, For convolution operation；
Hessian matrix computing module calculates smoothed image I at scale σσAt midpoint (x, y) (x, y)
Hessian matrix Hσ(x, y):
Eigenvalues analysis module, to the Hessian matrix Hσ(x, y) does Eigenvalues analysis, obtains eigenvalue λ1、λ2,
|λ1| < | λ2|；Blood vessel feature under scale s are as follows:
Wherein,β and C is preset constant；
Frangi filter result generation module, takes v under each scale0(s) maximum value is as the eyeground to be reinforced
The Frangi filter result v of image I (x, y):
Wherein, sminAnd smaxIt is smallest dimension and out to out respectively；
The Frangi filter result v is divided into several the second overlapped sons by the second subgraph division module
Image block.
Further, the vascular group judgment module includes:
Anisotropic filter setup module, setting direction are respectively θ1=0, 8 anisotropic filters；
Energy computation module, hypothesis direction are θiAnisotropic filter medium vessels region be Ω1, non-vascular region is
Ω2, calculate two respective energy in regionWith
Wherein v (x, y) is value of the Frangi filter result v at (x, y), N1It is Ω1Middle number of pixels, N2It is Ω2Middle pixel
Number；
Energy difference computing module calculatesWithEnergy difference:
Ceiling capacity difference determining module determines maximum energy difference in above-mentioned 8 directions:
Vascular group judging submodule, according to the EmaxVascular group is judged, if Emax>=T, then described second is sub
The eye fundus image for including in image block is thick blood vessel, is otherwise thin and delicate blood vessel.
Further, the angiosomes and its residual error weight and threshold residual value determining module include:
Angiosomes determining module, will be in anisotropic filter corresponding to maximum energy difference in 8 directions
Angiosomes are as the actual region Ω of blood vessel1, will be in anisotropic filter corresponding to maximum energy difference in 8 directions
Angiosomes as the actual region Ω of blood vessel2；
Residual error weight and threshold residual value determining module, by comprising eye fundus image be thick blood vessel the second subimage block
Angiosomes Ω1Residual error weight be set as 1, threshold residual value TR=T1；By comprising eye fundus image be thin and delicate blood vessel the second subgraph
As the angiosomes Ω of block1Residual error weight be set as 1/vmax, threshold residual value TR=T2, wherein vmaxIt is second subimage block
The maximum value of Frangi filter result.
Further, the sparse coefficient computing module includes:
The second subimage block vector is turned to x, i-th first sons in the blood vessel dictionary by image vector module
Image block is di；
First subimage block selecting module, by the first subimage block each in the blood vessel dictionary and second subgraph
Block x inner product the maximum is as first the first subimage block d chosenr0:
Wherein, k is the number of the first subimage block in the blood vessel dictionary, r0It is the call number of dictionary, < x, di> be x with
diInner product operation；
Sparse coefficient computational submodule calculates the first subimage block dr0Corresponding sparse coefficient αr0:αr0=< x, dr0>。
Further, the angiosomes residual computations module includes:
Residual error just calculates module, calculates the residual image R in the second subimage block medium vessels region:
R=x- < x, dr0>dr0；
Residual weighted module, by the residual error R multiplied by the residual error Weight in the second subimage block medium vessels region
Summation, the final residual error as the second subimage block medium vessels region.
Further, second subimage block of reconstruct are as follows:
Wherein, S is the set for multiple sparse coefficients that the sparse coefficient computing module is repeatedly determined, dr0It is described dilute
Maximum first subimage block of the inner product that sparse coefficient computing module is determined each time, αr0It is dr0Corresponding sparse coefficient.
Further, the eye fundus image reconstructed module is specifically used for:
Disjoint part of second subimage block of all reconstruct is merged, the eye fundus image completely enhanced.
The present invention learns picture construction blood vessel dictionary using eyeground；Utilization orientation is filtered the blood vessel in the second subimage block
It is classified as thick blood vessel and thin and delicate blood vessel, for the residual error weight and threshold residual value of thick blood vessel and thin and delicate blood vessel setting angiosomes；
Each first subimage block in second subimage block and dictionary is subjected to inner product, chooses maximum first subimage block of inner product, and
Calculate its corresponding sparse coefficient；Utilize the first subimage block of selection and the second subgraph of residual error weight calculation of angiosomes
The residual error in block medium vessels region repeats to choose the first subimage block and calculates the mistake of residual error if residual error is greater than threshold residual value
Journey；The second subimage block is reconstructed using sparse coefficient, and the second subimage block of each reconstruct is recombinated, the eyeground figure enhanced
Picture.The present invention reduces filter progress blood vessel enhancing by Frangi in the prior art ambient noise and thin and delicate blood vessel is brought to lose
The phenomenon that, the enhancing of eye fundus image is realized, eye fundus image visual effect is improved, can be used for the pre- place of eye fundus image analysis
Reason.
Detailed description of the invention
Fig. 1: the overall procedure schematic diagram of eye fundus image Enhancement Method provided by the invention；
Fig. 2: the main assembly schematic diagram of eye fundus image enhancing system provided by the invention；
The direction schematic diagram of Fig. 3: 8 anisotropic filters.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.
As shown in Figure 1, eye fundus image Enhancement Method provided by the invention, includes the following steps:
Step A: learning picture construction blood vessel dictionary using eyeground, includes the first subgraph of setting quantity in blood vessel dictionary
Block.
Step A is specifically included:
Step A1: by eyeground study image segmentation at identical first subimage block of several sizes, the first subimage block
Quantity should be greater than setting quantity.Eyeground study image segmentation is that 8*8 is big by the artificial segmentation result that can learn image according to eyeground
Small several first subimage blocks, these first subimage blocks will include the eyeground figures such as thick blood vessel, thin and delicate thin blood vessel, high bright part
As feature.
Step A2: each first subimage block is subjected to inner product two-by-two.Inner product is smaller, illustrates the phase of two the first subimage blocks
It is smaller like spending.
Step A3: it chooses the first subimage block of the smallest setting quantity of inner product and constructs blood vessel dictionary.Assuming that setting quantity
For K, chooses least similar K subimage block and constitute blood vessel dictionary.
Step B: Frangi filtering is carried out to eye fundus image to be reinforced, and the image that Frangi is filtered is divided into
Several the second overlapped subimage blocks.
Step B is specifically included:
Step B1: eye fundus image to be reinforced is set as I (x, y), the two-dimensional Gaussian function that scale is σ is G (x, y；σ), sharp
Eye fundus image I (x, y) to be reinforced is smoothed with two-dimensional Gaussian function, obtains smoothed image Iσ(x, y):
Wherein, For convolution operation.
Step B2: at scale σ, smoothed image I is calculatedσHessian matrix H at midpoint (x, y) (x, y)σ(x, y):
Step B3: to Hessian matrix Hσ(x, y) does Eigenvalues analysis, obtains eigenvalue λ1、λ2, | λ1| < | λ2|.If point
(x, y) belongs to tubular structure, then | λ1| ≈ 0, | λ2| value can be bigger, then the blood vessel feature under scale s are as follows:
Wherein,β and C is preset constant.
Step B4: under multiple dimensioned, v under each scale is taken0(s) maximum value is as eye fundus image I's (x, y) to be reinforced
Frangi filter result v:
Wherein, sminAnd smaxIt is smallest dimension and out to out respectively.
Step B5: Frangi filter result v is divided into several the second overlapped subimage blocks.
Step C: utilization orientation filter carries out trend pass filtering to the second subimage block, and is judged according to trend pass filtering result
The optical fundus blood vessel for including in second subimage block is thick blood vessel or thin and delicate blood vessel.
Step C is specifically included:
Step C1: setting direction is respectively θ1=0, 8 anisotropic filters (as shown in Figure 3).
Step C2: assuming that direction is θiAnisotropic filter medium vessels region (white area) be Ω1, non-vascular region is (black
Color region) it is Ω2, calculate two respective energy in regionWith
Wherein v (x, y) is value of the Frangi filter result v at (x, y), N1It is Ω1Middle number of pixels, N2It is Ω2Middle pixel
Number.
Step C3: it calculatesWithEnergy difference:
Step C4: maximum energy difference in above-mentioned 8 directions is determined:
Step C5: according to EmaxVascular group is judged, if Emax>=T, the then eye fundus image for including in the second subimage block
It is otherwise thin and delicate blood vessel for thick blood vessel.T is preset value.
Step D: the angiosomes in the second subimage block are determined, and according to the optical fundus blood vessel for including in the second subimage block
Type be arranged the second subimage block in angiosomes residual error weight and threshold residual value.
Step D is specifically included:
Step D1: by anisotropic filter corresponding to maximum energy difference in 8 directions (even if EmaxMaximum θiInstitute is right
The anisotropic filter answered) in angiosomes as the actual region Ω of blood vessel1, energy difference institute maximum in 8 directions is right
The anisotropic filter answered is (even if EmaxMaximum θiCorresponding anisotropic filter) in non-vascular region as non-vascular reality
Region Ω2。
Step D2: for comprising eye fundus image be thick blood vessel the second subimage block, by its angiosomes Ω1Residual error
Weight is set as 1, threshold residual value TR=T1；For comprising eye fundus image be thin and delicate blood vessel the second subimage block, by its area vasculosa
Domain Ω1Residual error weight be set as 1/vmax, threshold residual value TR=T2, wherein vmaxIt is the second subimage block Frangi filter result
Maximum value.
Step D may also include that
Step D3: it is sky, S=φ, the first subimage block that will be chosen that the first subimage block index set S chosen, which is arranged,
dr0Call number r0Set S, S=S ∪ r0 is added.
Step E: by each first subimage block inner product in the second subimage block and blood vessel dictionary, determine that wherein inner product is most
The first big subimage block, and calculate the corresponding sparse coefficient of maximum first subimage block of inner product.
Step E is specifically included:
Step E1: the second subimage block vector is turned into x, i-th of first subimage blocks are d in blood vessel dictionaryi。
Step E2: using the first subimage block each in blood vessel dictionary and second subimage block x inner product the maximum as choosing
First the first subimage block dr0:
Wherein, k is the number of the first subimage block in blood vessel dictionary, r0It is the call number of dictionary, < x, di> it is x and di's
Inner product operation.
Step E3: the first subimage block d is calculatedr0Corresponding sparse coefficient αr0:
αr0=< x, dr0>, the first subimage block d that will be chosenr0Call number r0Set S, S=S ∪ r0 is added.
Step F: residual image is calculated using maximum first subimage block of inner product and the second subimage block, and described in utilization
The residual error in residual error weight calculation the second subimage block medium vessels region of angiosomes.
Step F is specifically included:
Step F1: the residual image R in the second subimage block medium vessels region is calculated:
R=x- < x, dr0>dr0；
Step F2: residual error R is summed multiplied by the residual error Weight in the second subimage block medium vessels region, as second
The final residual error in subimage block medium vessels region.
Step G: when the norm of residual error is greater than threshold residual value, the second subimage block is set by residual image, and jump
To step E, otherwise, go to step H.The norm of residual error R i.e. in step F | | R | | it is greater than threshold residual value TR, then step is gone to
Otherwise rapid E goes to step H.
Step H: the second subimage block is reconstructed using sparse coefficient.Second subimage block of reconstruct are as follows:
Wherein, S is the set that multiple sparse coefficients that step E is determined are performed a plurality of times, dr0It is that execution step E is true each time
Maximum first subimage block of the inner product made, αr0It is dr0Corresponding sparse coefficient.
Step I: eye fundus image is reconstructed using the second subimage block of each reconstruct, thus the eye fundus image enhanced.
Step I includes:
Disjoint part of second subimage block of all reconstruct is merged, the eye fundus image completely enhanced.
As shown in Fig. 2, being based on aforementioned eye fundus image Enhancement Method, the present invention also provides a kind of enhancings of eye fundus image to be
System, comprising: blood vessel dictionary construct module 1, image filtering and division module 2, vascular group judgment module 5, angiosomes and its
Residual error weight and threshold residual value determining module 4, sparse coefficient computing module 3, angiosomes residual computations module 6, jump module
7, the second subimage block reconstructed module 8, eye fundus image reconstructed module 9.
Blood vessel dictionary constructs module 1 and learns picture construction blood vessel dictionary using eyeground, includes setting quantity in blood vessel dictionary
The first subimage block.It includes that eyeground learns image division module, the first subimage block inner product module that blood vessel dictionary, which constructs module 1,
Block, blood vessel dictionary construct submodule.
Eyeground learns image division module for eyeground study image segmentation into identical first subimage block of several sizes, the
The quantity of one subimage block is greater than setting quantity.
Each first subimage block is carried out inner product by the first subimage block inner product module two-by-two.
Blood vessel dictionary constructs submodule and chooses the smallest setting quantity of inner product the first subimage block building blood vessel dictionary.
Image filtering and division module 2 carry out Frangi filtering to eye fundus image to be reinforced, and Frangi is filtered
To image be divided into several the second overlapped subimage blocks.Image filtering and division module 2 include smothing filtering module,
Hessian matrix computing module, Eigenvalues analysis module, Frangi filter result generation module, the second subgraph division module.
Smothing filtering module sets eye fundus image to be reinforced as I (x, y), and the two-dimensional Gaussian function that scale is σ is G (x, y；
σ), eye fundus image I (x, y) to be reinforced is smoothed using two-dimensional Gaussian function, obtains smoothed image Iσ(x, y):
Wherein, For convolution operation.
Hessian matrix computing module calculates smoothed image I at scale σσHessian at midpoint (x, y) (x, y)
Matrix Hσ(x, y):
Eigenvalues analysis module is to Hessian matrix Hσ(x, y) does Eigenvalues analysis, obtains eigenvalue λ1、λ2, | λ1| < |
λ2|；Blood vessel feature under scale s are as follows:
Wherein,β and C is preset constant.
Frangi filter result generation module takes v under each scale0(s) maximum value as eye fundus image I to be reinforced (x,
Y) Frangi filter result v:
Wherein, sminAnd smaxIt is smallest dimension and out to out respectively.
Frangi filter result v is divided into several the second overlapped subimage blocks by the second subgraph division module.
5 utilization orientation filter of vascular group judgment module carries out trend pass filtering to the second subimage block, and according to direction
Filter result judges that the optical fundus blood vessel for including in second subimage block is thick blood vessel or thin and delicate blood vessel.Vascular group judges mould
Block 5 includes anisotropic filter setup module, energy computation module, energy difference computing module, ceiling capacity difference determining module, blood vessel
Type judging submodule.
Anisotropic filter setup module setting direction is respectively θ1=0, 8 anisotropic filters.
Energy computation module assumes that direction is θiAnisotropic filter medium vessels region be Ω1, non-vascular region is Ω2, meter
Calculate two respective energy in regionWith
Wherein v (x, y) is value of the Frangi filter result v at (x, y), N1It is Ω1Middle number of pixels, N2It is Ω2Middle pixel
Number.
Energy difference computing module calculatesWithEnergy difference:
Ceiling capacity difference determining module determines maximum energy difference in above-mentioned 8 directions:
Vascular group judging submodule is according to EmaxVascular group is judged, if Emax>=T is then wrapped in second subimage block
The eye fundus image contained is thick blood vessel, is otherwise thin and delicate blood vessel.
Angiosomes and its residual error weight and threshold residual value determining module 4 determine the angiosomes in the second subimage block,
And it is weighed according to the residual error that the angiosomes in the second subimage block are arranged in the type for the optical fundus blood vessel for including in the second subimage block
Weight and threshold residual value.Angiosomes and its residual error weight and threshold residual value determining module 4 include angiosomes determining module and residual
Poor weight and threshold residual value determining module.
Angiosomes determining module is by the angiosomes in anisotropic filter corresponding to maximum energy difference in 8 directions
As the actual region Ω of blood vessel1, the angiosomes in anisotropic filter corresponding to maximum energy difference in 8 directions are made
For the actual region Ω of blood vessel2。
Residual error weight and threshold residual value determining module by comprising eye fundus image be thick blood vessel the second subimage block blood
Area under control domain Ω1Residual error weight be set as 1, threshold residual value TR=T1；By comprising eye fundus image be thin and delicate blood vessel the second subgraph
The angiosomes Ω of block1Residual error weight be set as 1/vmax, threshold residual value TR=T2, wherein vmaxIt is second subimage block
The maximum value of Frangi filter result.
Sparse coefficient computing module 3 determines each first subimage block inner product in the second subimage block and blood vessel dictionary
Wherein maximum first subimage block of inner product out, and calculate the corresponding sparse coefficient of maximum first subimage block of inner product.It is sparse
Coefficients calculation block 3 includes image vector module, the first subimage block selecting module and sparse coefficient computational submodule.
Second subimage block vector is turned to x by image vector module, and i-th of first subimage blocks are d in blood vessel dictionaryi。
First subimage block selecting module is maximum by the first subimage block each in blood vessel dictionary and the second subimage block x inner product
Person is as first the first subimage block d chosenr0:
Wherein, k is the number of the first subimage block in blood vessel dictionary, r0It is the call number of dictionary, < x, di> it is x and di's
Inner product operation.
Sparse coefficient computational submodule calculates the first subimage block dr0Corresponding sparse coefficient αr0:αr0=< x, dr0>, it will select
In the first subimage block dr0Call number r0Set S, S=S ∪ r0 is added.
Angiosomes residual computations module 6 calculates residual error using maximum first subimage block of inner product and the second subimage block
Image, and the residual error in residual error weight calculation the second subimage block medium vessels region using the angiosomes.Angiosomes are residual
Poor computing module 6 includes that residual error just calculates module and residual weighted module.
Residual error just calculates the residual image R that module calculates the second subimage block medium vessels region:
R=x- < x, dr0>dr0；
Residual weighted module sums residual error R multiplied by the residual error Weight in the second subimage block medium vessels region, makees
For the final residual error in the second subimage block medium vessels region.
Jump module 7 sets the second subimage block for residual image when the norm of residual error is greater than threshold residual value, and
Sparse coefficient computing module 3 is jumped to, otherwise, jumps to the second subimage block reconstructed module 8.
Second subimage block reconstructed module 8 reconstructs the second subimage block using sparse coefficient.Second subimage block of reconstruct
Are as follows:
Wherein, S is the set for multiple sparse coefficients that the sparse coefficient computing module is repeatedly determined, dr0It is described dilute
Maximum first subimage block of the inner product that sparse coefficient computing module is determined each time, αr0It is dr0Corresponding sparse coefficient.
Eye fundus image reconstructed module 9 reconstructs eye fundus image using the second subimage block of each reconstruct, thus enhanced
Eye fundus image.Eye fundus image reconstructed module 9 is specifically used for:
Disjoint part of second subimage block of all reconstruct is merged, the eye fundus image completely enhanced.
The concrete operating principle of each module can refer to the correspondence step in aforementioned eye fundus image Enhancement Method in this system.
The above is merely preferred embodiments of the present invention, be not intended to limit the invention, it is all in spirit of the invention and
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within principle.
Claims (18)
1. a kind of eye fundus image Enhancement Method, which comprises the steps of:
Step A: learning picture construction blood vessel dictionary using eyeground, includes the first subgraph of setting quantity in the blood vessel dictionary
Block；
Step B: carrying out Frangi filtering to eye fundus image to be reinforced, and the image that Frangi is filtered be divided into it is several
The second overlapped subimage block；
Step C: utilization orientation filter carries out trend pass filtering to second subimage block, and is judged according to trend pass filtering result
The optical fundus blood vessel for including in second subimage block is thick blood vessel or thin and delicate blood vessel；
Step D: the angiosomes in second subimage block are determined, and according to the eyeground for including in second subimage block
The residual error weight and threshold residual value of the angiosomes in second subimage block is arranged in the type of blood vessel；
Step E: each first subimage block inner product in second subimage block and the blood vessel dictionary is determined in wherein
Maximum first subimage block of product, and calculate the corresponding sparse coefficient of maximum first subimage block of the inner product；
Step F: residual image is calculated using maximum first subimage block of the inner product and second subimage block, and is utilized
The residual error in the second subimage block medium vessels region described in the residual error weight calculation of the angiosomes；
Step G: when the norm of the residual error is greater than the threshold residual value, setting the second subimage block for residual image, and
Go to step E, and otherwise, go to step H；
Step H: second subimage block is reconstructed using the sparse coefficient；
Step I: the eye fundus image is reconstructed using the second subimage block of each reconstruct, thus the eye fundus image enhanced.
2. eye fundus image Enhancement Method as described in claim 1, which is characterized in that the step A includes:
Step A1: by eyeground study image segmentation at identical first subimage block of several sizes；First subgraph
The quantity of block is greater than the setting quantity；
Step A2: each first subimage block is subjected to inner product two-by-two；
Step A3: it chooses the first subimage block of the smallest setting quantity of inner product and constructs the blood vessel dictionary.
3. eye fundus image Enhancement Method as described in claim 1, which is characterized in that the step B includes:
Step B1: eye fundus image to be reinforced is set as I (x, y), the two-dimensional Gaussian function that scale is σ is G (x, y；σ), institute is utilized
It states two-dimensional Gaussian function to be smoothed the eye fundus image I (x, y) to be reinforced, obtains smoothed image Iσ(x, y):
Wherein, For convolution operation；
Step B2: at scale σ, smoothed image I is calculatedσHessian matrix H at midpoint (x, y) (x, y)σ(x, y):
Step B3: to the Hessian matrix Hσ(x, y) does Eigenvalues analysis, obtains eigenvalue λ1、λ2, | λ1|<|λ2|；Scale s
Under blood vessel feature are as follows:
Wherein,β and C is preset constant；
Step B4: under multiple dimensioned, v under each scale is taken0(s) maximum value is as the eye fundus image I's (x, y) to be reinforced
Frangi filter result v:
Wherein, sminAnd smaxIt is smallest dimension and out to out respectively；
Step B5: the Frangi filter result v is divided into several the second overlapped subimage blocks.
4. eye fundus image Enhancement Method as described in claim 1, which is characterized in that the step C includes:
Step C1: setting direction is respectively θ1=0, 8 anisotropic filters；
Step C2: assuming that direction is θiAnisotropic filter medium vessels region be Ω1, non-vascular region is Ω2, calculate the area Liang Ge
The respective energy in domainWith
Wherein v (x, y) is value of the Frangi filter result v at (x, y), N1It is Ω1Middle number of pixels, N2It is Ω2Middle number of pixels；
Step C3: it calculatesWithEnergy difference:
Step C4: maximum energy difference in above-mentioned 8 directions is determined:
Step C5: according to the EmaxVascular group is judged, if Emax>=T, the then eyeground for including in second subimage block
Image is thick blood vessel, is otherwise thin and delicate blood vessel.
5. eye fundus image Enhancement Method as claimed in claim 4, which is characterized in that the step D includes:
Step D1: using the angiosomes in anisotropic filter corresponding to maximum energy difference in 8 directions as blood vessel
Actual region Ω1, using the non-vascular region in anisotropic filter corresponding to maximum energy difference in 8 directions as
The actual region Ω of non-vascular2；
Step D2: for comprising eye fundus image be thick blood vessel the second subimage block, by its angiosomes Ω1Residual error weight
It is set as 1, threshold residual value TR=T1；For comprising eye fundus image be thin and delicate blood vessel the second subimage block, by its angiosomes
Ω1Residual error weight be set as 1/vmax, threshold residual value TR=T2, wherein vmaxIt is the second subimage block Frangi filter result
Maximum value.
6. eye fundus image Enhancement Method as described in claim 1, which is characterized in that the step E includes:
Step E1: the second subimage block vector is turned into x, i-th of first subimage blocks are d in the blood vessel dictionaryi；
Step E2: using the first subimage block each in the blood vessel dictionary and described second subimage block x inner product the maximum as choosing
In first the first subimage block dr0:
Wherein, k is the number of the first subimage block in the blood vessel dictionary, r0It is the call number of dictionary, < x, di> it is x and di's
Inner product operation；
Step E3: the first subimage block d is calculatedr0Corresponding sparse coefficient αr0:
αr0=< x, dr0>。
7. eye fundus image Enhancement Method as described in claim 1, which is characterized in that the step F includes:
Step F1: the residual image R in the second subimage block medium vessels region is calculated:
R=x- < x, dr0>dr0；
Step F2: the residual error R is summed multiplied by the residual error Weight in the second subimage block medium vessels region, as second
The final residual error in subimage block medium vessels region.
8. eye fundus image Enhancement Method as described in claim 1, which is characterized in that second subimage block of reconstruct are as follows:
Wherein, S is the set that multiple sparse coefficients that step E is determined are performed a plurality of times, dr0It is to execute step E each time to determine
Maximum first subimage block of inner product, αr0It is dr0Corresponding sparse coefficient.
9. eye fundus image Enhancement Method as described in claim 1, which is characterized in that the step I includes:
Disjoint part of second subimage block of all reconstruct is merged, the eye fundus image completely enhanced.
10. a kind of eye fundus image enhances system characterized by comprising
Blood vessel dictionary constructs module, learns picture construction blood vessel dictionary using eyeground, includes setting number in the blood vessel dictionary
First subimage block of amount；
Image filtering and division module carry out Frangi filtering to eye fundus image to be reinforced, and Frangi are filtered to obtain
Image be divided into several the second overlapped subimage blocks；
Vascular group judgment module, utilization orientation filter carry out trend pass filtering to second subimage block, and according to side
Judge that the optical fundus blood vessel for including in second subimage block is thick blood vessel or thin and delicate blood vessel to filter result；
Angiosomes and its residual error weight and threshold residual value determining module, determine the area vasculosa in second subimage block
Domain, and the area vasculosa in second subimage block is arranged according to the type for the optical fundus blood vessel for including in second subimage block
The residual error weight and threshold residual value in domain；
Sparse coefficient computing module, will be in each first subimage block in second subimage block and the blood vessel dictionary
Product, determines wherein maximum first subimage block of inner product, and it is corresponding dilute to calculate maximum first subimage block of the inner product
Sparse coefficient；
Angiosomes residual computations module utilizes maximum first subimage block of the inner product and the second subimage block meter
Residual image is calculated, and utilizes the residual error in the second subimage block medium vessels region described in the residual error weight calculation of the angiosomes；
Jump module sets the second subgraph for residual image when the norm of the residual error is greater than the threshold residual value
As block, and the sparse coefficient computing module is jumped to, otherwise, jumps to the second subimage block reconstructed module；
Second subimage block reconstructed module reconstructs second subimage block using the sparse coefficient；
Eye fundus image reconstructed module reconstructs the eye fundus image using the second subimage block of each reconstruct, to be enhanced
Eye fundus image.
11. eye fundus image as claimed in claim 10 enhances system, which is characterized in that the blood vessel dictionary constructs module packet
It includes:
Eyeground learns image division module, by eyeground study image segmentation at identical first subgraph of several sizes
Block；The quantity of first subimage block is greater than the setting quantity；
First subimage block inner product module, carries out inner product for each first subimage block two-by-two；
Blood vessel dictionary constructs submodule, chooses the first subimage block of the smallest setting quantity of inner product and constructs the blood vessel
Dictionary.
12. eye fundus image as claimed in claim 10 enhances system, which is characterized in that described image filtering and division module packet
It includes:
Smothing filtering module sets eye fundus image to be reinforced as I (x, y), and the two-dimensional Gaussian function that scale is σ is G (x, y；
σ), the eye fundus image I (x, y) to be reinforced is smoothed using the two-dimensional Gaussian function, obtains smoothed image
Iσ(x, y):
Wherein, For convolution operation；
Hessian matrix computing module calculates smoothed image I at scale σσHessian square at midpoint (x, y) (x, y)
Battle array Hσ(x, y):
Eigenvalues analysis module, to the Hessian matrix Hσ(x, y) does Eigenvalues analysis, obtains eigenvalue λ1、λ2, | λ1|<
|λ2|；Blood vessel feature under scale s are as follows:
Wherein,β and C is preset constant；
Frangi filter result generation module, takes v under each scale0(s) maximum value is as the eye fundus image I to be reinforced
The Frangi filter result v of (x, y):
Wherein, sminAnd smaxIt is smallest dimension and out to out respectively；
The Frangi filter result v is divided into several the second overlapped subgraphs by the second subgraph division module
Block.
13. eye fundus image as claimed in claim 10 enhances system, which is characterized in that the vascular group judgment module packet
It includes:
Anisotropic filter setup module, setting direction are respectively θ1=0, 8 anisotropic filters；
Energy computation module, hypothesis direction are θiAnisotropic filter medium vessels region be Ω1, non-vascular region is Ω2, meter
Calculate two respective energy in regionWith
Wherein v (x, y) is value of the Frangi filter result v at (x, y), N1It is Ω1Middle number of pixels, N2It is Ω2Middle number of pixels；
Energy difference computing module calculatesWithEnergy difference:
Ceiling capacity difference determining module determines maximum energy difference in above-mentioned 8 directions:
Vascular group judging submodule, according to the EmaxVascular group is judged, if Emax>=T, then second subgraph
The eye fundus image for including in block is thick blood vessel, is otherwise thin and delicate blood vessel.
14. eye fundus image as claimed in claim 13 enhances system, which is characterized in that the angiosomes and its residual error weight
Include: with threshold residual value determining module
Angiosomes determining module, by the blood vessel in anisotropic filter corresponding to maximum energy difference in 8 directions
Region is as the actual region Ω of blood vessel1, by the blood in anisotropic filter corresponding to maximum energy difference in 8 directions
Area under control domain is as the actual region Ω of blood vessel2；
Residual error weight and threshold residual value determining module, by comprising eye fundus image be thick blood vessel the second subimage block blood vessel
Region Ω1Residual error weight be set as 1, threshold residual value TR=T1；By comprising eye fundus image be thin and delicate blood vessel the second subimage block
Angiosomes Ω1Residual error weight be set as 1/vmax, threshold residual value TR=T2, wherein vmaxIt is second subimage block Frangi
The maximum value of filter result.
15. eye fundus image as claimed in claim 10 enhances system, which is characterized in that the sparse coefficient computing module packet
It includes:
The second subimage block vector is turned to x, i-th of first subgraphs in the blood vessel dictionary by image vector module
Block is di；
First subimage block selecting module, by the first subimage block each in the blood vessel dictionary and the second subimage block x
Inner product the maximum is as first the first subimage block d chosenr0:
Wherein, k is the number of the first subimage block in the blood vessel dictionary, r0It is the call number of dictionary, < x, di> it is x and di's
Inner product operation；
Sparse coefficient computational submodule calculates the first subimage block dr0Corresponding sparse coefficient αr0:
αr0=< x, dr0>。
16. eye fundus image as claimed in claim 10 enhances system, which is characterized in that the angiosomes residual computations module
Include:
Residual error just calculates module, calculates the residual image R in the second subimage block medium vessels region:
R=x- < x, dr0>dr0；
Residual weighted module sums the residual error R multiplied by the residual error Weight in the second subimage block medium vessels region,
Final residual error as the second subimage block medium vessels region.
17. eye fundus image as claimed in claim 10 enhances system, which is characterized in that second subimage block of reconstruct
Are as follows:
Wherein, S is the set for multiple sparse coefficients that the sparse coefficient computing module is repeatedly determined, dr0It is the sparse system
Maximum first subimage block of inner product that number computing module is determined each time, αr0It is dr0Corresponding sparse coefficient.
18. eye fundus image as claimed in claim 10 enhances system, which is characterized in that the eye fundus image reconstructed module is specific
For:
Disjoint part of second subimage block of all reconstruct is merged, the eye fundus image completely enhanced.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610822306.4A CN106504199B (en) | 2016-09-13 | 2016-09-13 | A kind of eye fundus image Enhancement Method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610822306.4A CN106504199B (en) | 2016-09-13 | 2016-09-13 | A kind of eye fundus image Enhancement Method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106504199A CN106504199A (en) | 2017-03-15 |
| CN106504199B true CN106504199B (en) | 2019-03-22 |
Family
ID=58290352
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610822306.4A Active CN106504199B (en) | 2016-09-13 | 2016-09-13 | A kind of eye fundus image Enhancement Method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106504199B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090899A (en) * | 2017-12-27 | 2018-05-29 | 重庆大学 | A kind of vessel extraction and denoising method |
| CN109816612A (en) * | 2019-02-18 | 2019-05-28 | 京东方科技集团股份有限公司 | Image enchancing method and device, computer readable storage medium |
| CN110473188B (en) * | 2019-08-08 | 2022-03-11 | 福州大学 | Fundus image blood vessel segmentation method based on Frangi enhancement and attention mechanism UNet |
| CN116630615A (en) * | 2023-04-24 | 2023-08-22 | 中国科学院空天信息创新研究院 | Infrared small target detection method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102842141A (en) * | 2012-07-03 | 2012-12-26 | 东南大学 | Rotary X-ray contrastographic picture iteration reconstruction method |
| CN103188988A (en) * | 2010-08-27 | 2013-07-03 | 索尼公司 | Image processing apparatus and method |
| CN103489203A (en) * | 2013-01-31 | 2014-01-01 | 清华大学 | Image coding method and system based on dictionary learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8068655B2 (en) * | 2007-10-02 | 2011-11-29 | Siemens Aktiengesellschaft | Method and system for vessel enhancement and artifact reduction in TOF MR angiography of brain |
-
2016
- 2016-09-13 CN CN201610822306.4A patent/CN106504199B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103188988A (en) * | 2010-08-27 | 2013-07-03 | 索尼公司 | Image processing apparatus and method |
| CN102842141A (en) * | 2012-07-03 | 2012-12-26 | 东南大学 | Rotary X-ray contrastographic picture iteration reconstruction method |
| CN103489203A (en) * | 2013-01-31 | 2014-01-01 | 清华大学 | Image coding method and system based on dictionary learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106504199A (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Soomro et al. | Impact of ICA-based image enhancement technique on retinal blood vessels segmentation | |
| Chakravarty et al. | RACE-net: a recurrent neural network for biomedical image segmentation | |
| CN110390650B (en) | OCT image denoising method based on dense connection and generation countermeasure network | |
| CN106504199B (en) | A kind of eye fundus image Enhancement Method and system | |
| WO2018049598A1 (en) | Ocular fundus image enhancement method and system | |
| CN110689083A (en) | Context pyramid fusion network and image segmentation method | |
| Tian et al. | Multi-path convolutional neural network in fundus segmentation of blood vessels | |
| Sinthanayothin | Image analysis for automatic diagnosis of diabetic retinopathy | |
| CN110544274B (en) | Multispectral-based fundus image registration method and system | |
| CN109658344A (en) | Image de-noising method, device, equipment and storage medium based on deep learning | |
| CN114612479A (en) | Medical image segmentation method based on global and local feature reconstruction network | |
| CN108986106A (en) | Retinal vessel automatic division method towards glaucoma clinical diagnosis | |
| CN109658393B (en) | Fundus image splicing method and system | |
| CN108764342B (en) | Semantic segmentation method for optic discs and optic cups in fundus image | |
| WO2017036231A1 (en) | Method and system for acquiring retina structure from optical coherence tomographic image | |
| CN109528196B (en) | Hepatic vein pressure gradient non-invasive evaluation method | |
| DE102007018077A1 (en) | Three-dimensional (3D) modeling of coronary arteries | |
| CN113205538A (en) | Blood vessel image segmentation method and device based on CRDNet | |
| Yin et al. | Fully-automatic segmentation of kidneys in clinical ultrasound images using a boundary distance regression network | |
| CN117495876B (en) | Coronary artery image segmentation method and system based on deep learning | |
| CN103295195B (en) | The enhanced method of vascular and its system of soft image | |
| CN113781403A (en) | Chest CT image processing method and device | |
| CN109919098A (en) | The recognition methods of target object and device | |
| CN110689080A (en) | Planar atlas construction method of blood vessel structure image | |
| Kop et al. | Kidney segmentation from ultrasound images using gradient vector force |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Wu Junhao Inventor after: Yang Hui Inventor after: Pei Jihong Inventor before: Yang Hui Inventor before: Zhang Zhengrui Inventor before: Pei Jihong |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |