CN107248161A - Retinal vessel extracting method is supervised in a kind of having for multiple features fusion - Google Patents

Retinal vessel extracting method is supervised in a kind of having for multiple features fusion Download PDF

Info

Publication number
CN107248161A
CN107248161A CN201710327857.8A CN201710327857A CN107248161A CN 107248161 A CN107248161 A CN 107248161A CN 201710327857 A CN201710327857 A CN 201710327857A CN 107248161 A CN107248161 A CN 107248161A
Authority
CN
China
Prior art keywords
mrow
pixel
image
feature
msub
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201710327857.8A
Other languages
Chinese (zh)
Inventor
梁礼明
刘博文
黄朝林
吴健
陈明理
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangxi University of Science and Technology
Original Assignee
Jiangxi University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangxi University of Science and Technology filed Critical Jiangxi University of Science and Technology
Priority to CN201710327857.8A priority Critical patent/CN107248161A/en
Publication of CN107248161A publication Critical patent/CN107248161A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G06T7/41—Analysis of texture based on statistical description of texture
    • G06T7/44—Analysis of texture based on statistical description of texture using image operators, e.g. filters, edge density metrics or local histograms
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20024—Filtering details
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30041—Eye; Retina; Ophthalmic
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Processing (AREA)

Abstract

The present invention relates to retinal vessel cutting techniques, retinal vessel extracting method is supervised in having for particularly a kind of multiple features fusion.The present invention includes following four step：Step 1, retinal vascular images are pre-processed；Step 2, retinal vascular images feature extraction；Step 3, random forest grader is trained；Step 4, retinal vascular images are post-processed.Experimental verification of the present invention on DRIVE and STARE eye fundus image databases, susceptibility is respectively 0.8354 and 0.8452, and accuracy rate is respectively 94.82% and 95.34%, and overall objective is better than existing retinal vascular images dividing method.Overcome simultaneously at blood vessels adjacent, at intersecting blood vessels and at capilary other methods deficiency, the blood vessel structure being partitioned into is closer to goldstandard and blood vessel full-size(d) value.

Description

Retinal vessel extracting method is supervised in a kind of having for multiple features fusion
Technical field
The present invention relates to retinal vessel cutting techniques, particularly a kind of multiple features fusion have supervision retinal vessel carry Take method.
Background technology
Eye ground blood vessel network has a complicated multi-level institutional framework, many ophthalmology and angiocardiopathy Lesion can be reflected directly in the change of retinal vessel network structure.Although blood vessel network has certain difference with background, But the brightness of blood vessel is gradually changed with extending in for blood vessel, the particularly tip of blood vessel and the contrast of background is relatively low, it is difficult to Complete parttion.Therefore, retinal vessel cutting techniques are always the focus and difficult point of eye fundus image analysis field.But retina The information such as blood vessel number, branch, angle, width can be as the diagnosis basis with retinal vessel relevant disease, and this is utilization Digital image processing techniques extract retinal vasculature, qualitative and quantitative analysis and judge that conditions of patients and research pathology are provided Basis.But oculist substantially carries out quantitative analysis to PVR using manual mode at present, subjectivity is strong, Accuracy and uniformity can not be ensured.
Current Segmentation Method of Retinal Blood Vessels is a variety of, such as based on matched filtering, blood vessel detection, morphology, deformation model and Blood vessel segmentation method of machine learning etc..Existing Segmentation Method of Retinal Blood Vessels be primarily present the capilary that is partitioned into easily from Dissipate, continuity is poor, susceptibility is relatively low, accuracy rate is not high and can not efficiently extract the defects such as vessel information.
The content of the invention
The purpose of the present invention is not enough the having there is provided a kind of multiple features fusion for existing Segmentation Method of Retinal Blood Vessels Retinal vessel extracting method is supervised, the capilary easily discrete, susceptibility that existing model is partitioned into is solved relatively low, it is impossible to maximum The problems such as extracting vessel information to limit.
Retinal vessel extracting method, including following four step are supervised in a kind of having for multiple features fusion：
Step 1, retinal vascular images are pre-processed：Pre- locate is used as to obtain the green channel retinal images of eye fundus image Manage object；First, the circular kernel etching mask image by r of radius carries out edge expansion to eye fundus image；Secondly, using double Side filters noise reduction, and the wherein a diameter of d of neighborhood of pixels, color space is d × 2, and coordinate space is d/2；Finally, by filtered figure As gray scale stretching to [0-255] so that blood vessel enhancing image background is more uniform, to ensure the characteristic area of blood vessel and background area Indexing is maximum；
Step 2, retinal vascular images feature extraction：Introduce linear character, textural characteristics, moment characteristics, Variance feature, ash Feature is spent, and using the image processing techniques such as many sizes, morphology, gaussian filtering measurement retinal vessel and the difference of background It is different, 34 features are extracted to each pixel in image, here shown as { Fi, i=1,2 ..., 34；The acquisition of each feature Method is specific as follows：
1. linear character：I is obtained to pretreatment image reversion firstrev, to reverse image IrevMake roof fall conversion to be gone Except optic disk and the image I of macula luteatop；Then to image ItopIn each pixel, calculating size be W × W pixel regions Average gray valueIts calculation formula is
Wherein I (x, y) is the pixel value at window midpoint (x, y)；
Each pixel starts from horizontal direction, with 15 ° for interval, has 12 different directions by the straight of the point Line, is designated asThe average gray value of this 12 curves is calculated, one of gray value maximum is defined as " winning Straight line ", calculates the gray value of triumph curveI.e.
It is the image average in W × W to subtract window againThe as blood vessel feature of the pointI.e.Its Middle W=(L+1)/2, by changing detection of straight lines L length, can obtain the linear character under different scale；Detect line length L={ 3,5,7,9,11 } is taken respectively, and 5 linear characters, i.e. F are extracted altogether1,F2,F3,F4,F5；
2. textural characteristics：Introduce local binary patterns (Local Binary Pattern, LBP) textural characteristics and describe view The local feature of film image；First using the height of three kinds of different size factor σ={ 2/6,4/6,6/6 } before texture feature extraction This template is filtered to blood vessel enhancing image, then uses pixel size to be detected for the sliding window of 3 × 3 and 5 × 5； If neighborhood point gray value subtracts central point gray scale not less than 2, this value is 1, otherwise is 0；Thus 8 binary systems are produced Number, takes its minimum value as characteristic value；The feature extraction of so each pixel has 3 × 2 kinds of combinations, adds up to 6 features F6,F7,F8,F9,F10,F11, and with size constancy and rotational invariance；
Here each pixel (x in retinal imagesc,yc) local binary patterns (LBP) operator definitions be
Wherein (xc,yc) centered on pixel coordinate, p-th of pixel of neighborhood of pixels, i centered on ppFor neighborhood territory pixel gray scale Value, icCentered on grey scale pixel value, I (ip-ic) it is indicating value function, it is defined as
3. moment characteristics：Vertical moment characteristics, three kinds of rectangle frames of horizontal moment characteristics and diagonal moment characteristics in Haar features are chosen to go Detect retinal vasculature；
Gaussian template first using three kinds of different size factor σ before moment characteristics are extracted is filtered to blood vessel enhancing image Ripple, while the sliding window size W={ 5,7,9 } of Moment Feature Extraction, therefore the feature extraction of each pixel has 3 × 3 kinds Combination, amounts to 9 feature F12,F13,…,F20；Vertical moment characteristics and horizontal moment characteristics with size constancy can so be obtained And the diagonal moment characteristics with size constancy and rotational invariance；Here moment characteristics h calculation formula is
The moment characteristics h of i.e. each pixel subtracts black region for white portion all pixels gray value sum in Haar operators All pixels gray value sum；
Equally, the feature extraction of retinal images is just completed after sliding window travels through complete image area；
4. Variance feature：Blood vessel is strengthened using the Gaussian template of three kinds of different size factor σ={ 2/6,4/6,6/6 } and schemed As being filtered, and the pixel local variance that sliding window size size is W={ 3,5,7 } is calculated respectively, make each pixel Obtain 9 feature F with size constancy21,F22,…,F29；
Here for the rectangular area that Pixel Dimensions are W × W, retina enhancing image IGauLocal variance var (Ix,y) Calculating is characterized as：
WhereinIt is p-th of pixel value in W area for dimensions length,It is being averaged in W area to refer to dimensions length Pixel value, calculation formula is：
5. gray feature：Blood vessel enhancing image is inverted and roof fall conversion, obtain feature F30；Then using size because Sub- σ={ 2/6,6/6 } Gaussian template roof fall is converted after image filtering and full Threshold segmentation, if then gray value is not or not blood vessel Become, otherwise gray value is changed into 0, gray feature F is obtained with this31,F32；Using size factor σ={ 6/6 } Gaussian template to reversion Image filtering and roof fall conversion, obtain feature F33；To feature F33Spy is obtained using size factor σ={ 9/6 } Gaussian template filtering Levy F34；
Afterwards, to the retinal vessel feature F of extraction1,F2,…,F34It is normalized, its numerical value is transformed into interval [0,1]；
Step 3, random forest grader is trained：1. training set is chosen：Unit is chosen by least random of pixel, at random Choose a number of pixel in retinal images and be used as sample set；2. sample random tree：Bootstrap is used to sample set Sampling obtains sample set θk, it is used as training every categorised decision tree T (X, θk) training set, set random forest scale as K； 3. decision tree is generated：Every random tree carries out sample attribute division according to information gain principle；Decision tree at utmost grows to the greatest extent, Reached to a certain degree if sample number is deep less than certain scale or tree, split vertexes sample is not also pure, then stop growing； Decision tree not beta pruning after the completion of generation；4. integrated classifier：K decision tree of generation is combined with relative majority ballot criterion, is obtained Thus the most class of number of votes obtained obtains the random forest grader under supervision as final output；
Step 4, retinal vascular images are post-processed：Because the textural characteristics and Haar features of focus and artifact are similar to blood Pipe feature, the extraction result of random forests algorithm usually contains a large amount of focuses and artifact；Therefore, blood-vessel image post processing is introduced, While vessel extraction susceptibility is ensured, reduction specificity improves blood vessel segmentation precision；Specific practice is as follows：
1. according to the gray value threshold value dividing background region of setting and angiosomes, exclude by macula lutea and optic disk center gray scale Blood-vessel image caused by value is too high is split by mistake, further enhances vessel information；
2. the evaluation index based on connected domain area Area and width W, high H, passes through the different geometric shape of operation definition Operator is learned to distinguish blood vessel, focus and artifact, to obtain the retinal vessel segmentation figure picture of high-accuracy.
Experimental verification of the present invention on DRIVE and STARE eye fundus image databases, susceptibility is respectively 0.8354 He 0.8452, accuracy rate is respectively 94.82% and 95.34%, and overall objective is better than existing retinal vascular images dividing method. Overcome simultaneously at blood vessels adjacent, at intersecting blood vessels and at capilary other methods deficiency, make the blood vessel structure being partitioned into It is closer to goldstandard and blood vessel full-size(d) value.
Brief description of the drawings
Fig. 1 is schematic flow sheet of the invention.
Fig. 2 strengthens image schematic diagram for the blood vessel obtained after pretreatment in embodiment；Wherein：(a) retina original graph Picture；(b) retina green channel images；(c) retinal vessel enhancing image；(d) retina original image local detail；(d) Retinal vessel strengthens image local details.
Fig. 3 is healthy retina image and lesion retinal images vessel extraction effect diagram in the embodiment of the present invention； Wherein：(a) healthy retina image；(b) healthy retina feature extraction result；(c) healthy retina post processing result；(d) Lesion retinal images；(e) lesion retinal feature extracts result；(f) lesion retina post processing result.
Embodiment
With reference to specific embodiment, the present invention is expanded on further.
Description of test：Embodiment data involved by the application of the present invention come from STARE databases.STARE databases are each There are 10 width to have lesion and do not have the eye fundus image of lesion, picture size is 605 × 700 pixels.Secondly during training Random Forest model Every width retinal images randomly select 7000 pixels of 3000 pixels of blood vessel sample and non-vascular (background) sample, altogether 200000 pixels of meter are used as training sample
The present embodiment includes four steps：It is retinal vascular images pretreatment, retinal vascular images feature extraction, random Forest classified device is trained and retinal vascular images post processing, as shown in Figure 1.
It is described in detail below：
1st, retinal vascular images are pre-processed
The present embodiment is used as pretreatment object using the green channel retinal images for obtaining eye fundus image.First using radius as 3 circular kernel etching mask image carries out edge expansion to eye fundus image.Secondly, it is guarantee blood-vessel image edge pixel matter Amount, using bilateral filtering noise reduction, wherein neighborhood of pixels a diameter of 25, color space is 25 × 2, and coordinate space is 25/2.Finally, Filtered gradation of image is stretched to [0-255] so that blood vessel enhancing image background is more uniform, to ensure blood vessel and background The characteristic area indexing in region is maximum, as shown in Fig. 2.
2nd, retinal vascular images feature extraction.Introduce linear character, textural characteristics, moment characteristics, Variance feature, gray scale special Levy, and using the image processing techniques such as many sizes, morphology, gaussian filtering measurement retinal vessel and the difference of background.It is right Each pixel of image extracts 34 features, here shown as { Fi, i=1,2 ..., 34.The acquisition methods of each feature It is specific as follows：
1. linear character：Enhanced image is inverted first and roof fall conversion, then using the side of linearity test Method obtains linear character.Present invention detection line length takes l={ 3,5,7,9,11 } respectively, and 5 linear characters, i.e. F are extracted altogether1, F2,F3,F4,F5.Defining the corresponding average gray values of detection line l for being W along dimensions length is：
With 15 ° for the anglec of rotation, 12 directions of calculatingIt is triumph detection line to take its maximum, and subtracts window for W Image average in × WThe as blood vessel feature of the point.When complete image area of sliding window traversal, detection line is just completed Size is the feature extraction under l.
2. textural characteristics：Present invention introduces the local feature that local binary patterns textural characteristics describe retinal images. Blood vessel is strengthened using the Gaussian template of three kinds of different size factor σ={ 2/6,4/6,6/6 } first before texture feature extraction and schemed As being filtered, pixel size is then used to be detected for the sliding window of 3 × 3 and 5 × 5, if neighborhood point gray value is subtracted Central point gray scale is not less than 2, then this value is 1, otherwise is 0.Thus 8 bits are produced, its minimum value are taken as spy Value indicative.The feature extraction of so each pixel has 3 × 2 kinds of combinations and amounts to 6 feature F6,F7,.F8,F9,F10,F11, and tool There are size constancy and rotational invariance.
3. moment characteristics：Choose in Haar vertical moment characteristics, horizontal moment characteristics and go detection to regard three kinds of rectangle frames of corner characteristics Retinal vasculature structure.First using the Gaussian template of three kinds of different size factor σ={ 2/6,4/6,6/6 } before moment characteristics are extracted Blood vessel enhancing image is filtered, while the sliding window size W={ 5,7,9 } of Moment Feature Extraction, therefore each pixel Feature extraction have 3 × 3 kinds of combinations, amount to 9 feature F12,F13,F14,F15,F16,F17,F18,F19,F20.It can so obtain Vertical moment characteristics, horizontal moment characteristics with size constancy and special to angular moment with size constancy and rotational invariance Levy.Equally, the feature extraction of retinal images is just completed after sliding window travels through complete image area.
4. Variance feature：Blood vessel is strengthened using the Gaussian template of three kinds of different size factor σ={ 2/6,4/6,6/6 } and schemed As being filtered, and the pixel local variance that sliding window size size is W={ 3,5,7 } is calculated respectively, make each pixel Obtain 9 feature F with size constancy21,F22,F23,F24,F25,F26,F27,F28,F29。
5. gray feature：Blood vessel enhancing image is inverted and roof fall conversion, obtain feature F30；Then using size because Sub- σ={ 2/6,6/6 } Gaussian template roof fall is converted after image filtering and full Threshold segmentation, if then gray value is not or not blood vessel Become, otherwise gray value is changed into 0, gray feature F is obtained with this31,F32；Using size factor σ={ 6/6 } Gaussian template to reversion Image filtering and roof fall conversion, obtain feature F33；To feature F33Spy is obtained using size factor σ={ 9/6 } Gaussian template filtering Levy F34.Feature F31,F32,F33,F34Primarily to the large scale blood vessel of description retina, while suppressing the production of noise characteristic It is raw, the noise immunity for the blood vessel grader that the raising later stage is trained based on random forests algorithm.
In addition, to the retinal vessel feature F of extraction1,F2,…,F34It is normalized, its numerical value is transformed into interval [0,1]。
3rd, random forest grader is trained.1. training set is chosen：Unit is chosen by least random of pixel, is randomly selected A number of pixel is used as sample set in retinal images.2. sample random tree：Sample set is sampled using bootstrap Obtain sample set θk, it is used as training every categorised decision tree T (X, θk) training set, set random forest scale as 100.③ Generate decision tree：Every random tree carries out sample attribute division according to information gain principle.Decision tree at utmost grows to the greatest extent, if Sample number is less than 50 or tree has reached 20 deeply, treats that split vertexes sample is not yet pure, then stops growing.Decision-making after the completion of generation Set not beta pruning.4. integrated classifier：100 decision trees of generation combine with relative majority ballot criterion, obtain poll most Thus class obtains the random forest grader under supervision as final output.
4th, retinal vascular images are post-processed.Because the textural characteristics and Haar features of focus and artifact are special similar to blood vessel Levy, the extraction result of random forests algorithm usually contains a large amount of focuses and artifact.Therefore, introducing blood-vessel image post processing, protecting While demonstrate,proving vessel extraction susceptibility, reduction specificity improves blood vessel segmentation precision.Specific practice is as follows：
1. with 190 for gray value threshold value dividing background region and angiosomes, exclude by macula lutea and optic disk center gray value Blood-vessel image caused by too high is split by mistake, further enhances vessel information.
2. the evaluation index based on connected domain area Area and width W, high H information, passes through the different geometry of operation definition Morphological operator improves the segmentation precision of retinal vessel, as shown in Fig. 3 to remove artifact and focus.The present embodiment geometry is calculated Sub-definite is as follows：
(1)0.4<W/H<2.5；
(2)W×H<3.5Area；
(3)Area<30。

Claims (1)

1. a kind of multiple features fusion has supervision retinal vessel extracting method, it is characterized in that, including following four step：
Step 1, retinal vascular images are pre-processed：Using obtain eye fundus image green channel retinal images be used as pretreatment pair As；First, the circular kernel etching mask image by r of radius carries out edge expansion to eye fundus image；Secondly, using bilateral filter The a diameter of d of ripple noise reduction, wherein neighborhood of pixels, color space is d × 2, and coordinate space is d/2；Finally, by filtered image ash Degree is stretched to [0-255] so that blood vessel enhancing image background is more uniform, is indexed with the characteristic area for ensureing blood vessel and background area It is maximum；
Step 2, retinal vascular images feature extraction：Introduce linear character, textural characteristics, moment characteristics, Variance feature, gray scale special Levy, and using the image processing techniques such as many sizes, morphology, gaussian filtering measurement retinal vessel and the difference of background, it is right Each pixel extracts 34 features in image, here shown as { Fi, i=1,2 ..., 34；The acquisition methods tool of each feature Body is as follows：
1. linear character：I is obtained to pretreatment image reversion firstrev, to reverse image IrevMake roof fall conversion to be eliminated The image I of optic disk and macula luteatop；Then to image ItopIn each pixel, it is the flat of W × W pixel regions to calculate size Equal gray valueIts calculation formula is
<mrow> <msubsup> <mi>I</mi> <mrow> <mi>a</mi> <mi>v</mi> <mi>g</mi> </mrow> <mi>W</mi> </msubsup> <mo>=</mo> <mfrac> <mn>1</mn> <msup> <mi>W</mi> <mn>2</mn> </msup> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>x</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>W</mi> </munderover> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>y</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>W</mi> </munderover> <mi>I</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> </mrow>
Wherein I (x, y) is the pixel value at window midpoint (x, y)；
Each pixel starts from horizontal direction, with 15 ° for interval, has the straight line that 12 different directions pass through the point, note For(i=1,2 ... 12)；The average gray value of this 12 curves is calculated, one of gray value maximum is defined as " winning straight Line ", calculates the gray value of triumph curveI.e.
<mrow> <msubsup> <mi>I</mi> <mrow> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> <mi>L</mi> </msubsup> <mo>=</mo> <mi>max</mi> <mo>{</mo> <msubsup> <mi>I</mi> <mi>i</mi> <mi>L</mi> </msubsup> <mo>}</mo> <mo>,</mo> <mrow> <mo>(</mo> <mi>i</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mn>2</mn> <mo>,</mo> <mo>...</mo> <mn>12</mn> <mo>)</mo> </mrow> </mrow>
It is the image average in W × W to subtract window againThe as blood vessel feature of the pointI.e.Wherein W =(L+1)/2, by changing detection of straight lines L length, can obtain the linear character under different scale；Detect line length difference L={ 3,5,7,9,11 } is taken, 5 linear characters, i.e. F are extracted altogether1,F2,F3,F4,F5；
2. textural characteristics：Introduce the local feature that local binary patterns textural characteristics describe retinal images；It is special extracting texture Gaussian template before levying first using three kinds of different size factor σ={ 2/6,4/6,6/6 } is filtered to blood vessel enhancing image, Then pixel size is used to be detected for the sliding window of 3 × 3 and 5 × 5；If neighborhood point gray value subtracts central point gray scale not Less than 2, then this value is 1, otherwise is 0；Thus 8 bits are produced, its minimum value are taken as characteristic value；It is so each The feature extraction of pixel has 3 × 2 kinds of combinations, adds up to 6 feature F6,F7,F8,F9,F10,F11, and with size constancy And rotational invariance；
Here each pixel (x in retinal imagesc,yc) local binary pattern operator be defined as
<mrow> <mi>L</mi> <mi>B</mi> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>c</mi> </msub> <mo>,</mo> <msub> <mi>y</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mo>-</mo> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>p</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>p</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <msup> <mn>2</mn> <mi>p</mi> </msup> <mi>I</mi> <mrow> <mo>(</mo> <msub> <mi>i</mi> <mi>p</mi> </msub> <mo>-</mo> <msub> <mi>i</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> </mrow>
Wherein (xc,yc) centered on pixel coordinate, p-th of pixel of neighborhood of pixels, i centered on ppFor neighborhood territory pixel gray value, ic Centered on grey scale pixel value, I (ip-ic) it is indicating value function, it is defined as
<mrow> <mi>I</mi> <mrow> <mo>(</mo> <msub> <mi>i</mi> <mi>p</mi> </msub> <mo>-</mo> <msub> <mi>i</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <mi>i</mi> <mi>f</mi> <mrow> <mo>(</mo> <msub> <mi>i</mi> <mi>p</mi> </msub> <mo>-</mo> <msub> <mi>i</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;GreaterEqual;</mo> <mn>2</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>e</mi> <mi>l</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>
3. moment characteristics：Choose vertical moment characteristics, three kinds of rectangle frames of horizontal moment characteristics and diagonal moment characteristics in Haar features and go detection Retinal vasculature；
Gaussian template first using three kinds of different size factor σ before moment characteristics are extracted is filtered to blood vessel enhancing image, together When Moment Feature Extraction sliding window size W={ 5,7,9 }, therefore the feature extraction of each pixel has 3 × 3 kinds of combinations, Amount to 9 feature F12,F13,…,F20；Vertical moment characteristics and horizontal moment characteristics and tool with size constancy can so be obtained There are size constancy and the diagonal moment characteristics of rotational invariance；Here moment characteristics h calculation formula is
<mrow> <mi>h</mi> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>W</mi> <mi>h</mi> <mi>i</mi> <mi>t</mi> <mi>e</mi> </mrow> </munder> <mi>I</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>B</mi> <mi>l</mi> <mi>a</mi> <mi>c</mi> <mi>k</mi> </mrow> </munder> <mi>I</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> </mrow>
The moment characteristics h of i.e. each pixel subtracts black region for white portion all pixels gray value sum in Haar operators to be owned Grey scale pixel value sum；
Equally, the feature extraction of retinal images is just completed after sliding window travels through complete image area；
4. Variance feature：Blood vessel enhancing image is entered using the Gaussian template of three kinds of different size factor σ={ 2/6,4/6,6/6 } Row filtering, and the pixel local variance that sliding window size size is W={ 3,5,7 } is calculated respectively, obtain each pixel 9 feature F with size constancy21,F22,…,F29；
Here for the rectangular area that Pixel Dimensions are W × W, retina enhancing image IGauLocal variance var (Ix,y) calculate It is characterized as：
<mrow> <mi>v</mi> <mi>a</mi> <mi>r</mi> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <msup> <mi>W</mi> <mn>2</mn> </msup> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>p</mi> <mo>=</mo> <mn>1</mn> </mrow> <msup> <mi>W</mi> <mn>2</mn> </msup> </munderover> <mrow> <mo>(</mo> <msubsup> <mi>&amp;eta;</mi> <mi>p</mi> <mi>W</mi> </msubsup> <mo>-</mo> <msubsup> <mover> <mi>I</mi> <mo>&amp;OverBar;</mo> </mover> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mi>W</mi> </msubsup> <mo>)</mo> </mrow> </mrow>
WhereinIt is p-th of pixel value in W area for dimensions length,It is the mean pixel in W area to refer to dimensions length It is worth, calculation formula is：
<mrow> <msubsup> <mover> <mi>I</mi> <mo>&amp;OverBar;</mo> </mover> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mi>W</mi> </msubsup> <mo>=</mo> <mfrac> <mn>1</mn> <msup> <mi>W</mi> <mn>2</mn> </msup> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>p</mi> <mo>=</mo> <mn>1</mn> </mrow> <msup> <mi>W</mi> <mn>2</mn> </msup> </munderover> <msubsup> <mi>&amp;eta;</mi> <mi>p</mi> <mi>W</mi> </msubsup> </mrow>
5. gray feature：Blood vessel enhancing image is inverted and roof fall conversion, obtain feature F30；Then size factor σ is used ={ 2/6,6/6 } Gaussian template is to the image filtering and full Threshold segmentation after roof fall conversion, if then gray value is constant for blood vessel, instead Gray value be changed into 0, gray feature F is obtained with this31,F32；Reverse image is filtered using size factor σ={ 6/6 } Gaussian template Involve roof fall conversion, obtain feature F33；To feature F33Feature F is obtained using size factor σ={ 9/6 } Gaussian template filtering34；
Afterwards, to the retinal vessel feature F of extraction1,F2,…,F34Be normalized, its numerical value be transformed into it is interval [0, 1]；
Step 3, random forest grader is trained：1. training set is chosen：Unit is chosen by least random of pixel, is randomly selected A number of pixel is used as sample set in retinal images；2. sample random tree：Sample set is sampled using bootstrap Obtain sample set θk, it is used as training every categorised decision tree T (X, θk) training set, set random forest scale as K；3. give birth to Into decision tree：Every random tree carries out sample attribute division according to information gain principle；Decision tree at utmost grows to the greatest extent, if sample This number is deep less than certain scale or tree have been reached to a certain degree, and split vertexes sample is not also pure, then stop growing；Generation After the completion of decision tree not beta pruning；4. integrated classifier：K decision tree of generation is combined with relative majority ballot criterion, obtains ticket Thus the most class of number obtains the random forest grader under supervision as final output；
Step 4, retinal vascular images are post-processed：Because the textural characteristics and Haar features of focus and artifact are special similar to blood vessel Levy, the extraction result of random forests algorithm usually contains a large amount of focuses and artifact；Therefore, introducing blood-vessel image post processing, protecting While demonstrate,proving vessel extraction susceptibility, reduction specificity improves blood vessel segmentation precision；Specific practice is as follows：
1. according to the gray value threshold value dividing background region of setting and angiosomes, exclude by macula lutea and optic disk center gray value mistake Blood-vessel image caused by height is split by mistake, further enhances vessel information；
2. the evaluation index based on connected domain area Area and width W, high H, is calculated by the different geometric state of operation definition Son is to distinguish blood vessel, focus and artifact, to obtain the retinal vessel segmentation figure picture of high-accuracy.
CN201710327857.8A 2017-05-11 2017-05-11 Retinal vessel extracting method is supervised in a kind of having for multiple features fusion Pending CN107248161A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710327857.8A CN107248161A (en) 2017-05-11 2017-05-11 Retinal vessel extracting method is supervised in a kind of having for multiple features fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710327857.8A CN107248161A (en) 2017-05-11 2017-05-11 Retinal vessel extracting method is supervised in a kind of having for multiple features fusion

Publications (1)

Publication Number Publication Date
CN107248161A true CN107248161A (en) 2017-10-13

Family

ID=60017171

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710327857.8A Pending CN107248161A (en) 2017-05-11 2017-05-11 Retinal vessel extracting method is supervised in a kind of having for multiple features fusion

Country Status (1)

Country Link
CN (1) CN107248161A (en)

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108230306A (en) * 2017-12-28 2018-06-29 中山大学 Eyeground color picture blood vessel and arteriovenous recognition methods
CN108230301A (en) * 2017-12-12 2018-06-29 哈尔滨理工大学 A kind of spine CT image automatic positioning dividing method based on active contour model
CN108416344A (en) * 2017-12-28 2018-08-17 中山大学中山眼科中心 Eyeground color picture optic disk and macula lutea positioning identifying method
CN108921133A (en) * 2018-07-27 2018-11-30 福州大学 Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features
CN109447947A (en) * 2018-09-28 2019-03-08 上海理工大学 A kind of central fovea localization method based on lesion colour retinal fundus images
CN109448006A (en) * 2018-11-01 2019-03-08 江西理工大学 A kind of U-shaped intensive connection Segmentation Method of Retinal Blood Vessels of attention mechanism
CN109448038A (en) * 2018-11-06 2019-03-08 哈尔滨工程大学 Sediment sonar image feature extracting method based on DRLBP and random forest
CN109471853A (en) * 2018-09-18 2019-03-15 平安科技（深圳）有限公司 Data noise reduction, device, computer equipment and storage medium
CN109473148A (en) * 2018-10-26 2019-03-15 武汉工程大学 A kind of ion concentration prediction technique, device and computer storage medium
CN109615614A (en) * 2018-11-26 2019-04-12 北京工业大学 The extracting method and electronic equipment of eye fundus image medium vessels based on multi-feature fusion
CN110189327A (en) * 2019-04-15 2019-08-30 浙江工业大学 Eye ground blood vessel segmentation method based on structuring random forest encoder
CN110288588A (en) * 2019-07-01 2019-09-27 齐鲁工业大学 Retinal images blood vessel segmentation method and system based on gray variance and standard deviation
CN110472600A (en) * 2019-08-20 2019-11-19 北京百度网讯科技有限公司 The identification of eyeground figure and its training method, device, equipment and storage medium
CN110874597A (en) * 2018-08-31 2020-03-10 福州依影健康科技有限公司 Blood vessel feature extraction method, device and system for fundus image and storage medium
CN113011340A (en) * 2021-03-22 2021-06-22 华南理工大学 Cardiovascular surgery index risk classification method and system based on retina image
CN113781514A (en) * 2021-09-14 2021-12-10 西南交通大学 Retina blood vessel segmentation method with multi-low-level feature fusion
CN113989246A (en) * 2021-10-29 2022-01-28 南开大学 Transparent blood vessel image segmentation method based on blood flow characteristics
CN113989170A (en) * 2021-10-29 2022-01-28 南开大学 Transparent blood vessel type identification method based on blood flow characteristics
CN114913099A (en) * 2021-12-28 2022-08-16 天翼数字生活科技有限公司 Method and system for processing video file
CN116091451A (en) * 2023-01-09 2023-05-09 中国科学院苏州生物医学工程技术研究所 Retinal pigment epithelial cell image segmentation method and system
CN117689552A (en) * 2024-02-02 2024-03-12 科普云医疗软件(深圳)有限公司 Coronary angiography enhancement method for intracardiac interventional therapy

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090206838A1 (en) * 2007-12-13 2009-08-20 Isis Innovation Ltd. Systems and Methods for Correction of Inhomogeneities in Magnetic Resonance Images
CN106248559A (en) * 2016-07-14 2016-12-21 中国计量大学 A kind of leukocyte five sorting technique based on degree of depth study

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090206838A1 (en) * 2007-12-13 2009-08-20 Isis Innovation Ltd. Systems and Methods for Correction of Inhomogeneities in Magnetic Resonance Images
CN106248559A (en) * 2016-07-14 2016-12-21 中国计量大学 A kind of leukocyte five sorting technique based on degree of depth study

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
关亚勇: "医学图像的多特征融合和识别研究与应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
康朝红: "基于支持向量机的X线图像分割技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
朱承璋: "基于多特征融合和随机森林的视网膜血管分割", 《计算机辅助设计与图形学学报》 *
高向军: "基于多尺度线性检测的视网膜血管分割", 《科学技术与工程》 *

Cited By (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108230301A (en) * 2017-12-12 2018-06-29 哈尔滨理工大学 A kind of spine CT image automatic positioning dividing method based on active contour model
CN108416344B (en) * 2017-12-28 2021-09-21 中山大学中山眼科中心 Method for locating and identifying eyeground color optic disk and yellow spot
CN108416344A (en) * 2017-12-28 2018-08-17 中山大学中山眼科中心 Eyeground color picture optic disk and macula lutea positioning identifying method
CN108230306A (en) * 2017-12-28 2018-06-29 中山大学 Eyeground color picture blood vessel and arteriovenous recognition methods
CN108921133A (en) * 2018-07-27 2018-11-30 福州大学 Multimode unsupervised learning retinal vessel segmenting system based on Fusion Features
CN110874597A (en) * 2018-08-31 2020-03-10 福州依影健康科技有限公司 Blood vessel feature extraction method, device and system for fundus image and storage medium
CN110874597B (en) * 2018-08-31 2023-03-24 福州依影健康科技有限公司 Blood vessel feature extraction method, device and system for fundus image and storage medium
US11321287B2 (en) 2018-09-18 2022-05-03 Ping An Technology (Shenzhen) Co., Ltd. Data noise reduction method, device, computer apparatus and storage medium
WO2020056968A1 (en) * 2018-09-18 2020-03-26 平安科技（深圳）有限公司 Data denoising method and apparatus, computer device, and storage medium
CN109471853A (en) * 2018-09-18 2019-03-15 平安科技（深圳）有限公司 Data noise reduction, device, computer equipment and storage medium
CN109471853B (en) * 2018-09-18 2023-06-16 平安科技（深圳）有限公司 Data noise reduction method, device, computer equipment and storage medium
CN109447947A (en) * 2018-09-28 2019-03-08 上海理工大学 A kind of central fovea localization method based on lesion colour retinal fundus images
CN109447947B (en) * 2018-09-28 2021-10-01 上海理工大学 Fovea location method based on focus color retina fundus image
CN109473148A (en) * 2018-10-26 2019-03-15 武汉工程大学 A kind of ion concentration prediction technique, device and computer storage medium
CN109448006A (en) * 2018-11-01 2019-03-08 江西理工大学 A kind of U-shaped intensive connection Segmentation Method of Retinal Blood Vessels of attention mechanism
CN109448038A (en) * 2018-11-06 2019-03-08 哈尔滨工程大学 Sediment sonar image feature extracting method based on DRLBP and random forest
CN109615614B (en) * 2018-11-26 2020-08-18 北京工业大学 Method for extracting blood vessels in fundus image based on multi-feature fusion and electronic equipment
CN109615614A (en) * 2018-11-26 2019-04-12 北京工业大学 The extracting method and electronic equipment of eye fundus image medium vessels based on multi-feature fusion
CN110189327A (en) * 2019-04-15 2019-08-30 浙江工业大学 Eye ground blood vessel segmentation method based on structuring random forest encoder
CN110288588A (en) * 2019-07-01 2019-09-27 齐鲁工业大学 Retinal images blood vessel segmentation method and system based on gray variance and standard deviation
CN110472600A (en) * 2019-08-20 2019-11-19 北京百度网讯科技有限公司 The identification of eyeground figure and its training method, device, equipment and storage medium
CN113011340A (en) * 2021-03-22 2021-06-22 华南理工大学 Cardiovascular surgery index risk classification method and system based on retina image
CN113011340B (en) * 2021-03-22 2023-12-19 华南理工大学 Cardiovascular operation index risk classification method and system based on retina image
CN113781514A (en) * 2021-09-14 2021-12-10 西南交通大学 Retina blood vessel segmentation method with multi-low-level feature fusion
CN113989170A (en) * 2021-10-29 2022-01-28 南开大学 Transparent blood vessel type identification method based on blood flow characteristics
CN113989246B (en) * 2021-10-29 2023-01-24 南开大学 Transparent blood vessel image segmentation method based on blood flow characteristics
CN113989170B (en) * 2021-10-29 2023-01-24 南开大学 Transparent blood vessel type identification method based on blood flow characteristics
CN113989246A (en) * 2021-10-29 2022-01-28 南开大学 Transparent blood vessel image segmentation method based on blood flow characteristics
CN114913099A (en) * 2021-12-28 2022-08-16 天翼数字生活科技有限公司 Method and system for processing video file
CN116091451A (en) * 2023-01-09 2023-05-09 中国科学院苏州生物医学工程技术研究所 Retinal pigment epithelial cell image segmentation method and system
CN117689552A (en) * 2024-02-02 2024-03-12 科普云医疗软件(深圳)有限公司 Coronary angiography enhancement method for intracardiac interventional therapy
CN117689552B (en) * 2024-02-02 2024-04-05 科普云医疗软件(深圳)有限公司 Coronary angiography enhancement method for intracardiac interventional therapy

Similar Documents

Publication Publication Date Title
CN107248161A (en) Retinal vessel extracting method is supervised in a kind of having for multiple features fusion
CN109493954B (en) SD-OCT image retinopathy detection system based on category distinguishing and positioning
US20200234445A1 (en) Method and system for classifying diabetic retina images based on deep learning
Lu et al. Automatic optic disc detection from retinal images by a line operator
CN104463140B (en) A kind of colored eye fundus image optic disk automatic positioning method
CN105261015B (en) Eye fundus image blood vessel automatic division method based on Gabor filter
CN105160660B (en) Active contour vessel extraction method and system based on multiple features Gauss curve fitting
CN101984916B (en) Blood vessel diameter measuring method based on digital image processing technology
CN106530283A (en) SVM (support vector machine)-based medical image blood vessel recognition method
CN102982542B (en) Fundus image vascular segmentation method based on phase congruency
CN101667289B (en) Retinal image segmentation method based on NSCT feature extraction and supervised classification
CN104794721B (en) A kind of quick optic disk localization method based on multiple dimensioned spot detection
CN106886991A (en) A kind of fuzziness automatic grading method based on colored eyeground figure
CN104809480A (en) Retinal vessel segmentation method of fundus image based on classification and regression tree and AdaBoost
CN107330876A (en) A kind of image automatic diagnosis method based on convolutional neural networks
CN106355599A (en) Non-fluorescent eye fundus image based automatic segmentation method for retinal blood vessels
CN107358612A (en) A kind of retinal vessel segmenting system combined based on fractal dimension with gaussian filtering and method
CN107292835A (en) A kind of method and device of eye fundus image retinal vessel Automatic Vector
CN107229937A (en) A kind of retinal vessel sorting technique and device
Rodrigues et al. Retinal vessel segmentation using parallel grayscale skeletonization algorithm and mathematical morphology
Dikkala et al. A comprehensive analysis of morphological process dependent retinal blood vessel segmentation
CN106327490A (en) Nucleus segmentation method based on white blood cell detection
CN113506284A (en) Fundus image microangioma detection device and method and storage medium
CN103985113B (en) Tongue is as dividing method
CN105930811A (en) Palm texture feature detection method based on image processing

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20171013

RJ01 Rejection of invention patent application after publication