CN104408453B - Near-infrared venae subcutaneae dividing method based on multi-characters clusterl - Google Patents

Near-infrared venae subcutaneae dividing method based on multi-characters clusterl Download PDF

Info

Publication number
CN104408453B
CN104408453B CN201410515436.4A CN201410515436A CN104408453B CN 104408453 B CN104408453 B CN 104408453B CN 201410515436 A CN201410515436 A CN 201410515436A CN 104408453 B CN104408453 B CN 104408453B
Authority
CN
China
Prior art keywords
blood vessel
vessel
branch
image
segmentation
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
Application number
CN201410515436.4A
Other languages
Chinese (zh)
Other versions
CN104408453A (en
Inventor
杨健
王涌天
刘越
宋宪政
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.)
Beijing Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
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 Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN201410515436.4A priority Critical patent/CN104408453B/en
Publication of CN104408453A publication Critical patent/CN104408453A/en
Application granted granted Critical
Publication of CN104408453B publication Critical patent/CN104408453B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20036—Morphological image processing
    • 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 devises a kind of venae subcutaneae blood vessel segmentation method based near infrared imaging, can realize the multi-feature extraction and automatic cluster of vein blood vessel.First, segmentation and the edge end effect of skin area are realized using NiBlack and Morphology Algorithm；Second step, analyze to obtain blood vessel similarity image, vessel directions figure, blood vessel scalogram by multiple dimensioned IUWT and Hessian matrix and just split blood vessel；3rd step, extracted using first segmentation blood vessel and vessel directions figure and repair vessel branch center line, using the position and direction of the method amendment branch hub line of segmentation spline-fit；4th step, based on the coordinate mapping relations of vessel branch direction calculating artwork to branch's contour images, after IUWT enhancing images and blood vessel similarity graph picture are respectively mapped into contour images space, extract normalized second order Gauss feature and blood vessel similarity feature；5th step, K means algorithms are used to cluster contour images for skin, blood vessel and the class of fuzzy region 3 using obtained blood vessel feature.

Description

Near-infrared venae subcutaneae dividing method based on multi-characters clusterl
Technical field
The present invention relates to a kind of venae subcutaneae blood vessel segmentation method, and in particular to a kind of near-infrared based on multi-characters clusterl Venae subcutaneae dividing method, it is mainly used in the fields such as venae subcutaneae injection, vein identification.
Background technology
As researcher grinds for black light spectral imaging technology and the continuous of human tissue structure light spectrum image-forming characteristic Study carefully, infrared spectrum is imaged in tissue, especially shows splendid enhancing effect in venae subcutaneae imaging.With X-ray and Ultrasonic imaging is compared, infrared vein imaging more safety and conveniently.The vein enhancing display of infrared imaging substantially comes from blood vessel With the spectral response difference of skin, this makes it apply in the intravenous injection process such as children, the elderly, traumatic patient, obese patient Still there is stable vein enhancing effect during the larger special population of difficulty.Meanwhile vein blood vessel and fingerprint etc. other tradition Physiological characteristic is high compared to hiding, is not easy to forge, has security and uniqueness, and except No operation is intervened, its structure is not with the time Changed with the age, there is stability.Therefore, in recent years, the clinical vein based on infrared vein imaging injects auxiliary equipment And the identity recognizing technology based on vein blood vessel feature is all one of the study hotspot in respective field, such as Oscar Cristi Medical Group The intravenous injection accessory system of (Christie Medical Holdings, Inc.) and the palm vein identification of Fuji Tsu System etc..
Infrared vein imaging can be divided into 800nm~1100nm near infrared imaging and 8um~14um according to spectral region 2 kinds of far infrared imagery.Near infrared imaging can be divided into reflective and transmission-type according to imaging mode and be imaged 2 kinds again.Far infrared into As being imaging and passive imaging, because of the influence of the environmental factors such as time, temperature, sweat stain, the change of its image quality is acutely and imaging device is held high It is expensive；Near infrared imaging is Active Imaging, smaller by such environmental effects, and its image quality is relatively stable and equipment is relatively cheap. In near infrared imaging system, transmissive system has a stronger contrast under suitable light conditions, but system suitability compared with Difference, its imaging object are confined to the tissue of thinner thickness, and imaging results are extremely uneven with difference in thickness intensity profile, the intensity of light source It need to be adjusted with testee tissue thickness difference；And in reflective system, reflected image intensity profile is uniform, meanwhile, 3~8mm of near infrared light penetration power can meet the needs of most of vein imaging.Therefore, in actual applications, transmissive system Be suitably applied the imaging of finger vena, and it is reflective be imaged then be suitably applied the vein in other regions such as the back of the hand, arm into As in.However, unless using expensive professional near-infrared image forming apparatus, the aggregate performance of near-infrared vein image is image comparison Spend relatively low, vessel boundary is fuzzy and background has noise.
In order to preferably meet the requirement of the applications such as assisted IV injection, identification, the blood vessel enhancing of vein image and Accurate segmentation is still the key factor for improving application system performance, and one of researcher's focus of attention in field.So And because near-infrared vein image quality is generally poor, the limitation of following several respects often be present in existing vein dividing method Property：
1. because environmental factor is difficult to avoid that, therefore, resulting image irradiation heterogeneity has a strong impact on venous blood Contrast of the pipe in different zones, greatly enhance vein blood vessel enhancing, measurement and segmentation difficulty in shadow region；
2. because near-infrared image image quality is poor, the threshold value class dividing method based on gradation of image information is often difficult to Obtain smooth vessel boundary, there is also substantial amounts of noise, the ability processing of segmentation details are poor in segmentation result.Use shape merely State algorithm solves the problems, such as that two classes can have a strong impact on the accuracy of segmentation again above；
3. also due to the reason for image quality, it is complete that the join domain in blood vessel is often difficult to segmentation, between vessel branch Easily it is broken, while also increases the difficulty of noise filtering.
The content of the invention
In order to solve the above problems, the present invention splits blood vessel structure feature applied to accurate vein blood vessel, proposes one Near-infrared venae subcutaneae dividing method of the kind based on multi-characters clusterl, has considered Image Multiscale half-tone information and blood vessel knot Structure information, realize the accurate segmentation of vein blood vessel.
A kind of near-infrared venae subcutaneae dividing method based on multi-characters clusterl, comprises the following steps：
The first step, the pretreatment of near-infrared vein image：Global threshold is calculated to extract the skin in near-infrared vein image Region Mask, the edge extension of Mask dorsal areas is realized based on Morphology Algorithm；
Second step, vein blood vessel enhancing and blood vessel feature extraction：First by IUWT wavelet decompositions and reconstruct realize it is non- Uniform illumination carries out image enhaucament while correction, multiple dimensioned vessel information is calculated subsequently, based on sea is gloomy, by analyzing Hai Sen The feature of matrix is worth to blood vessel similarity image and just segmentation blood vessel, and blood vessel is obtained by the characteristic vector for analyzing Hessian matrix Directional diagram, scalogram picture is obtained by scale factor；
The extraction and measurement of 3rd step, vessel branch：Based on first segmentation vessel extraction vessel branch center line, blood vessel is utilized Directional diagram realizes the connection between fracture vessel branch, and the fitting of vessel branch is completed using SPL；
4th step, the calculating of contour images and the extraction of blood vessel feature：Vessel branch direction and scalogram based on amendment Picture, the curved blood vessel branch in blood vessel similarity image and IUWT enhancing images and neighborhood region are each mapped to straight line and square The contour images of shape, and the distribution of second order horizontal Gausian and two kinds of blood vessel spies of blood vessel similarity are calculated respectively in contour images Sign；
The segmentation and post processing of 5th step, vein blood vessel：Based on both the above blood vessel feature, by each point in contour images point For skin, blood vessel and the class of fuzzy region three, and segmentation result reflection is mapped under original image coordinate system and realizes that branch merges, then Blood vessel cavity caused by anti-mapping point approximation is filled up using Morphology Algorithm.
Beneficial effects of the present invention：
1. the inhomogeneous illumination eliminated in former near-infrared image is decomposed and rebuild using IUWT to be influenceed, and realizes vein The enhancing of blood vessel；
2. extract to have obtained including blood vessel similarity, just on the whole using the multiple dimensioned blood vessel feature based on Hessian matrix Split a variety of blood vessel features such as blood vessel, vessel directions and blood vessel yardstick, abundant information is provided for follow-up blood vessel Local treatment；
3. realizing the repairing of vessel branch center line based on vessel directions figure, blood vessel segmentation structural integrity ensure that.With Center line spline-fit afterwards further have modified position of center line and vessel directions；
4. the vessel centerline of bending and its neighborhood are each mapped to straight line and square by the vessel branch direction based on amendment The contour images of shape, unify to greatly facilitate the special zone of blood vessel feature in subsequent central line neighborhood while vessel directions and divide Class.
5. the horizontal second order Gauss distribution response and blood vessel similarity for blood vessel classification can accurately extract contour images In tubulose feature.Each point in contour images is divided into 3 classes while adaptive classification is realized using K-means algorithms also to arrange Except the interference of fuzzy region, the accuracy of segmentation ensure that.
Brief description of the drawings
Fig. 1 is vein blood vessel dividing method flow chart proposed by the invention；
Fig. 2 is vein blood vessel enhancing and blood vessel characteristic image extraction flow chart；
Fig. 3 is extraction and the measurement procedure figure of vessel branch；
Fig. 4 is the distribution of vessel centerline neighborhood and blood vessel tracing schematic diagram；
Fig. 5 is calculating and the extraction flow chart of blood vessel feature of contour images.
Embodiment
As shown in Figure 1, the near-infrared venae subcutaneae blood vessel segmentation method based on multi-characters clusterl specifically includes following several Individual step：
Step S101, the pretreatment of near-infrared vein image.
Near-infrared vein image includes three background, skin and vein blood vessel regions.Wherein, the ash of skin and venosomes Angle value is presented as that two parts region has obvious line of demarcation apparently higher than background area in image histogram.Therefore, in order to Downscaled images process range removes edge effect side by side, and the present invention splits to obtain skin and quiet first with Niblack global thresholds Arteries and veins angiosomes, its threshold calculations such as formula (1)：
Tb=Mean-b × std (1)
Wherein, Mean and std is respectively the global average and mean square deviation of image；B is threshold coefficient, under fixed-illumination, Single chooses the requirement that can meet partitioning into skin manually.Then, in order to ensure the segmentation of skin area accurately and completely, this hair It is bright that the noise region of fritter in segmentation result is eliminated using connected domain algorithm, and retain largest connected domain as skin and vein Angiosomes.Meanwhile edge-smoothing and hole-filling are realized using medium filtering and closing operation of mathematical morphology respectively, obtain skin region Domain masking-out, is designated as Mask.
In addition, in order to simplify BORDER PROCESSING of the Subsequent vessel enhancing algorithm in masking-out, the present invention uses a kind of alternative manner Realize the edge end effect of skin area.Iterations is relevant with the number of neighborhood operation and template size in enhancing algorithm, Its iterative process is as follows：
1. first, defining an interim tMask=Mask, and once expansion is completed to tMask using 3 × 3 templates and is grasped Make, then treat that continuation point set is represented by：
Set (p)=dilate (tMask)-tMask；
2. then, Set (p) gray value is replaced by the average gray of the point set in tMask in neighborhood；
3. last, renewal tMask regions, i.e. tMask=dilate (tMask).
Edge extension algorithm is handled on original image, and its output image and masking-out image will strengthen collectively as blood vessel And the input of partitioning algorithm, it is unnecessary to be not necessarily the progress of masking-out marginal point again while vessel extraction image procossing scope is limited Judge.
Step S102, vein blood vessel enhancing and the extraction of blood vessel characteristic image.
In order to realize fine blood vessel segmentation, the present invention is using the multi-scale enhancement side analyzed with reference to IUWT and Hessian matrix Method, blood vessel enhancing is carried out in terms of local gray level distribution and tubular structure analysis two respectively to vein image, finally gives blood vessel Characteristic image, its flow are as shown in Figure 2.
IUWT includes picture breakdown and rebuilds two processes, and decomposable process can decompose from image and obtain the details of each yardstick Information, process of reconstruction are then selectively to combine the detailed information of each yardstick.
Picture breakdown process obtains scalogram as C firstly the need of by multiple dimensioned low pass filteri, calculating process is public as follows Formula：
Wherein, f0For edge extension image；hiFor i rank scaling kernels.Due to kernel function hiIt is isotropism and independence , its Two dimensional Distribution can be represented by one-dimensional distribution.0 rank kernel function is one-dimensional to be distributed as [1 464 1]/16, and size is 5 × 5.i Rank kernel function hiOne-dimensional be distributed asSize is (2i+1+5)×(2i+1+ 5) kernel function.Then, wavelet decomposition image wiIt can be obtained by adjacent yardstick image subtraction：wi+1=Ci- Ci+1.Therefore, during picture breakdown, n+1 scalogram picture can obtain n wavelet decomposition image, and the edge as input prolongs Open up image f0It is represented byImage reconstruction be Degree of the details selection combination process, experiment show blood vessel feature from More obvious in the continuous range scale of one of p to q, therefore, for Selective long-range DEPT blood vessel feature, reconstructing blood vessel image can It is expressed as
IUWT methods realize the details enhancing based on local gray level distribution, but also enhance the partial noise in image. Therefore, the present invention filters out image using a kind of multi-scale enhancement method based on Hessian matrix analysis image local form information Noise, realize that tubulose knot strengthens, smooth vessel boundary, so as to further reduce blood vessel segmentation and the difficulty of parameter measurement.Image Local form information analysis can realize that, for the point x in figure, its Taylor expansion L is by Taylor expansion：
Wherein,With 0, gradient vectors and Hessian matrix of the respectively point x under yardstick s.Utilize Gaussian linear yardstick The property in space, n ranks (n=1,2) derivation in formula can be converted into the convolution with n rank Gaussian functions.Conversion formula is：
Wherein, parameter γ is the dimension normalization factor, it is ensured that each yardstick derivation response range is consistent.In two dimensional image In, G is two-dimensional Gaussian function, normalization factor γ=2.
Based on the second order local derviation information in Image neighborhood information, tubular structure analysis divides the characteristic value based on Hessian matrix Solution is realized.According to the definition of two-dimentional Hessian matrix, image midpoint p (u, v) Hessian matrix is represented by：
Each component represents the second-order partial differential coefficient in point p 4 directions in neighborhood respectively in matrix, directly represent point p neighborhoods Interior grey scale change degree.According to analyses of the Frangi to Hessian matrix in two dimensional image, the information can also be indirectly for part Structural analysis.It is specifically described as：By carrying out Eigenvalues Decomposition to the Hessian matrix of p points, two eigenvalue λs are can obtain1, λ2.It is false If | λ1| ＜ | λ2|, then work as λ1, λ2Meet simultaneously | λ1| ≈ 0, | λ1| ＜ ＜ | λ2| when, p points meet tubulose feature.Work as tubular area For dead color when, should also meet λ2＞ 0, conversely, when tubular area is light tone, meet λ2＜ 0.Meanwhile λ1, λ2Corresponding feature Vector v1, v2The directional information of two-dimentional tubular structure is reflected respectively.Wherein, λ1Corresponding characteristic vector v1Reflect curvature compared with Small direction, i.e. vessel directions；λ2Corresponding characteristic vector v2Then reflect the larger direction of grey scale change, i.e. blood vessel normal side To.Compared to the method using multi-angle template matches, the method calculating blood vessel angle based on Hessian matrix is quicker, simultaneously It it also avoid the discreteness of template direction sampling process.
The characteristic value for decomposing to obtain based on Hessian matrix, Frangi and Sato et al. define different blood vessel similarities and sentenced According to by experimental comparative analysis, the present invention uses the blood vessel similarity measure that Zhou Shoujun is defined, obtains yardstick σkUnder blood vessel phase Like property function V0(x,σk)：
Wherein, t is filterable agent, can filter out obvious non-vascular region in blood vessel similarity image, suppresses noise jamming. It further can obtain multiple dimensioned blood vessel similarity function V：
In formula, σmaxAnd σminThe out to out and smallest dimension of gaussian kernel function respectively in Hessian matrix calculating.
Above-mentioned blood vessel calculating formula of similarity is applied to IUWT enhancing images by the present invention, takes t=t1, it is similar to obtain blood vessel Spend image VP.Meanwhile characteristic vector v corresponding to each point in image2Vessel directions figure and blood vessel chi therefore can be formed respectively with yardstick Degree figure.
In addition, appropriate threshold value is chosen to blood vessel similarity graph picture can obtain just segmentation blood vessel.First cutting procedure is from image Entirety is set out, using image Hessian matrix characteristic value information, withFor threshold value, most image Mask regions (i.e. skin at last Skin region) press the larger eigenvalue λ of Hessian matrix2Size be divided into two classes.First segmentation threshold t2T should be met2＞ t1, ensure just to divide Cut the part that result is real blood vessels.
Step S103, the extraction and measurement of vessel branch.
In order to carry out the region division based on vessel branch and feature extraction, the present invention utilizes just segmentation blood vessel and blood vessel side The extraction, repairing and fitting of vessel branch are realized to figure, its flow is as shown in Figure 3.
After to just segmentation blood vessel refines to obtain vessel centerline, branch's extraction by center line feature point extraction and branch with Two parts of track form.Center line feature point is the whole story point of vessel branch, including end points and the class of bifurcation two, its extraction process The main neighborhood characteristics using each point on center line.As shown in figure 4, it is p1-p8 by centerline points neighbourhood signatures, clockwise p1->p2...p8->In p1 Index process, bifurcation and end points (can be more than 2 times and 1 time) according to " 0-1 " transition times respectively Extraction obtains, and its value characterizes the traversal number of tracking process., can be real along the tracking process of center line after obtaining branch's whole story point The extraction of existing branch.It is worth noting that, according to the difference of starting point, tracking can be divided into one direction tracking and multi-direction tracking two Kind situation, a tracking inceptive direction need to be randomly choosed for the latter.And during tracking, as Fig. 2 medium vesselses tracked Shown in journey, the direction of subsequent point can be limited in 5 candidate points, meanwhile, 4 neighborhoods o'clock have more compared with 8 neighborhood points in candidate point High priority.Finally, after each characteristic point has been traversed corresponding number, vessel centerline is divided into independent point Branch.
Mending course is from end points.If P (end) is the end points of branch, end is the call number of the point, and AM represents vessel directions Figure, initialization procedure obtain Plast=P (end-L), Pcur=P (end), then branch's repairing are realized by following iterative manner.It is first First, P is calculatedlastTo PcurDirection θ1, and the P that will be recorded in direction AMcurDirection θ2It is compared, works as θ1With θ2Angle is big When 90 °, θ2=θ2+ π, otherwise θ2It is constant.Then, P is madelast=Pcur, and with PcurIt is starting point along θ2Stepping L in direction is reached Pnext, update Pcur=Pnext.Finally, by less than or equal to SmaxAfter the iteration of step, if PnextIntersect with just segmentation blood vessel, then will Branch is sequentially added into by node in extension path, otherwise, enters repairing for next end points after retaining 1/3 extension node Mend.It is worth noting that：L selected conference to cause in the case where extended line intersects with first segmentation blood vessel, PnextCross and just divide Cut region；And L selections are too small, the direction of stepping and anticipated orientation deviation can be caused excessive.Meanwhile SmaxIt is excessive also to cause Degree repairing.
After repairing, the space coordinates distribution of vessel branch center line is no longer uniform, and the center line for refining to obtain is adjacent Point spacing is 1, and it is L to repair obtained center line consecutive points spacing, and therefore, center line needs interpolation in repairing area.Meanwhile The vessel directions figure obtained based on Hessian matrix often has relatively large deviation at intersecting blood vessels, this calculating to contour images and Analysis causes strong influence.Therefore, need to be to center in order to obtain uniform center line distribution and more accurate directional information Line carries out curve interpolation and fitting.A kind of method of the segmentation spline-fit of the invention mentioned using Bankhead etal. is simultaneously Fitting and the interpolation of vessel branch are realized, and is calculated according to the parametric equation of matched curve and updates the direction of center line.In order to Ensure that center line interpolation space coordinate is evenly distributed, the present invention have selected the fitting strategy of Euclidean distance.
Step S104, the calculating of contour images and the extraction of blood vessel feature, flow are as shown in Figure 5.
In analysis of two-dimensional blood vessel, blood vessel transversal has special meaning, contains abundant blood vessel feature.However, by Each vessel branch is curve in image, and its blood vessel section has different directions, therefore to original image medium vessels section Analysis and processing are often very troublesome and time-consuming.Therefore, based on known vessel branch directional information, the present invention is by center The wide uniform sampling of line normal direction obtains the contour images of branch vessel.Sampled point calculates and uses bilinear interpolation, its coordinate Calculation formula is：
Wherein, (x, y) represents sample point coordinate, (x0,y0) blood vessel center line coordinates corresponding to expression section, t expressions (x, Y) with (x0,y0) between distance；N represents section half width, relevant with vessel segment Scale；θ represents blood vessel normal direction.In observation For formula it can be found that although θ and θ ± π represents identical section in image, the sign change of cos θ sin θs can cause profile to intercept Direction it is inconsistent.Therefore, the present invention after starting point normal direction is obtained, to remaining vessel centerline normal direction by with Lower formula is adjusted：
Wherein, θlastThe normal direction of any in expression.
After above-mentioned coordinate and direction transformation relation is obtained, the contour images of arbitrary characteristics image are got in return by the change Arrive.In contour images, vessel directions are unified to vertical direction, are the section of any in vessel centerline per a line, because And the processing to the continuous section of multiple spot on center line becomes very easily, to greatly simplifie the feature extraction of vessel branch.
Based on contour images, two category features that the present invention is extracted in single branch vessel contour images are used to cluster:(1) by sea The blood vessel feature that the blood vessel similarity feature and (2) that gloomy matrix defines are defined by second order Gauss matched filter.Feature (1) carries Take and be based on blood vessel similarity image, after contour images space is mapped that to, the overall normalizing of blood vessel similitude contour images Change result and be characterized 1.The extraction of feature 2 is based on IUWT images, need to equally map that to contour images space.Due to blood vessel Direction is unified to vertical direction in contour images, and the present invention meets second order Gauss distribution merely with a horizontal direction, hung down Nogata is l to length is uniformly distributedyTwo dimension pattern plate carry out blood vessel feature extraction, template can be expressed as：
Wherein, σxFor Gaussian function standard deviation, blood vessel half width is characterized.
In addition, in the extraction process of feature 2, IUWT contour images are along vessel directions intensity profile and uneven, therefore The blood vessel feature of weak angiosomes is still suppressed.In order to solve this problem, present invention employs different from feature 1 to return One changes method, carries out gray scale normalization processing by row to IUWT contour images so that the blood vessel feature of weak angiosomes obtains Recover.Then, using second order Gauss template and contour images convolution, and result is integrally normalized, obtains blood vessel Feature 2.
Step S105, the segmentation and post processing of arteries and veins blood vessel.
Near-infrared vein blood vessel image border is very fuzzy, and manual segmentation process is very cumbersome, and segmentation result varies with each individual, Subjectivity is extremely strong, and the sorting algorithm sample based on supervised learning is difficult to obtain.Therefore, the present invention uses non-supervisory K-means Clustering algorithm is classified to contour images.Blood vessel segmentation process can be regarded as 2 classification problems, however, due to venous blood tube edge Edge is more fuzzy, contour images is divided into two classes non-vascular fuzzy region easily is divided into blood vessel.Therefore, in order to ensure blood vessel point The accuracy cut, the present invention use K-means to cluster contour images for blood vessel, fuzzy region and the class of background area 3.Obviously, The blood vessel feature of blood vessel class is the strongest, and its class center should be farthest away from origin.
After cluster result is obtained, the post processing of image method that the present invention uses includes image reconstruction and morphology repairing two Part.After the cluster result of each branch's contour images of blood vessel is obtained, result need to be mapped in artwork image space, so as to obtain The vessel segmentation of structural integrity.However, because coordinate mapping process has coordinate approximation, process of reconstruction can cause segmentation to be tied The hole of some single pixel ranks in fruit be present.Therefore, mending course uses closing operation of mathematical morphology in segmentation result first Small cavity is filled up.Then, small block noise is there may be in segmentation result in order to filter out, the present invention connects to result again Domain filters.
Although with reference to preferred embodiment, present invention is described, and example described above does not form present invention protection model The restriction enclosed, any modifications, equivalent substitutions and improvements in the spirit and principle of the present invention etc., should be included in the present invention's In claims.

Claims (5)

1. a kind of near-infrared venae subcutaneae dividing method based on multi-characters clusterl, it is characterised in that comprise the following steps：
The first step, the pretreatment of near-infrared vein image：Global threshold is calculated to extract the skin area in near-infrared vein image Mask, the edge extension of Mask dorsal areas is realized based on Morphology Algorithm；
Second step, vein blood vessel enhancing and blood vessel feature extraction：First by IUWT wavelet decompositions and reconstruct realize it is non-homogeneous Image enhaucament is carried out while illumination correction, multiple dimensioned vessel information is calculated subsequently, based on sea is gloomy, by analyzing Hessian matrix Feature be worth to blood vessel similarity image and just segmentation blood vessel, vessel directions are obtained by the characteristic vector for analyzing Hessian matrix Figure, scalogram picture is obtained by scale factor；
The extraction and measurement of 3rd step, vessel branch：Based on first segmentation vessel extraction vessel branch center line, vessel directions are utilized Figure realizes the connection between fracture vessel branch, and the fitting of vessel branch is completed using SPL；
4th step, the calculating of contour images and the extraction of blood vessel feature：Vessel branch direction and scalogram picture based on amendment, will Blood vessel similarity image and curved blood vessel branch in IUWT enhancing images and neighborhood region are each mapped to straight line and rectangle Contour images, and the distribution of second order horizontal Gausian and two kinds of blood vessel features of blood vessel similarity are calculated respectively in contour images；
The segmentation and post processing of 5th step, vein blood vessel：Based on both the above blood vessel feature, each point in contour images is divided into skin Skin, blood vessel and the class of fuzzy region three, and segmentation result reflection is mapped under original image coordinate system and realizes that branch merges, then use Morphology Algorithm fills up blood vessel cavity caused by anti-mapping point approximation.
A kind of 2. near-infrared venae subcutaneae dividing method based on multi-characters clusterl as claimed in claim 1, it is characterised in that The second step comprises the following steps that：
[1] IUWT wavelet decompositions are carried out to pretreatment image and obtains each yardstick detail pictures；
[2] select suitable number of scalogram picture to carry out wavelet reconstruction, strengthen blood vessel and remove illumination interference；
[3] in the Hessian matrix image of multiple dimensioned lower calculating IUWT enhancings image, and the sea under different scale is calculated in every bit The characteristic value and characteristic vector of gloomy matrix；
[4] the blood vessel similarity under calculating is multiple dimensioned；
[5] vessel directions figure is reversely obtained according to characteristic vector；
[6] blood vessel scalogram is obtained according to multiple dimensioned blood vessel Similarity Measure Mesoscale information；
[7] suitable threshold value is selected, obtains blood vessel just segmentation result.
3. a kind of near-infrared venae subcutaneae dividing method based on multi-characters clusterl as claimed in claim 1 or 2, its feature exist In the 3rd step comprises the following steps that：
[1] based on blood vessel just segmentation result extraction vessel centerline；
[2] center line is divided into by vessel branch using the method for blood vessel tracing；
[3] vessel directions figure and vessel branch center line information are based on, branch hub line is repaired；
[4] position correction is further realized using segmentation spline-fit to the branch hub line of repairing；
[5] functional form based on SPL, vessel directions after amendment are calculated.
4. a kind of near-infrared venae subcutaneae dividing method based on multi-characters clusterl as claimed in claim 1 or 2, its feature exist In the 4th step comprises the following steps that：
[1] according to the coordinate mapping relations of its neighborhood of branch hub line direction calculating to rectangular profile image；
[2] branch that IiUuWwtT blood vessels enhancing image and blood vessel similarity graph picture are respectively obtained using the coordinate mapping relations is taken turns Wide image；
[3] branch's contour images corresponding to the image of IUWT enhancings are subjected to gray scale normalization processing by row；
[4] the normalized result of second order horizontal Gaussian function convolution is used, obtains the second order Gauss response of branch vessel；
[5] it is i.e. available to carry out global normalization's processing for blood vessel similarity and the second order Gauss response respectively to branch's contour images Two blood vessel features.
A kind of 5. near-infrared venae subcutaneae dividing method based on multi-characters clusterl as claimed in claim 4, it is characterised in that Described contour images coordinate mapping relations formula is as follows：
<mrow> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>x</mi> <mo>=</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <mi>t</mi> <mi> </mi> <mi>c</mi> <mi>o</mi> <mi>s</mi> <mi>&amp;theta;</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>y</mi> <mo>=</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> <mo>+</mo> <mi>t</mi> <mi> </mi> <mi>s</mi> <mi>i</mi> <mi>n</mi> <mi>&amp;theta;</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>,</mo> <mi>&amp;theta;</mi> <mo>&amp;Element;</mo> <mrow> <mo>(</mo> <mo>-</mo> <mi>&amp;pi;</mi> <mo>,</mo> <mi>&amp;pi;</mi> <mo>)</mo> </mrow> <mo>,</mo> <mi>t</mi> <mo>=</mo> <mo>{</mo> <mo>-</mo> <mi>n</mi> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mo>-</mo> <mn>1</mn> <mo>,</mo> <mn>0</mn> <mo>,</mo> <mn>1</mn> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mi>n</mi> <mo>}</mo> </mrow>
Wherein, (x, y) represents sample point coordinate, (x0,y0) represent section corresponding to blood vessel center line coordinates, t represent (x, y) with (x0,y0) between distance；N represents section half width, relevant with vessel segment yardstick；θ represents blood vessel normal direction.
CN201410515436.4A 2014-09-29 2014-09-29 Near-infrared venae subcutaneae dividing method based on multi-characters clusterl Active CN104408453B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410515436.4A CN104408453B (en) 2014-09-29 2014-09-29 Near-infrared venae subcutaneae dividing method based on multi-characters clusterl

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410515436.4A CN104408453B (en) 2014-09-29 2014-09-29 Near-infrared venae subcutaneae dividing method based on multi-characters clusterl

Publications (2)

Publication Number Publication Date
CN104408453A CN104408453A (en) 2015-03-11
CN104408453B true CN104408453B (en) 2017-12-15

Family

ID=52646084

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410515436.4A Active CN104408453B (en) 2014-09-29 2014-09-29 Near-infrared venae subcutaneae dividing method based on multi-characters clusterl

Country Status (1)

Country Link
CN (1) CN104408453B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107229899B (en) * 2017-03-24 2020-07-24 杭州电子科技大学 Finger vein identification method based on two-dimensional Gaussian maximum curvature
CN108205807A (en) * 2017-12-29 2018-06-26 四川和生视界医药技术开发有限公司 Retinal vascular images edit methods and retinal vascular images editing device
CN109919022A (en) * 2019-01-29 2019-06-21 浙江工业大学 A kind of adaptive inside and outside OCT fingerprint extraction method
CN109903298B (en) * 2019-03-12 2021-03-02 数坤（北京）网络科技有限公司 Method, system and computer storage medium for repairing blood vessel segmentation image fracture
CN110189295A (en) * 2019-04-16 2019-08-30 浙江工业大学 Eye ground blood vessel segmentation method based on random forest and center line
CN111339828B (en) * 2020-01-19 2021-04-30 华中科技大学同济医学院附属协和医院 Vein developing identification method based on combination of infrared image and ultrasonic Doppler
CN111666953B (en) * 2020-06-04 2022-07-01 电子科技大学 Tidal zone surveying and mapping method and device based on semantic segmentation
CN112022346B (en) * 2020-08-31 2022-02-18 同济大学 Control method of full-automatic venipuncture recognition integrated robot
CN115375719A (en) * 2022-08-18 2022-11-22 华科精准（北京）医疗科技有限公司 Blood vessel image segmentation method and device
CN116636866B (en) * 2023-05-17 2024-03-19 北京唯迈医疗设备有限公司 Method, device and storage medium for accurately displaying angiographic image stent

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101006923A (en) * 2007-01-23 2007-08-01 天津理工大学 Identity recognition instrument based on characteristics of subcutaneous vein of dorsum of hand and recognition method
CN101561870A (en) * 2009-04-10 2009-10-21 北方工业大学 Hand back vein collection system and method thereof
CN102813504A (en) * 2012-08-29 2012-12-12 北京理工大学 Multi-spectral three-dimensional venous image display device
CN103337071A (en) * 2013-06-19 2013-10-02 北京理工大学 Device and method for structure-reconstruction-based subcutaneous vein three-dimensional visualization

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR2944203A1 (en) * 2009-04-08 2010-10-15 Gen Electric SYSTEM AND METHOD FOR DETERMINING THE POSITION OF A MEDICAL INSTRUMENT

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101006923A (en) * 2007-01-23 2007-08-01 天津理工大学 Identity recognition instrument based on characteristics of subcutaneous vein of dorsum of hand and recognition method
CN101561870A (en) * 2009-04-10 2009-10-21 北方工业大学 Hand back vein collection system and method thereof
CN102813504A (en) * 2012-08-29 2012-12-12 北京理工大学 Multi-spectral three-dimensional venous image display device
CN103337071A (en) * 2013-06-19 2013-10-02 北京理工大学 Device and method for structure-reconstruction-based subcutaneous vein three-dimensional visualization

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Real-Time Non-invasive Imaging of Subcutaneous Blood Vessels;Xianzheng Song et al.;《Advances in Image and Graphics Technologies,Springer Berlin Heidelberg》;20131231;第1-10页 *
基于多尺度Hessian矩阵和Gabor滤波的造影图像冠脉中心线提取;李颖超 等;《中国医学影像技术》;20131231;第23卷(第1期);第133-136页 *

Also Published As

Publication number Publication date
CN104408453A (en) 2015-03-11

Similar Documents

Publication Publication Date Title
CN104408453B (en) Near-infrared venae subcutaneae dividing method based on multi-characters clusterl
Nasr-Esfahani et al. Vessel extraction in X-ray angiograms using deep learning
US9317761B2 (en) Method and an apparatus for determining vein patterns from a colour image
CN104361626B (en) Venae subcutaneae three-dimensional rebuilding method based on mixing matching strategy
CN101266645B (en) Iris positioning method based on multi-resolution analysis
Thanh et al. Blood vessels segmentation method for retinal fundus images based on adaptive principal curvature and image derivative operators
Ouabida et al. Vander Lugt Correlator based active contours for iris segmentation and tracking
CN107358612A (en) A kind of retinal vessel segmenting system combined based on fractal dimension with gaussian filtering and method
CN110517198A (en) High frequency sensitivity GAN network for LDCT image denoising
CN101667289A (en) Retinal image segmentation method based on NSCT feature extraction and supervised classification
Kakade et al. Identification of nerve in ultrasound images using U-net architecture
Shinde et al. Performance analysis of machine learning algorithm of detection and classification of brain tumor using computer vision
CN109523484B (en) Fractal feature-based finger vein network repair method
Sun et al. Liver lesion segmentation in CT images with MK-FCN
Marattukalam et al. Segmentation of palm vein images using U-Net
Lin et al. YoloCurvSeg: You only label one noisy skeleton for vessel-style curvilinear structure segmentation
CN105488460A (en) Physiological feature based image processing method
Goyal et al. Automatic lesion boundary segmentation in dermoscopic images with ensemble deep learning methods
Parmar et al. Automated melanoma types and stages classification for dermoscopy images
Guo et al. A novel algorithm of dorsal hand vein image segmentation by integrating matched filter and local binary fitting level set model
Olveres et al. Texture descriptor approaches to level set segmentation in medical images
Ma et al. Deep learning models for segmentation of lesion based on ultrasound images
Tang et al. Vein pattern recognition based on RGB images using Monte Carlo simulation and ridge tracking
Singh Diagnosis of skin cancer using novel computer vision and deep learning techniques
Altamimi et al. An improved skin lesion detection solution using multi-step preprocessing features and NASNet transfer learning model

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant