CN106919950B - The brain MR image segmentation of probability density weighting geodesic distance - Google Patents
The brain MR image segmentation of probability density weighting geodesic distance Download PDFInfo
- Publication number
- CN106919950B CN106919950B CN201710053148.5A CN201710053148A CN106919950B CN 106919950 B CN106919950 B CN 106919950B CN 201710053148 A CN201710053148 A CN 201710053148A CN 106919950 B CN106919950 B CN 106919950B
- Authority
- CN
- China
- Prior art keywords
- pixel
- super
- image
- segmentation
- seed point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/20076—Probabilistic 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
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of brain MR image segmentations of probability density weighting geodesic distance, belong to technical field of image processing.Include: to read in several width simulation brain database images, carries out statistics with histogram, obtain the sample value in most integrated distribution section；Multilayer networks are carried out to each pixel on the image to be processed of selection as priori knowledge using obtained sample value；Super-pixel segmentation is carried out to image to be processed based on probability density function；Super-pixel after segmentation is scanned, non-compliant super-pixel is filtered out to be divided, all pixels in super-pixel are gathered again for two classes with FCM algorithm, connected region is found according to classification results, and using the pixel in each connected region as new one kind, super-pixel classification results matrix is updated；It is clustered on the basis of all updated super-pixel with FCM algorithm, obtains image brain tissue to be processed segmentation result.The present invention improves the accuracy of super-pixel segmentation and brain tissue's segmentation.
Description
Technical field
The present invention relates to technical field of image processing, particularly relate to a kind of brain MR figure of probability density weighting geodesic distance
As dividing method.
Background technique
Image segmentation is one of most classic research topic in the fields such as image procossing, image analysis and computer vision,
It is the largest one of difficult point, image Segmentation Technology plays each in pivotal player and image in many medical image applications
The basis that the pathology of kind tissue and organ is further analyzed mentions region more interested in image by utilizing image segmentation
It takes out, provides foundation for clinical diagnosis and treatment etc., and brain is the vitals of human body, therefore study point of brain area
Technology is cut for the equal important in inhibiting of diagnosis of the research of brain three-dimensional reconstruction, neural circuitry and clinical brain diseases.
Super-pixel segmentation is a kind of image over-segmentation algorithm, can be used as the pretreatment work in some image applications, example
Such as segmentation, conspicuousness detection, recognition of face.Super-pixel can capture the redundancy in image, greatly reduce subsequent image processing
The complexity of task.A kind of existing effective superpixel segmentation method is based on geodesic distance (Geodesic Distance)
Super-pixel, with geodesic distance rather than Euclidean distance measurement pixel between similarity, divide for natural image
Effect is pretty good, but for brain MR (Magnetic Resonance, abbreviation MR) image, this method can not be separated accurately often
One tiny brain tissue's region super-pixel block.
Fuzzy C-Means Cluster Algorithm (Fuzzy C-Means, abbreviation FCM) is the fuzzy clustering image being most widely used
Partitioning algorithm.Relative to other dividing methods, FCM can retain more information of initial pictures.However, traditional FCM is calculated
Method fails to consider the gray feature of each point and its correlation degree of neighborhood territory pixel in image segmentation, result in the algorithm for
Noise and gray scale are unevenly more sensitive, in view of the above-mentioned problems, it has been proposed that many improved FCM algorithms, although improved
Method anti-noise or in terms of be improved to some extent, but due to the high complexity of brain image, cannot still obtain
Satisfactory segmentation result, therefore actual requirement is not able to satisfy using traditional single method segmentation.
Summary of the invention
The present invention provides a kind of brain MR image segmentation of probability density weighting geodesic distance, and which raises super-pixel
The accuracy of segmentation and brain tissue's segmentation.
In order to solve the above technical problems, present invention offer technical solution is as follows:
A kind of brain MR image segmentation of probability density weighting geodesic distance, comprising:
Step 1: read in several width simulation brain database images, statistics with histogram carried out to it, obtain white matter, grey matter or
The sample value in cerebrospinal fluid most integrated distribution section；
Step 2: piece image is randomly selected from the simulation brain database images as image to be processed, using obtaining
The most integrated distribution section sample value as priori knowledge on image to be processed each pixel carry out probability it is close
Degree estimation, obtains probability density function；
Step 3: super-pixel segmentation being carried out to image to be processed based on the obtained probability density function, and is recorded super
Pixel classifications matrix of consequence；
Step 4: the super-pixel after segmentation being scanned, is filtered out according to super-pixel color standard difference non-compliant
Super-pixel is divided, and when division, is gathered all pixels in super-pixel for two classes again with FCM algorithm, is tied later according to classification
Fruit finds connected region, and using the pixel in each connected region as new one kind, updates the super-pixel classification results square
Battle array；
Step 5: according to updated super-pixel classification results matrix, with FCM algorithm in all updated super-pixel bases
It is clustered on plinth, obtains image brain tissue to be processed segmentation result.
The invention has the following advantages:
The brain MR image segmentation of probability density weighting geodesic distance of the invention, first obtains statistics with histogram
Sample value as priori value on image each pixel carry out Multilayer networks, then with based on probability density weight
The super-pixel method of geodesic distance carries out super-pixel segmentation to image, then scans super-pixel and filters out non-compliant super picture
Element is divided, and is carried out merokinesis optimization using super-pixel internal feature, is updated super-pixel classification results square after the completion of division
Battle array, finally completes the segmentation of brain image based on being clustered again on the basis of the super-pixel divided with FCM algorithm.This method
It at least has the advantages that (1) designs the new weights influence factor to define geodesic distance, has incorporated probability density function, made
Compared between brain different tissues it is more obvious, gradient calculate it is more reasonable；(2) increase the process of local segmentation post-processing, it is right
Super-pixel carries out merokinesis, and pixel is more accurately sorted out, further improves the accuracy of super-pixel segmentation, that is, uses
Most simple most traditional FCM carries out last cluster, as a result still very well；It (3) will be based on probability density weighting geodesic distance
The methods of super-pixel segmentation technology, FCM are connected, and brain tissue's segmentation result is obtained on the basis of super-pixel.
Detailed description of the invention
Fig. 1 is the flow diagram for the brain MR image segmentation that probability density of the invention weights geodesic distance；
Fig. 2 is the schematic illustration for the brain MR image segmentation that probability density of the invention weights geodesic distance；
Fig. 3 is that the process of step 3 in the brain MR image segmentation of probability density weighting geodesic distance of the invention is shown
It is intended to；
Fig. 4 is the detailed process signal for the brain MR image segmentation that probability density of the invention weights geodesic distance
Figure；
Fig. 5 (a): for image to be processed of the invention: synthesizing brain MR example image；
Fig. 5 (b): for the super-pixel segmentation figure for corresponding to Fig. 5 (a) and synthesizing brain MR example image of the invention；
Fig. 5 (c): for regions such as the brain MR example image realized according to the present invention final white matter, grey matter, cerebrospinal fluid
Segmentation result figure；
Fig. 6 (a): for image to be processed of the invention: true brain MR example image；
Fig. 6 (b): for the super-pixel segmentation figure for corresponding to the true brain MR example image of Fig. 6 (a) of the invention；
Fig. 6 (c): for regions such as the brain MR example image realized according to the present invention final white matter, grey matter, cerebrospinal fluid
Segmentation result figure.
Specific embodiment
To keep the technical problem to be solved in the present invention, technical solution and advantage clearer, below in conjunction with attached drawing and tool
Body embodiment is described in detail.
The present invention provides a kind of brain MR image segmentation of probability density weighting geodesic distance, as shown in figures 1 to 6, packet
It includes:
Step 1: read in several width simulation brain database images, statistics with histogram carried out to it, obtain white matter, grey matter or
The sample value in cerebrospinal fluid most integrated distribution section；
In this step, the simulation brain database images of reading are the synthesis brain MR image of bmp format, read in its gray scale
Value.
Step 2: from simulation brain database images piece image is randomly selected as image to be processed, using obtaining most
The sample value in integrated distribution section carries out Multilayer networks to each pixel on image to be processed as priori knowledge, obtains
To probability density function；
Step 3: super-pixel segmentation being carried out to image to be processed based on obtained probability density function, and records super-pixel
Classification results matrix；
Step 4: the super-pixel after segmentation being scanned, is filtered out according to super-pixel color standard difference non-compliant
Super-pixel is divided, and when division, is gathered all pixels in super-pixel for two classes again with FCM algorithm, is tied later according to classification
Fruit finds connected region, and using the pixel in each connected region as new one kind, updates super-pixel classification results matrix；
Step 5: according to updated super-pixel classification results matrix, with FCM algorithm in all updated super-pixel bases
It is clustered on plinth, obtains image brain tissue to be processed segmentation result.
The brain MR image segmentation of probability density weighting geodesic distance of the invention, first obtains statistics with histogram
Sample value as priori value on image each pixel carry out Multilayer networks, then with based on probability density weight
The super-pixel method of geodesic distance carries out super-pixel segmentation to image, then scans super-pixel and filters out non-compliant super picture
Element is divided, and is carried out merokinesis optimization using super-pixel internal feature, is updated super-pixel classification results square after the completion of division
Battle array, finally completes the segmentation of brain image based on being clustered again on the basis of the super-pixel divided with FCM algorithm.This method
It at least has the advantages that (1) designs the new weights influence factor to define geodesic distance, has incorporated probability density function, made
Compared between brain different tissues it is more obvious, gradient calculate it is more reasonable；(2) increase the process of local segmentation post-processing, it is right
Super-pixel carries out merokinesis, and pixel is more accurately sorted out, further improves the accuracy of super-pixel segmentation, that is, uses
Most simple most traditional FCM carries out last cluster, as a result still very well；It (3) will be based on probability density weighting geodesic distance
The methods of super-pixel segmentation technology, FCM are connected, and brain tissue's segmentation result is obtained on the basis of super-pixel.
Preferably, step 1 is further are as follows:
It reads in several width simulation brain MR and (obtains) image from BrainWeb database, K-means algorithm first is used to it
Preliminary classification is carried out, N number of section is divided into after then normalizing gray value, to the gray value of its white matter, grey matter or cerebrospinal fluid
Range is counted, and K section of white matter, grey matter or cerebrospinal fluid most integrated distribution is counted, in this K section, each
M gray value is chosen as sample in section, and the numerical value of the sample is the sample value in most integrated distribution section.In this step, N,
K, m are the integer greater than 0.
In the present invention, due to brain image grey matter, white matter, cerebrospinal fluid different tissues grey scale pixel value between have coincidence
Place, none exact division boundary between them, the classification that each pixel belongs to be it is fuzzy, directly to it
Segmentation is difficult.So the present invention carries out statistics with histogram first, using obtained sample value as priori knowledge, it is used to
Multilayer networks are carried out to each pixel on image in next step.
Further, step 2 includes:
Step 21: K probability Estimation model of each pixel in image to be processed is calculated separately out according to formula (1):
Wherein, x is the gray value of each pixel in image, xiIt is a section of kth (k=1,2 ..., K) as priori
M gray value sample of value, h are bandwidth (control parameters)；
Step 22: K obtained probability Estimation model being calculated into mixing probability density function by formula (2), is obtained every
A possibility that Multilayer networks value of a pixel, i.e., each pixel belongs to white matter, grey matter or cerebrospinal fluid:
Wherein, p (k) is impact factor of each probability Estimation model to data point, between 0~1；
Step 23: obtained mixing probability density function is normalized, probability density function is obtained:
Wherein, wmax and wmin is the maximum value and minimum value of P (x).
In the present invention, probability density function has been incorporated, can have been made between the brains different tissues such as grey matter, white matter, cerebrospinal fluid
Comparison it is more obvious, and be accurately calculated us subsequent gradient.
As an improvement of the present invention, as shown in figure 3, step 3 includes:
Step 31: initialization seed point is first uniformly distributed n/2 seed point, then with adaptive hexagon on the image
(Adaptive Hexagonal) method is inserted into other seed points, and the complexity of each seed point hexagon is calculated with formula (4), is looked for
To the maximum hexagon of complexity, it is divided into the small hexagon of six overlappings, is inserted into the maximum small hexagon of complexity
One new seed point, successively iteration, until sampling n cluster centre, complexity is defined as:
Wherein, HiIt is seed point siHexagonal area, N is the pixel quantity of image I, and M is to meet condition in braces
Pixel p quantity, α is control parameter, It is the ladder of pixel p on image
Degree, GσIt is the Gaussian function with standard deviation sigma, ω is an adjustment parameter, is preventedThe case where being zero；
Step 32: upsetting seed point, each seed point is moved to the minimum gradient locations in its part region 3*3, and remember
Its X, Y coordinates information is recorded as new seed point；
Step 33: the seed point sensitising gradient based on probability density is calculated with formula (5):
PSSG(si, G (t))=| | Sp(si,G(t))|| (5)
Wherein,SpIt is the P calculated on geodetic path with sobel operatorw(x) gradient；
Step 34: it is calculated using FMM (fast marching method) algorithm and geodesic distance is weighted based on probability density,
Row bound of going forward side by side diffusion generates a series of pixels with like attribute and forms super-pixel block；
In this step, A: geodesic distance is calculated by the following method:
From seed point siTo a description of any one pixel p probability density weighting geodesic distance are as follows: from seed point siIt opens
Begin to reach pixel p along a shortest path, on path every multiplied by a weighting function W (si, G (t)) minimum arc
Long integral, is defined as:
Wherein, G (t) is from seed point siTo a geodetic path between pixel, t be it is continually changing, take 0~1
Between value, weight W is set as the gradient that a pixel belongs to white matter, grey matter or cerebrospinal fluid possibility, for defining from kind
Sub- point siDistance increment onto some pixel path G:
Geodesic distance is calculated by the expanding policy of the Fast Marching Method (FMM) with appropriate velocity field iterative diffusion
, velocity function is defined based on formula (8), its calculation formula is:
B: boundary diffusion generates super-pixel block:
In diffusion process, the new speed of each pixel be no longer it is static, dependent on the seed point nearest from it
(weighting the pixel of geodesic distance with shortest probability density with it), since given seed point, is counted with formula (5)
The gradient for calculating its neighborhood territory pixel point, extends along the pixel with maximum speed formula (8), the pixel color value after extension
It will be replaced by the color value of seed point, the probability seed point sensitising gradient of next pixel can constantly pass through diffusing through for FMM
The value of the current neighborhood territory pixel point of journey is calculated with formula (5) to be updated, and is spread forward every time from the minimum (speed of seed point geodesic distance
It is maximum to spend function) pixel, finished until all pixels are all spread, obtain the classification results matrix of super-pixel.
Step 35: position and the color of seed point are updated with publicity (9)；
Wherein, SlIt is first of super-pixel, slIt is super-pixel SlSeed point, xl', cl' respectively be update after seed point position
It sets and color value,Measurement pixel is subordinate to the weighting function of seed point degree；
Step 36: repeating step 33, step 34, step 35, algorithm is intended to optimize an energy function, is defined as formula
(10), when the change of energy function in subsequent iteration twice threshold specific less than one just stops, preliminary super-pixel
Segmentation is completed,
In the present invention, for the high complexity of brain image, the new weights influence factor is designed to define geodesic distance, is made
With new gradient calculation method, a more obvious boundary is had at the fuzzy region edge of brain MR image, it can be calibrated
True Ground Split goes out each tiny brain tissue's region super-pixel block.
Further, step 4 includes:
Step 41: the super-pixel after scanning segmentation, if the color standard difference of super-pixel is greater than some threshold value Tc, then to full
The super-pixel of this condition of foot is divided, and is calculated is defined as:
C(sl)=λ Stdl> Tc (11)
Wherein, StdlIt is the standard deviation of pixel color in each super-pixel, TcIt is threshold value, λ is control parameter；
Step 42: needing all pixels in the super-pixel divided to gather again for two classes, local FCM for each with FCM algorithm
Objective function are as follows:
Wherein, C is the number of cluster, QnIt is the number of pixel in the super-pixel of Current Scan, μijIt is j-th of pixel category
In the fuzzy membership function of ith cluster, meet constraint μij∈[0,1],M is acted on fuzzy membership
Weighting function, viIt is ith cluster center, xjIt is the color value of pixel in the super-pixel of Current Scan；
Step 43: finding connected region according to obtained super-pixel classification results, and the pixel in each connected region
As new one kind, i.e. a new super-pixel block, the matrix of consequence of super-pixel classification is updated；
Step 44: step 42, step 43 are repeated until all super pictures for meeting splitting condition to the super-pixel that needs divide
Element all divisions are completed.
In the present invention, local segmentation post-processing is carried out again after the completion of primary segmentation, utilizes super-pixel internal feature carry out office
Optimization is split in part, sorts out pixel more accurately, improves the segmentation accuracy of super-pixel, allows the division of super-pixel more
Accurately brain different tissues are separated by further clustering.
Preferably, step 5 is further are as follows:
Input by super-pixel (rather than pixel) as FCM algorithm is gathered on the basis of the super-pixel divided
Class completes the segmentation of brain image, the FCM objective function based on super-pixel are as follows:
Wherein, C' is the number of cluster, Qn' be super-pixel in image number, μ 'ijIt is that j-th of super-pixel belongs to i-th
The fuzzy membership function of cluster meets constraintM is the weighting function acted on fuzzy membership,
v′iIt is ith cluster center, ξiIt is super-pixel SjColor mean value.
In the present invention, the super-pixel segmentation technology based on probability density weighting geodesic distance, FCM algorithm are connected,
Brain tissue's segmentation result is obtained on the basis of super-pixel, improves segmentation efficiency and accuracy.
As an improvement of the present invention, the image of reading may be the true brain MR image of dcm format, true brain
The processing mode of portion's MR image is similar with the synthesis processing mode of brain MR image, and difference is, synthesis brain MR image is read in
Be gray value, and what true brain MR image was read in is the density value limited by window width and window level.
Processing result image is analyzed below, is such as directed to Fig. 5 (a), the picture that Fig. 6 (a) is provided utilizes side above
Method is handled, and shown in processing result such as Fig. 5 (c), Fig. 6 (c), Fig. 5 (b), Fig. 6 (b) are super-pixel segmentation figure.
In Fig. 5 (b) and Fig. 6 (b), it can be seen that super-pixel segmentation of the present invention is accurately boundary from visual effect
It is bonded edge, can accurately separate each tiny region, separates white matter (WM), grey matter (GM), spinal fluid to be subsequent
(CSF) etc. the cluster work of brain tissues provides solid foundation.Our effect is even clustered with most simple most traditional FCM
Fruit is still fine.
In Fig. 5 (c) and Fig. 6 (c), it can be seen that the present invention can accurately be partitioned into brain tissue's image, maximum journey
Remain to degree the raw information of image.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the art
For, without departing from the principles of the present invention, it can also make several improvements and retouch, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (3)
1. a kind of brain MR image segmentation of probability density weighting geodesic distance characterized by comprising
Step 1: reading in several width simulation brain database images, statistics with histogram is carried out to it, obtains white matter, grey matter or brain ridge
The sample value in liquid most integrated distribution section；
Step 2: randomly selecting piece image from the simulation brain database images as image to be processed, utilize obtained institute
The sample value for stating most integrated distribution section carries out probability density to each pixel on image to be processed as priori knowledge and estimates
Meter, obtains probability density function；
Step 3: super-pixel segmentation being carried out to image to be processed based on the obtained probability density function, and records super-pixel
Classification results matrix；
Step 4: the super-pixel after segmentation being scanned, non-compliant super picture is filtered out according to super-pixel color standard difference
Element is divided, and when division, is gathered all pixels in super-pixel for two classes again with FCM algorithm, is sought later according to classification results
Connected region is looked for, and using the pixel in each connected region as new one kind, updates the super-pixel classification results matrix；
Step 5: according to updated super-pixel classification results matrix, with FCM algorithm on the basis of all updated super-pixel
It is clustered, obtains image brain tissue to be processed segmentation result；
The step 1 is further are as follows:
Several width simulation brain MR images are read in, preliminary classification first are carried out with K-means algorithm to it, then by gray value normalizing
Be divided into N number of section after change, the intensity value ranges of its white matter, grey matter or cerebrospinal fluid counted, count white matter, grey matter or
K section of person's cerebrospinal fluid most integrated distribution, in this K section, m gray value is chosen as sample in each section；
The step 2 includes:
Step 21: K probability Estimation model of each pixel in image to be processed is calculated separately out according to formula (1):
Wherein, x is the gray value of each pixel in image, xiIt is m gray value sample of k-th of section as priori value, k
=1,2 ..., K, h be control parameter；
Step 22: K obtained probability Estimation model being calculated into mixing probability density function by formula (2), obtains each picture
A possibility that Multilayer networks value of vegetarian refreshments, i.e., each pixel belongs to white matter, grey matter or cerebrospinal fluid:
Wherein, p (k) is impact factor of each probability Estimation model to data point, between 0~1；
Step 23: obtained mixing probability density function is normalized, probability density function is obtained:
Wherein, wmax and wmin is the maximum value and minimum value of P (x)；
The step 3 includes:
Step 31: initialization seed point is first uniformly distributed n/2 seed point on the image, is then inserted with adaptive hexagon method
Enter other seed points, the complexity of each seed point hexagon is calculated with formula (4), finds the maximum hexagon of complexity,
It is divided into the small hexagon of six overlappings, a new seed point is inserted into the maximum small hexagon of complexity, successively iteration,
Until sampling n cluster centre, complexity is defined as:
Wherein, HiIt is seed point siHexagonal area, N is the pixel quantity of image I, and M is the picture for meeting condition in braces
The quantity of vegetarian refreshments p, α are control parameters, It is the gradient of pixel p on image, Gσ
It is the Gaussian function with standard deviation sigma, ω is an adjustment parameter, is preventedThe case where being zero；
Step 32: upsetting seed point, each seed point is moved to the minimum gradient locations in its part region 3*3, and record it
X, Y coordinates information is as new seed point；
Step 33: the seed point sensitising gradient based on probability density is calculated with formula (5):
PSSG(si, G (t))=| | Sp(si,G(t))|| (5)
Wherein,SpIt is the P calculated on geodetic path with sobel operatorw(x) gradient；
Step 34: being calculated using FMM algorithm and geodesic distance is weighted based on probability density, row bound of going forward side by side diffusion generates a series of
Pixel with like attribute forms super-pixel block；
In this step, A: geodesic distance is calculated by the following method:
From seed point siTo a description of any one pixel p probability density weighting geodesic distance are as follows: from seed point siStart edge
A shortest path reach pixel p, on path every multiplied by a weighting function W (si, G (t)) minimum arc length product
Point, is defined as:
Wherein, G (t) is from seed point siTo a geodetic path between pixel, t be it is continually changing, take between 0~1
Value, weight W is set as the gradient that a pixel belongs to white matter, grey matter or cerebrospinal fluid possibility, for defining from seed point si
Distance increment onto some pixel path G:
Geodesic distance is calculated by the expanding policy of the Fast Marching Method FMM with appropriate velocity field iterative diffusion, base
Velocity function is defined in formula (7), its calculation formula is:
B: boundary diffusion generates super-pixel block:
In diffusion process, the new speed of each pixel be no longer it is static, dependent on the seed point nearest from it, i.e., with
Its pixel with shortest probability density weighting geodesic distance calculates its neighbour with formula (5) since given seed point
The gradient of domain pixel is extended along the pixel with maximum speed formula (8), and the pixel color value after extension will be by planting
The color value of son point replaces, and the probability seed point sensitising gradient of next pixel can be constantly by the diffusion process of FMM with working as
The value of preceding neighborhood territory pixel point is calculated with formula (5) to be updated, and is spread forward every time from seed point geodesic distance minimum i.e. speed letter
The maximum pixel of number, finishes until all pixels are all spread, obtains super-pixel classification results matrix；
Step 35: position and the color of seed point are updated with publicity (9)；
Wherein, SlIt is first of super-pixel, slIt is super-pixel SlSeed point, xl', cl' respectively be update after seed point position and
Color value,Measurement pixel is subordinate to the weighting function of seed point degree；
Step 36: step 33, step 34, step 35 are repeated, algorithm is intended to optimize an energy function, is defined as formula (10),
When the change of energy function in subsequent iteration twice threshold specific less than one just stops, preliminary super-pixel segmentation
It completes,
2. the brain MR image segmentation of probability density weighting geodesic distance according to claim 1, which is characterized in that
The step 4 includes:
Step 41: the super-pixel after scanning segmentation, if the color standard difference of super-pixel is greater than some threshold value Tc, then to meeting this
The super-pixel of condition is divided, and is calculated is defined as:
C(sl)=λ Stdl> Tc (11)
Wherein, StdlIt is the standard deviation of pixel color in each super-pixel, TcIt is threshold value, λ is control parameter；
Step 42: needing all pixels in the super-pixel divided to gather again for two classes, the mesh of local FCM for each with FCM algorithm
Scalar functions are as follows:
Wherein, C is the number of cluster, QnIt is the number of pixel in the super-pixel of Current Scan, μijIt is that j-th of pixel belongs to
The fuzzy membership function of i cluster meets constraintM is the weight acted on fuzzy membership
Function, viIt is ith cluster center, xjIt is the color value of pixel in the super-pixel of Current Scan；
Step 43: find connected region according to obtained super-pixel classification results, and using the pixel in each connected region as
New one kind, i.e. a new super-pixel block, update the super-pixel classification results matrix；
Step 44: to the super-pixel that divides of needs repeat step 42, step 43 until all super-pixel for meeting splitting condition all
Division is completed.
3. the brain MR image segmentation of probability density weighting geodesic distance according to claim 2, which is characterized in that
The step 5 is further are as follows:
Using super-pixel as the input of FCM algorithm, is clustered on the basis of the super-pixel divided, complete brain image
Segmentation, the FCM objective function based on super-pixel are as follows:
Wherein, C' is the number of cluster, Q 'nIt is the number of super-pixel in image, μ 'ijIt is that j-th of super-pixel belongs to ith cluster
Fuzzy membership function, meet constraintM is the weighting function acted on fuzzy membership, v 'i
It is ith cluster center, ξiIt is super-pixel SjColor mean value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710053148.5A CN106919950B (en) | 2017-01-22 | 2017-01-22 | The brain MR image segmentation of probability density weighting geodesic distance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710053148.5A CN106919950B (en) | 2017-01-22 | 2017-01-22 | The brain MR image segmentation of probability density weighting geodesic distance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106919950A CN106919950A (en) | 2017-07-04 |
| CN106919950B true CN106919950B (en) | 2019-10-25 |
Family
ID=59454078
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710053148.5A Active CN106919950B (en) | 2017-01-22 | 2017-01-22 | The brain MR image segmentation of probability density weighting geodesic distance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106919950B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107516314B (en) * | 2017-08-25 | 2020-03-27 | 山东大学 | Medical image hyper-voxel segmentation method and device |
| CN108596236A (en) * | 2018-04-18 | 2018-09-28 | 东南大学 | It is a kind of to roll into a ball partition method based on the thalamic nuclei of global connection features and geodesic distance |
| CN112132842A (en) * | 2020-09-28 | 2020-12-25 | 华东师范大学 | Brain image segmentation method based on SEEDS algorithm and GRU network |
| CN113052859A (en) * | 2021-04-20 | 2021-06-29 | 哈尔滨理工大学 | Super-pixel segmentation method based on self-adaptive seed point density clustering |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103353986A (en) * | 2013-05-30 | 2013-10-16 | 山东大学 | Brain MR image segmentation method based on superpixel fuzzy clustering |
| CN104463870A (en) * | 2014-12-05 | 2015-03-25 | 中国科学院大学 | Image salient region detection method |
| CN105678797A (en) * | 2016-03-04 | 2016-06-15 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Image segmentation method based on visual saliency model |
| CN106296654A (en) * | 2016-07-26 | 2017-01-04 | 中国科学技术大学 | A kind of image superpixel dividing method keeping edge |
-
2017
- 2017-01-22 CN CN201710053148.5A patent/CN106919950B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103353986A (en) * | 2013-05-30 | 2013-10-16 | 山东大学 | Brain MR image segmentation method based on superpixel fuzzy clustering |
| CN104463870A (en) * | 2014-12-05 | 2015-03-25 | 中国科学院大学 | Image salient region detection method |
| CN105678797A (en) * | 2016-03-04 | 2016-06-15 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Image segmentation method based on visual saliency model |
| CN106296654A (en) * | 2016-07-26 | 2017-01-04 | 中国科学技术大学 | A kind of image superpixel dividing method keeping edge |
Non-Patent Citations (8)
| Title |
|---|
| 《Geodesic Matting: A Framework for Fast Interactive Image and Video Segmentation and Matting》;Xue Bai等;《International Journal of Computer Vision》;20081115(第82期);第113-132页 * |
| 《TurboPixels: Fast Superpixels Using Geometric Flows》;Alex Levinshtein等;《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》;20091231;第31卷(第12期);第2290-2297页 * |
| 《交互式前景抠图技术综述》;沈洋等;《计算机辅助设计与图形学学报》;20140430;第26卷(第4期);第511-519页 * |
| 《几种新超像素算法的研究》;张晓平;《控制工程》;20150930;第22卷(第5期);第902-907页 * |
| 《基于测地距离的超像素生成方法》;王爱齐等;《大连理工大学学报》;20120731;第52卷(第4期);第610-614页 * |
| 《基于超像素和模糊聚类的医学超声图像分割算法》;陈放等;《半导体光电》;20160229;第37卷(第1期);第146-150页 * |
| 《基于高斯超像素的快速Graph Cuts图像分割方法》;韩守东等;《自动化学报》;20110131;第37卷(第1期);第11-20页 * |
| 《近似测地距离度量下的图像抗噪分割方法》;高珊珊等;《计算机辅助设计与图形学学报》;20141231;第26卷(第12期);第2214-2222页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106919950A (en) | 2017-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108038476B (en) | A kind of facial expression recognition feature extracting method based on edge detection and SIFT | |
| CN106919950B (en) | The brain MR image segmentation of probability density weighting geodesic distance | |
| CN105512684B (en) | Logo automatic identifying method based on principal component analysis convolutional neural networks | |
| Manoharan | Performance analysis of clustering based image segmentation techniques | |
| Rohlfing et al. | Bee brains, B-splines and computational democracy: Generating an average shape atlas | |
| Kannan et al. | Strong fuzzy c-means in medical image data analysis | |
| CN108932479A (en) | A kind of human body anomaly detection method | |
| CN105913081B (en) | SAR image classification method based on improved PCAnet | |
| CN110120042A (en) | A kind of crop map based on SLIC super-pixel and automatic threshold segmentation is as pest and disease damage method for extracting region | |
| CN112418337B (en) | Multi-feature fusion data classification method based on brain function hyper-network model | |
| CN110021019B (en) | AI-assisted hair thickness distribution analysis method for AGA clinical image | |
| Jin et al. | White matter hyperintensity segmentation from T1 and FLAIR images using fully convolutional neural networks enhanced with residual connections | |
| CN111460966B (en) | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement | |
| CN112785603A (en) | Brain tissue segmentation method based on Unet and superpixel | |
| Amarapur | An automated approach for brain tumor identification using ANN classifier | |
| Remya et al. | Enhanced DWT Filtering technique for brain tumor detection | |
| Alagarsamy et al. | Identification of Brain Tumor using Deep Learning Neural Networks | |
| Borman et al. | Classification of Medicinal Wild Plants Using Radial Basis Function Neural Network with Least Mean Square | |
| Naeini et al. | Improving the dynamic clustering of hyperspectral data based on the integration of swarm optimization and decision analysis | |
| Shelke et al. | A Survey on Automated Brain Tumor Detection and Segmentation from MRI | |
| KR20020079742A (en) | Convolution filtering of similarity data for visual display of enhanced image | |
| Wisaeng et al. | Brain tumor segmentation using fuzzy otsu threshold morphological algorithm | |
| CN113723281B (en) | High-resolution image classification method based on local adaptive scale ensemble learning | |
| Mehena et al. | Medical image segmentation and detection of MR images based on spatial multiple-kernel fuzzy C-means algorithm | |
| CN104573746B (en) | Trypetid category identification method based on magnetic resonance imaging |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |