CN100545865C - A kind of automatic division method that image initial partitioning boundary is optimized - Google Patents
A kind of automatic division method that image initial partitioning boundary is optimized Download PDFInfo
- Publication number
- CN100545865C CN100545865C CNB2007100629899A CN200710062989A CN100545865C CN 100545865 C CN100545865 C CN 100545865C CN B2007100629899 A CNB2007100629899 A CN B2007100629899A CN 200710062989 A CN200710062989 A CN 200710062989A CN 100545865 C CN100545865 C CN 100545865C
- Authority
- CN
- China
- Prior art keywords
- neighbour
- point
- image
- value
- boundary
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 101
- 238000000638 solvent extraction Methods 0.000 title claims abstract description 37
- 230000011218 segmentation Effects 0.000 claims abstract description 41
- 238000005457 optimization Methods 0.000 claims abstract description 17
- 230000000694 effects Effects 0.000 claims abstract description 13
- 230000005540 biological transmission Effects 0.000 claims description 4
- 238000004422 calculation algorithm Methods 0.000 abstract description 19
- 238000003709 image segmentation Methods 0.000 abstract description 13
- 238000012545 processing Methods 0.000 abstract description 6
- 239000003550 marker Substances 0.000 abstract description 2
- 238000004088 simulation Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 24
- 230000006870 function Effects 0.000 description 19
- 238000001514 detection method Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 238000005192 partition Methods 0.000 description 4
- 238000000926 separation method Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000000877 morphologic effect Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 229910003460 diamond Inorganic materials 0.000 description 2
- 239000010432 diamond Substances 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 102000000584 Calmodulin Human genes 0.000 description 1
- 108010041952 Calmodulin Proteins 0.000 description 1
- 206010058314 Dysplasia Diseases 0.000 description 1
- 206010028980 Neoplasm Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000005290 field theory Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000007943 implant Substances 0.000 description 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000012113 quantitative test Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 231100000241 scar Toxicity 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to image processing techniques, particularly a kind of automatic optimization method of the initial partitioning boundary based on neighbour's Function Criterion comprises step: detect initial partitioning boundary point and with arabic numeral by the initial segmentation zone marker; To each frontier point, calculate the neighbour's functional value in certain neighborhood; Calculate the membership function value of working as fore boundary point according to neighbour's functional value; Carry out the classification again of frontier point according to membership function value, obtain the partitioning boundary of optimizing; Repeat to implement top step, guarantee that the partitioning boundary of entire image all reaches optimization effect.The inventive method simulation human eye some function when handling image, can be optimized by automatically inaccurate partitioning boundary, and overcome picture noise, local volume effect effectively, intensity is overlapping and the influence of intensity non-uniformity, existing partitioning algorithm is one well replenishes.In medical image segmentation, remote sensing images are cut apart, and fields such as Target Recognition have important use to be worth.
Description
Technical field
The present invention relates to image processing techniques, particularly a kind of Automatic Optimal algorithm of the initial partitioning boundary based on neighbour's Function Criterion.
Background technology
So-called image segmentation is meant that the zones of different that will have special connotation in the image makes a distinction, and these zones do not intersect mutually, and the consistance of specific region is all satisfied in each zone.Cutting apart from the process object angle is the location of definite target of being concerned about image array.Obviously, have only in this way " interested target object " extracted from the scene of complexity, just might further carry out quantitative test or identification, and then image is understood each sub regions.Image segmentation can with feature comprise gradation of image, color, texture, partial statistics characteristic or spectrum signature etc., utilize the difference of these features can differentiate between images in the different target object.Since we can only utilize some Partial Feature cut zone in the image information, so the whole bag of tricks must have limitation and specific aim, can only select suitable dividing method at the demand of various practical application area.
Utilize the difference of character according to partitioning algorithm, image partition method mainly can be divided into two big classes: class methods are based on the method in zone, utilize the zones of different in the homogeneity recognition image in the same area usually; Another kind of method is the edge dividing method; usually utilize interregional heterogeneity (as gray scale uncontinuity in the zone) to mark off the separatrix between each zone; these class methods can cause incomplete part segmentation result usually, are interrupted phenomenon or obtain wrong edge such as existing in the segmentation result.In recent years, along with the application in image segmentation such as statistical theory, fuzzy set theory, neural network, morphology theory, wavelet theory are day by day extensive, new method that genetic algorithm, metric space, multiresolution method, nonlinear diffusion equations etc. are emerged in large numbers in the recent period and new thought also constantly are used to solve segmentation problem, and Chinese scholars has proposed much image partition method targetedly.Below we simply introduce some representational image partition methods.
Dividing method based on the zone
Comprise Threshold Segmentation, region growing and division merging and statistics dividing method etc.These class methods all are to cut apart according to the property difference between the zones of different in the image, do not have or seldom use boundary information in the image.Threshold Segmentation is the dividing method in modal parallel detection zone, and its advantage is simply to be easy to realize, and when inhomogeneous object gray-scale value or other eigenwerts differed greatly, it can be cut apart image very effectively.Its shortcoming is not to be suitable for the image that multichannel image and eigenwert are more or less the same, and for the gray-scale value scope that does not have tangible each object of gray scale difference XOR big overlapping image is arranged, and is difficult to obtain segmentation result accurately.It is two kinds of typical serial region segmentation method that region growing and division merge, and is characterized in cutting procedure is decomposed into the step of a plurality of orders, and wherein subsequent step will be judged according to the result of preceding step and determines.Its basic thought is that the pixel that will have similar quality puts together the formation zone, and this method need be chosen a seed points earlier, successively the similar pixel around the sub pixel is merged in the zone at sub pixel place then.The advantage of region growing algorithm is to calculate simply, is specially adapted to cut apart little structure such as tumour and scar, and shortcoming is that it needs man-machine interactively to obtain seed points, and the user must implant a seed points in each zone that need extract like this.Simultaneously, region growing method is also to noise-sensitive, and there is the cavity in the zone that causes extracting or the zone that will separate originally under the situation of local bulk effect couples together.The statistics dividing method is the method for carrying out statistical classification according to the image-region feature, comprises the markov Random Field Theory, clustering algorithm etc.The calculated amount of this class algorithm is generally bigger, is not easy to real-time processing.
Dividing method based on the edge
Also be called edge detection algorithm, comprise differentiating operator, methods such as curve fitting.Edge detecting technology can be divided into serial rim detection and parallel rim detection according to processing sequence.In the serial edge detecting technology, whether current pixel belongs to the testing result that edge that desire detects depends on first preceding pixel; And in parallel edge detecting technology, the edge whether pixel belongs to detection is only relevant with current pixel and neighbor thereof, can simultaneously all pixels in the image be detected like this, thereby is referred to as parallel edge detecting technology.
The method of calmodulin binding domain CaM and boundary information
Dividing method based on the zone tends to cause over-segmentation, is about to image segmentation and becomes too much zone.If in based on the framework in zone, do not comprise the measure on border, may cause the inner cavity that occurs of noise margin and object in the decision phase.People often will combine based on the method for area information and the method for rim detection, but adopt what mode combination, and how in conjunction with the advantage that just can give full play to separately, obtaining good segmentation result is the emphasis of research.
Method based on fuzzy set theory
Image segmentation problem is typical dysplasia problem, and fuzzy set theory has the ability of describing bad problem, thus there is the researcher that fuzzy theory is incorporated into Flame Image Process and analysis field, comprising solving segmentation problem with fuzzy theory.Image partition method based on fuzzy theory comprises fuzzy threshold segmentation method, fuzzy clustering dividing method and fuzzy degree of connection dividing method etc.
Based on neural network method
In the late nineteen eighties, the main flow field in Flame Image Process, pattern-recognition and computer vision is subjected to the artificial intelligence Influence and Development, the way of recognition system occurred higher level inference mechanism is used for.This thinking also begins to influence image Segmentation Technology, when solving concrete medical problem, the dividing method based on neural network model occurred.The neuron network simulation biology is the learning process of human brain particularly, and it is made of a large amount of parallel nodes, and each node can both be carried out some basic calculating, and learning process realizes by the weights of adjusting internodal annexation and connection.
Method based on mathematical morphology
Mathematical morphology day by day comes into one's own in recent years in Application in Image Processing, and more system all adopts morphological operator to come image is carried out pre-service or aftertreatment.Morphological images is handled in the mode that moves a structural element and carry out convolution in image and is carried out, and structural element can have any size.Basic morphological operation is corrosion and expands that their some fundamental operations mutually combine and can produce complicated effect, and they are suitable for constructing look-up tables'implementation with relevant hardware.The application of morphology theory in image segmentation is more representational to be watershed algorithm, and the researchist has proposed the morphology dividing method of multiple use dividing ridge method so far.Though dividing ridge method successfully is used for image classification, they need the user alternately or accurately about the priori of picture structure.
By top introduction as can be known, more or less all there are some problems in existing partitioning algorithm, should select pointed algorithm in actual use.The goldstandard of judging the segmentation result quality in the reality does not exist, and expert's subjective determination is main examination criteria.
Neighbour's Function Criterion in the pattern-recognition has a lot of similarities with human eye mechanism aspect Flame Image Process, as shown in Figure 1, so we have proposed the optimized Algorithm of image segmentation on this basis, can handle the coarse partitioning boundary of image well.Being one for above-mentioned all partitioning algorithms well replenishes.
Summary of the invention
Prior art can not be optimized by automatically inaccurate partitioning boundary, when handling image, picture noise, local volume effect are arranged, intensity is overlapping and the influence of intensity non-uniformity, the automatic optimization method that the purpose of this invention is to provide a kind of initial partitioning boundary based on neighbour's Function Criterion, considered the mechanism of human eye to Flame Image Process, use neighbour's Function Criterion that the border of cutting apart is cut apart again, improve the accuracy of segmentation result as far as possible.
Core concept of the present invention is to propose a kind of brand-new adaptive method, adopts the full-automatic optimization of neighbour's Function Criterion to initial coarse segmentation result.This method comprises following step: the mark boundaries point; Calculate neighbour's functional value; Calculate membership function value; Frontier point is classified again; Repeat to implement above four steps to guarantee to optimize effect.
Based on above-mentioned purpose and thought, the present invention is based on the Automatic Optimal algorithm of the initial partitioning boundary of neighbour's Function Criterion, a kind of automatic division method that image initial partitioning boundary is optimized that provides comprises:
(1) detect frontier point in the initial segmentation result, and carry out mark by initial segmentation result, all frontier points in first zone are labeled as arabic numeral " 1 ", and the frontier point in second zone is labeled as " 2 ", by that analogy;
(2) calculate neighbour's functional value between having a few in certain neighborhood of each frontier point that is labeled;
(3) calculate the membership function value of working as fore boundary point according to neighbour's functional value, that is to say when fore boundary point is under the jurisdiction of certain regional possibility and measure;
(4) carry out the classification again of frontier point according to membership function value, obtain the partitioning boundary of optimizing;
(5) repeating step 1～step 4 guarantees that the partitioning boundary of entire image all reaches optimization effect.
The present invention simulates human eye some function when handling image, can be optimized by automatically inaccurate partitioning boundary, and overcome picture noise, local volume effect effectively, intensity is overlapping and the influence of intensity non-uniformity, the optimized Algorithm of the image segmentation that the present invention proposes, can handle the coarse partitioning boundary of image well, existing partitioning algorithm is one well replenish.In medical image segmentation, remote sensing images are cut apart, and fields such as Target Recognition have important use to be worth.
Description of drawings
Fig. 1. neighbour's Function Criterion synoptic diagram.As shown in the figure, ω
1The class comparatively dense, ω
2Class is more sparse, sees p intuitively
iBelong to ω
2Class is comparatively reasonable.But by Euclidean distance, p
kBe p
iArest neighbors, if but by neighbour's Function Criterion, p
jBe p
iArest neighbors because p
jWith p
iBetween neighbour's functional value be 1, and p
kWith p
iBetween neighbour's functional value be 5.
Fig. 2. (a) the frontier point mark of initial segmentation.ω
1The frontier point in zone is labeled as " 1 ", ω
2The frontier point in zone is labeled as " 2 "; (b) seek neighbour's point of working as fore boundary point.By Euclidean distance from closely to seeking N neighbour's point in two zones far respectively, N is 11 among the figure.
Fig. 3. the validity with the composograph verification algorithm (a) is original composograph; (e) be initial partitioning boundary, artificially initial partitioning boundary dragged away from the validity of correct split position with the checking optimization method; (b) (c) (d) added noise in various degree on original image, and (f) (g) is the optimization of the inventive method to the noise image segmentation result (h).
Fig. 4. the optimization comparison example on self-adaptation maximum a posteriori estimation (aMAP) segmentation result.First classifies original image as, and second classifies the segmentation result of aMAP as, and the 3rd classifies the result that the inventive method is optimized as, and the 4th classifies the manual result of cutting apart of experienced clinician (being taken as is reference) as.
Fig. 5. the optimization comparison example on level set (LS) segmentation result.First classifies original image as, and second classifies the segmentation result of LS as, and the 3rd classifies the result that the inventive method is optimized as, and the 4th classifies the manual result of cutting apart of experienced clinician (being taken as is reference) as.
Fig. 6. the inventive method is estimated the optimized amount fractional analysis of (aMAP), level set (LS) initial segmentation result to the self-adaptation maximum a posteriori; Wherein the Diamond spot solid line is the aMAP accuracy of separation, and the star point dotted line is the degree of accuracy after the inventive method is optimized the aMAP segmentation result; The round dot solid line is the LS accuracy of separation, and the cross pecked line is the degree of accuracy after the inventive method is optimized the LS segmentation result.
Embodiment
Below in conjunction with accompanying drawing the present invention is described in detail, the optimization method of cutting apart of the present invention mainly may further comprise the steps:
(1) detect frontier point in the initial segmentation result, and carry out mark by initial segmentation result, all frontier points in first zone are labeled as arabic numeral " 1 ", and the frontier point in second zone is labeled as " 2 ", by that analogy;
(2) calculate neighbour's functional value between having a few in certain neighborhood of each frontier point that is labeled;
(3) calculate the membership function value of working as fore boundary point according to neighbour's functional value, that is to say when fore boundary point is under the jurisdiction of certain regional possibility and measure;
(4) carry out the classification again of frontier point according to membership function value, obtain the partitioning boundary of optimizing.
(5) step 1～step 4 repeats to implement, and guarantees that the partitioning boundary of entire image all reaches optimization effect.
Specific embodiment according to the inventive method is as follows:
Step 1: detect initial partitioning boundary point and press the initial segmentation zone marker with arabic numeral:
In the coarse segmentation result of image initial, find its frontier point of cutting apart, and the frontier point in the different cut zone is made different marks, shown in Fig. 2 (a), ω
1The frontier point in zone is labeled as " 1 ", ω
2The frontier point in zone is labeled as " 2 "; If the zone of cutting apart is more than two, just the available marks such as " 3,4,5 ... " of other regional frontier point.This step suitable simple, only need once travel through just and can obtain the result image.
Step 2: calculate the neighbour's functional value between having a few in certain neighborhood of each frontier point that is labeled:
After mark is finished, get a frontier point that is labeled as current process points, we make p at note
0Next step, we will calculate current process points p
0Neighbour's functional value between having a few in certain field.
At first, we will determine current process points p
0Certain neighborhood in have a few which all arranged.In the methods of the invention, we are from closing on current process points p
0Several cut zone in select N neighbour's point respectively, shown in Fig. 2 (b), square is current process points p
0, from closing on current process points p
0Two cut zone in find N neighbour's point respectively.Positive trigpoint is regional ω among the figure
1In neighbour's point, the set of these points is represented as S
1(p
0, N); Inverted triangle point is regional ω
2In neighbour's point, the set of these points is represented as S
2(p
0, N).The set of all these points is the current process points p that we select
0Neighborhood, be expressed as: S (p
0, N)=S
1(p
0, N) ∪ S
2(p
0, N), if the zone of cutting apart is more than two, then
Next step calculating for convenience makes S ' (p again
0, N)=S (p
0, N) ∪ { p
0.In the methods of the invention, N generally gets 25～40, can obtain optimal result because experiment shows the value in this scope.
Then, calculate S ' (p
0, N) the neighbour's functional value between middle the having a few.P sets up an office
iAnd p
jBelong to S ' (p
0, N), then put p
iAnd p
jBetween neighbour's functional value as follows:
a
ij＝s+t-2
Wherein s represents a p
iBe a p
jS neighbour, t represents a p
jBe a p
iT neighbour.
The following describes the value of how specifically calculating s and t, when calculating the neighbour, distance between points is a generalized distance, unites definition according to Euclidean distance and gradation of image value difference are different, and it is defined as follows:
d(p
i，p
j)＝c
1·d
E(p
i，p
j)+c
2·d
G(p
i，p
j)，c
1+c
2＝1，
d
G(p
i，p
j)＝|I(p
i)-I(p
j)|，
D wherein
E(p
i, p
j) expression point-to-point transmission Euclidean distance, d
G(p
i, p
j) absolute value of gray scale difference of expression point-to-point transmission, c
1And c
2Be constant between two 0～1, I (p
i) expression point p
iThe gradation of image value at place.
From a p
iTo S ' (p
0, N) in the distance of all other points form a set, be expressed as:
We this distance by sorting from small to large:
Carry out the sequence number of ordering and point corresponding one by one then.Shown in following formula, some p
iTo current process points p
0The sequence number of distance in ordering be t, we are with regard to mark (p
i→ p
0)=t.Repeat this step, obtain S ' (p
0, N) the middle current process points p that has a few
0Sequence number and current process points p
0To S ' (p
0, the sequence number of being had a few in N) just can obtain current process points p
0To S ' (p
0, N) middle arbitrfary point p
iNeighbour's functional value:
a
0i＝(p
0→p
i)+(p
i→p
0)-2
＝s+t-2
Define current process points p more on this basis
0To neighbour's S set ' (p
0, total neighbour's functional value N) and total distance value:
Wherein
And α
IjAll represent some p
iAnd p
jBetween neighbour's functional value.
Step 3: calculate the membership function value of working as fore boundary point according to neighbour's functional value, that is to say when fore boundary point is under the jurisdiction of certain regional possibility and measure:
The degree of membership that the inventive method adopts is: P
i(p
0)=P
I1(p
0)+P
I2(p
0), i=1,2 ..., C represents that current process points is under the jurisdiction of regional ω
iPossibility.Degree of membership is made up of two parts, and first is by current process points p
0Nearest neighbor point p
0' decision, second portion by
With
The associating decision.
First is expressed as: P
I1(p
0)=(p
0' ∈ ω
i) 50%, (p wherein
0∈ ω
i) be logical expression, expression formula is true, its value is 1, otherwise then is 0.
Second portion is expressed as:
Wherein sgn () is a sign function, and its value is for-1,0 or 1.
Step 4: carry out the classification again of frontier point according to membership function value, obtain the partitioning boundary of optimizing:
On the basis of membership function value, it is optimum to judge which zone is current process points belong on earth.Postulated point p
iBe current process points p
08 Euclidean neighbour points, we judge current process points p by following expression so
0Ownership:
p
0∈ω
i?if?P
i(p
0)≥F
I=1 wherein, 2 ..., C, F=δ max (P
j(p
0)), j ≠ i, δ are constant, and δ ∈ [1,2].If i.e. p
0In initial segmentation result, belong to ω
i, and still belong to ω after judging
i, represent that so initial segmentation result is correct; If p
0In initial segmentation result, belong to ω
i, and do not belong to ω after judging
i, represent that so initial segmentation result is incorrect, initial segmentation result has just obtained optimization.
Step 5: step 1～step 4 repeats to implement, and guarantees that the partitioning boundary of entire image all reaches optimization effect.
Operation result:
In order to verify the inventive method, we test with simulated data and true medical magnetic resonance image.
The simulated data experiment (a) is original composograph as shown in Figure 3, and our boundary in two zones has carried out the degree of difficulty that Fuzzy Processing is cut apart with increase; In order to detect the validity of the inventive method, we have dragged initial partitioning boundary away from correct split position artificially, shown in the blue lines of Fig. 3 (e).(b), (c) and (d) be respectively on the basis of (a), to have added 10%, 20% and 40% Gaussian noise.(f), (g) and (h) be on the basis of initial segmentation, respectively (b), (c) and border (d) to be optimized the result.From this experiment as can be seen, reach in 40% at noise, optimization effect also is tangible! But along with the increase of noise, some problems still occurred, some excessive modifications occurred such as the zone, the lower left corner in (h) at regional area.
In order further to verify the inventive method, we have chosen the real magnetic resonance image (MRI) of 20 width of cloth and have carried out medical image segmentation result's optimization.These images adopt GE 1.5T or 3.0T magnetic resonance imaging system, use the T1 imaging to obtain.These images are cut apart very difficulty owing to be subjected to the influence of factors such as noise, local volume effect, intensity non-uniformity and disperse anisotropy.As Fig. 4, shown in Figure 5, first classifies original image as, second classifies initial segmentation result as, estimate that by the self-adaptation maximum a posteriori (aMAP) and level set (LS) dividing method obtain respectively, the 3rd classifies the result that the inventive method is optimized as, and the 4th classifies the manual result of cutting apart of experienced clinician (being taken as is reference) as.From these two examples as can be seen, the inventive method can be optimized the border of cutting apart greatly, makes segmentation result follow the manual segmentation result of doctor more approaching.
Experiment shows, the inventive method-based on the automatic optimization method of the initial partitioning boundary of neighbour's Function Criterion-effectively raise and cut apart accuracy, reached the purpose of optimizing partitioning boundary.We estimate that with dividing method of the present invention and self-adaptation maximum a posteriori (aMAP) partitioning algorithm, level set partitioning algorithm (LS) compare, the result as shown in Figure 6, wherein the Diamond spot solid line is the aMAP accuracy of separation, and the star point dotted line is the degree of accuracy after the inventive method is optimized the aMAP segmentation result; The round dot solid line is the LS accuracy of separation, the degree of accuracy of cross pecked line after to be the inventive method to the LS segmentation result optimize, can see shown in 20 examples in, method of the present invention all has significantly aMAP and LS methods and results to be optimized.
The above; only be the embodiment among the present invention; but protection scope of the present invention is not limited thereto; anyly be familiar with the people of this technology in the disclosed technical scope of the present invention; can understand conversion or the replacement expected; all should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.
Claims (4)
1. the automatic division method that image initial partitioning boundary is optimized is characterized in that, based on neighbour's Function Criterion is classified automatically in the coarse border of image again, may further comprise the steps:
Step 1: detect the frontier point in the initial segmentation result, and carry out mark by initial segmentation result, all frontier points in first zone are labeled as arabic numeral " 1 ", and the frontier point in second zone is labeled as " 2 ", by that analogy;
Step 2: calculate the neighbour's functional value between having a few in certain neighborhood of each frontier point that is labeled;
Step 3: calculate the membership function value of working as fore boundary point according to neighbour's functional value, that is to say when fore boundary point is under the jurisdiction of certain regional possibility and measure;
Step 4: carry out the classification again of frontier point according to membership function value, obtain the partitioning boundary of optimizing;
Step 5: step 1～step 4 repeats to implement, and guarantees that the partitioning boundary of entire image all reaches optimization effect.
2. method according to claim 1 is characterized in that, in step 2, calculates neighbour's functional value of being had a few in certain neighborhood of each frontier point, establishes S (p
0, N) be a p
0At i cut zone ω
iIn N the set that the neighbour is ordered, establish S ' (p again
0, N)=S (p
0, N) ∪ { p
0; Calculate S ' (p then
0, N) the neighbour's functional value between middle the having a few, p sets up an office
iAnd p
jBe frontier point p
0The neighborhood point, p then
iAnd p
jBetween neighbour's functional value as follows:
a
ij＝s+t-2
Wherein s represents a p
iBe a p
jS neighbour, t represents a p
jBe a p
iT neighbour; Distance between points is a generalized distance, unites definition according to Euclidean distance and gradation of image value difference are different, and it is defined as follows:
d(p
i，p
j)＝c
1·d
E(p
i，p
j)+c
2·d
G(p
i，p
j)，c
1+c
2＝1，
d
G(p
i，p
j)＝|I(p
i)-I(p
j)|，
D wherein
E(p
i, p
j) expression point-to-point transmission Euclidean distance, d
G(p
i, p
j) absolute value of gray scale difference of expression point-to-point transmission, c
1And c
2Be two constants, I (p
i) expression point p
iThe gradation of image value at place;
Point p
0S set ' (p to adjoint point
0, total neighbour's functional value N) and total distance value are respectively:
Wherein
And a
IjAll represent some p
iAnd p
jBetween neighbour's functional value.
3. method according to claim 2 is characterized in that, in step 3, and the subordinate function below adopting:
P
i(p
0)＝P
i1(p
0)+P
i2(p
0)，i＝1，2，…，C
I represents cut zone in the following formula, P
i(p
0) expression p
0Belong to i cut zone ω
iPossibility tolerance, in the formula:
P
i1(p
0)＝(p
0′∈ω
i)·50％
(p wherein
0' ∈ ω
i) be logical expression;
Wherein sgn () is a sign function.
4. method according to claim 3 is characterized in that, in step 4, calculates after the membership function value, determines current some p according to following formula
0Membership:
p
0∈ω
i?if?P
i(p
0)≥F
F=δ max (P wherein
j(p
0)), j ≠ i, δ are constant, and δ ∈ [1,2].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007100629899A CN100545865C (en) | 2007-01-24 | 2007-01-24 | A kind of automatic division method that image initial partitioning boundary is optimized |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007100629899A CN100545865C (en) | 2007-01-24 | 2007-01-24 | A kind of automatic division method that image initial partitioning boundary is optimized |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101231745A CN101231745A (en) | 2008-07-30 |
| CN100545865C true CN100545865C (en) | 2009-09-30 |
Family
ID=39898191
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2007100629899A Expired - Fee Related CN100545865C (en) | 2007-01-24 | 2007-01-24 | A kind of automatic division method that image initial partitioning boundary is optimized |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100545865C (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220358682A1 (en) * | 2020-09-16 | 2022-11-10 | China Tobacco Yunnan Industrial Co., Ltd | Device and method for detecting cigarette fly ash by gray-scale difference based on machine vision (mv) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101404085B (en) * | 2008-10-07 | 2012-05-16 | 华南师范大学 | Partition method for interactive three-dimensional body partition sequence image and application |
| CN101499165B (en) * | 2009-03-09 | 2011-01-26 | 广东威创视讯科技股份有限公司 | Partition method for crossing-overlapped chromosome |
| CN102013017B (en) * | 2010-11-26 | 2012-07-04 | 华中科技大学 | Method for roughly sorting high-resolution remote sensing image scene |
| US9271652B2 (en) * | 2011-03-07 | 2016-03-01 | Koninklijke Philips N.V. | MR segmentation using nuclear emission data in hybrid nuclear imaging/MR |
| US20150347862A1 (en) * | 2014-05-30 | 2015-12-03 | Fuji Xerox Co., Ltd. | Image processing apparatus, image processing method, image processing system, and non-transitory computer readable medium storing program |
| CN104715483B (en) * | 2015-03-20 | 2018-05-11 | 中国科学院自动化研究所 | The automatic division method of lung CT image abnormal area |
| DE102015217429A1 (en) * | 2015-09-11 | 2017-03-16 | Siemens Healthcare Gmbh | Diagnostic system and diagnostic procedure |
| CN107507201A (en) * | 2017-09-22 | 2017-12-22 | 深圳天琴医疗科技有限公司 | A kind of medical image cutting method and device |
| CN110910396A (en) * | 2019-10-18 | 2020-03-24 | 北京量健智能科技有限公司 | Method and device for optimizing image segmentation result |
| CN113311006A (en) * | 2020-02-26 | 2021-08-27 | 乐达创意科技股份有限公司 | Automatic optical detection system and method for detecting edge defects of contact lens |
| CN115281584B (en) * | 2022-06-30 | 2023-08-15 | 中国科学院自动化研究所 | Flexible endoscope robot control system and flexible endoscope robot simulation method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1388948A (en) * | 2000-09-07 | 2003-01-01 | 皇家菲利浦电子有限公司 | Segmentation of digital images |
| CN1418353A (en) * | 2000-01-18 | 2003-05-14 | 芝加哥大学 | Automated method and system for segmentation of lung regions in computed tomography scans |
| US20040120574A1 (en) * | 2002-12-18 | 2004-06-24 | Xerox Corporation | Systems and method for automatically choosing visual characteristics to highlight a target against a background |
| CN1711556A (en) * | 2002-11-13 | 2005-12-21 | 皇家飞利浦电子股份有限公司 | A method of and system for detecting uniform color segments |
| JP2006346465A (en) * | 2005-06-15 | 2006-12-28 | Canon Inc | Method, device, and recording medium for detecting heart boundary, chest boundary, and diaphergm boundary |
-
2007
- 2007-01-24 CN CNB2007100629899A patent/CN100545865C/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1418353A (en) * | 2000-01-18 | 2003-05-14 | 芝加哥大学 | Automated method and system for segmentation of lung regions in computed tomography scans |
| CN1388948A (en) * | 2000-09-07 | 2003-01-01 | 皇家菲利浦电子有限公司 | Segmentation of digital images |
| CN1711556A (en) * | 2002-11-13 | 2005-12-21 | 皇家飞利浦电子股份有限公司 | A method of and system for detecting uniform color segments |
| US20040120574A1 (en) * | 2002-12-18 | 2004-06-24 | Xerox Corporation | Systems and method for automatically choosing visual characteristics to highlight a target against a background |
| JP2006346465A (en) * | 2005-06-15 | 2006-12-28 | Canon Inc | Method, device, and recording medium for detecting heart boundary, chest boundary, and diaphergm boundary |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220358682A1 (en) * | 2020-09-16 | 2022-11-10 | China Tobacco Yunnan Industrial Co., Ltd | Device and method for detecting cigarette fly ash by gray-scale difference based on machine vision (mv) |
| US11670007B2 (en) * | 2020-09-16 | 2023-06-06 | China Tobacco Yunnan Industrial Co., Ltd | Device and method for detecting cigarette fly ash by gray-scale difference based on machine vision (MV) |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101231745A (en) | 2008-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100545865C (en) | A kind of automatic division method that image initial partitioning boundary is optimized | |
| CN110287932B (en) | Road blocking information extraction method based on deep learning image semantic segmentation | |
| CN110298321B (en) | Road blocking information extraction method based on deep learning image classification | |
| Farsadnia et al. | Identification of homogeneous regions for regionalization of watersheds by two-level self-organizing feature maps | |
| CN103098090B (en) | Multiparameter three-dimensional magnetic resonance imaging brain tumor partition method | |
| CN105930868A (en) | Low-resolution airport target detection method based on hierarchical reinforcement learning | |
| CN106096561A (en) | Infrared pedestrian detection method based on image block degree of depth learning characteristic | |
| CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN105975913B (en) | Road network extraction method based on adaptive cluster learning | |
| Moussa et al. | A new technique for automatic detection and parameters estimation of pavement crack | |
| CN101894275B (en) | Weakly supervised method for classifying SAR images | |
| CN107527351A (en) | A kind of fusion FCN and Threshold segmentation milking sow image partition method | |
| Sumer et al. | An adaptive fuzzy-genetic algorithm approach for building detection using high-resolution satellite images | |
| CN105260738A (en) | Method and system for detecting change of high-resolution remote sensing image based on active learning | |
| CN103400151A (en) | Optical remote-sensing image, GIS automatic registration and water body extraction integrated method | |
| CN104331716A (en) | SVM active learning classification algorithm for large-scale training data | |
| CN106340016A (en) | DNA quantitative analysis method based on cell microscope image | |
| CN103353986A (en) | Brain MR image segmentation method based on superpixel fuzzy clustering | |
| CN104346801A (en) | Image-composition evaluating device, information processing device and method thereof | |
| CN106611421A (en) | SAR image segmentation method based on feature learning and sketch line constraint | |
| CN111047603B (en) | Aerial image hybrid segmentation algorithm based on novel Markov random field and region combination | |
| CN109871875A (en) | A kind of building change detecting method based on deep learning | |
| CN104463885A (en) | Partition method for multiple-sclerosis damage area | |
| CN104573701A (en) | Automatic detection method of corn tassel traits |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090930 Termination date: 20210124 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |