CN110097565A - Deep layer condition random field combination Linearization apparatus is to river SAR image split plot design - Google Patents
Deep layer condition random field combination Linearization apparatus is to river SAR image split plot design Download PDFInfo
- Publication number
- CN110097565A CN110097565A CN201910197015.4A CN201910197015A CN110097565A CN 110097565 A CN110097565 A CN 110097565A CN 201910197015 A CN201910197015 A CN 201910197015A CN 110097565 A CN110097565 A CN 110097565A
- Authority
- CN
- China
- Prior art keywords
- pixel
- river
- discrete
- linearization apparatus
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013461 design Methods 0.000 title claims abstract description 14
- 230000011218 segmentation Effects 0.000 claims abstract description 43
- 238000010606 normalization Methods 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 5
- 238000000605 extraction Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 238000000034 method Methods 0.000 description 31
- 230000000694 effects Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 230000000052 comparative effect Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 235000013399 edible fruits Nutrition 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000003709 image segmentation Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 238000011524 similarity measure Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 210000001210 retinal vessel Anatomy 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- G06T5/90—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- 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
- G06T2207/30181—Earth observation
Abstract
A kind of deep layer condition random field combination Linearization apparatus includes the following steps: (1) input picture to river SAR image split plot design;(2) rough Threshold segmentation;(3) image normalization establishes soft label (4) and establishes more size discrete sliding windows;(5) discrete group and Linearization apparatus are established as penalty factor;(6) judge similitude;(7) the soft label of iteration;(8) it extracts;(9) computational accuracy.The present invention can reduce that common sliding window bring is excessively smooth and bring the reciprocation between remote pixel, and the interference that the introducing of Linearization apparatus reduces identification target to noise and pseudo- river has good resistance by discrete sliding window.
Description
Technical field
The invention belongs to field of image processing, in particular to a kind of deep layer condition random field combination Linearization apparatus is to river SAR
Image segmentation.
Background technique
SAR radar satellite is a kind of active observation system over the ground, mountable in flying platforms such as aircraft, satellites, entirely
It when, it is round-the-clock over the ground implement observe and have certain ground penetrating ability.Therefore, SAR system is in disaster monitoring, environment
There is unique advantage in many applications such as monitoring, marine monitoring and military aspect.Information above SAR image is ground object target
The image information that reflection to radar beam, the mainly back scattering of ground object target are formed, for distribution objectives, it is understood that there may be
A large amount of speckle noise, reduce the spatial resolution of image, obscured the marginal information of image, so that the essence of interpretation of images
Degree reduces.
Tae-Jung Kown et al. 2013 in IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE
" the ETVOS:An Enhanced Total Variation Optimization that SENSING, VOL.51, NO.2 are delivered above
ETVOS algorithm is referred in Segmentation Approach for SAR Sea-Ice Image Segmentation ", it should
Method, in the entire variable optimizing phase, passes through the pact of minimum entire variable to divide various ocean ice types including two stages
Beam goes to estimate by improving the total variate model of Rudin-Osher-Fatemi and implementing it iteration from the steady state of non-segmentation
The steady state of segmentation.The sorting phase in Finite mixture model is logical using expectation maximization method is executed based on pixel distribution
It is finally similar right to cross gauss hybrid models estimation, the entire variable optimizing phase is then estimated using maximum likelihood classification technology
The classification of each pixel in product.
Linlin Xu et al. is in 2016 in IEEE JOURNAL OF SELECTED TOPICS IN APPLIED
" Fully Connected has been delivered above EARTH OBSERVATIONS AND REMOTE SENSING, VOL.9, NO.7
Continuous Conditional Random Field With Stochastic Cliques for Dark-Spot
Detection In SAR Imagery".It designs and constructs random full-mesh condition of continuity random field (SFCCRF) method
It establishes SAR image and carries out soft label reasoning, to realize a kind of effective detection algorithm.SFCCRF be based on they
The degree of approach in feature space and image space determines the connectivity of two pixels in a random basis.Since SFCCRF is extensive
The modeling of space correlation effect provides a kind of effective method, therefore obtained soft label can resist the influence of speckle noise
And the difference between prominent dim spot and background.Binaryzation is carried out by the soft label estimated SFCCRF to realize that dim spot detects.
Although random full-mesh condition of continuity random field (SFCCRF) method utilizes the influence of soft label speckle noise, this method is still
Have unfortunately: applying for extensive space has paid very big cost on the calculating time, can not carry out to dash area
Processing well, it is serious to weaken edge and detail section using window smoothing computation.
Summary of the invention
The purpose of the present invention is in view of the above shortcomings of the prior art, and provide a kind of deep layer condition random field combine it is linear
Device can be applied to the fields such as the extraction, segmentation and target following of image to river SAR image split plot design, this method, be containing making an uproar
The dividing method of the image of sound reduces the erroneous segmentation of pseudo- shade by joined Linearization apparatus, is subtracted by the application of discrete window
The phenomenon that few smoothed image, further improve nicety of grading.
As above design, the technical scheme is that a kind of deep layer condition random field combination Linearization apparatus is to river SAR image
Split plot design, characterized by the following steps:
(1) picture is inputted: one river synthetic aperture radar sar image to be split of input;
(2) rough Threshold segmentation: rough cutting operation is first carried out to river sar image using threshold value, after segmentation
Image be divided into river region, pseudo- river region and non-river region, to river region and pseudo- river region all labelled 1,
Labelled for non-river region 0;
(3) image is normalized and establishes soft label:
(3a) carries out image normalization processing to original image, and value range is between [1,2];
(3b) takes with the pixel value after image normalization as soft label;
(4) the discrete sliding window of more sizes is established:
(4a) discrete window of various sizes of n*n is established above in rough segmented image;
(4b) passes through the label divided roughly that step (2) obtain, then by judging whether window interior label value is identical,
Then the discrete pixel calculated is participated in required for judging;
(4c) reformulates a discrete sliding window by these discrete pixels;
(5) discrete group and penalty factor Linearization apparatus are established:
(5a) judges that the similitude between pixel, determination range are not limited solely to adjacent pixel by similar function
Point, but the similitude between pixel is judged in image different range by adjusting window size, these similar pixels
Form random group;
(5b) takes 24 different angles linear as large scale centered on a pixel in the window of 15*15
Device, then the small size Linearization apparatus of 24 different angles is established in the window of 9*9, the Linearization apparatus of two different scales combines conduct
Penalty factor;
(6) judge similitude:
(6a) judges the continuity between soft label and pixel value by unitary potential function;
(6b) judges the similitude between the soft label an of pixel and the soft label of other pixels by binary potential function;
(6c), as penalty factor, judges noise and pseudo- target by Linearization apparatus;
(7) the soft label of iteration:
The soft label in condition random field formula being made of continuous iteration unitary potential function and binary potential function,
Reach a stable value;
(8) it extracts: stablizing soft label by what last iteration obtained, compared with segmentation threshold, obtain segmentation knot to the end
Fruit figure.
(9) computational accuracy:
(9a) is calculated identical with ground truth class label in the river target of extraction by ground truth
Number, calculate its percentage in the object pixel number of extraction;
(9b) calculates ground truth identical number of class label in the target of extraction, then calculate again its
Percentage in ground truth.
Threshold segmentation formula described in the step (2) is as follows:
Thrd=mean (S)-ε * std (S)
The wherein soft tally set of S representative image, mean calculate the average value of the soft label of image, and std calculates the soft label of image
Standard deviation, ε are a coefficients, and experience value is that 1, thrd is the threshold value obtained.
The step (5a) judges whether two center pixels are similar by Similarity measures, if between two pixels
Similar value between [0,1], just include another pixel in the group centered on one of pixel, otherwise with
Group centered on one of pixel does not then include that another pixel judges that the function of similitude is as follows:
dijIt indicates to judge the similitude between two pixels by calculating two discrete windows,Represent with i as
The average value of image pixel intensities in the discrete window of center pixel,Similarly, var { Wi+WjIt is pooled variance, calculation method is
Formula (3), WiRepresent the sum of the intensity of all pixels point in discrete window, WjSimilarly, niRepresent pixel in this discrete window
Total number, njSimilarly;
Formula (7) passes throughIt is similar to want to judge, discrete group is established, it is 0.1 that γ, which takes empirical value,Value range be [0,
1]。
The judgment formula of step (5b) Linearization apparatus are as follows:
For the average value of all pixels in large scale Linearization apparatus,For " winner line " wired middle gray scale
Value is maximum, and wherein W represents the window size of large scale Linearization apparatus, and L represents the Linearization apparatus window size of small size.
Unitary potential function described in the step (6a) is
In formula, xiRepresent the pixel value of i point, siThe soft label of i point is represented, L is equivalent number,For unitary gesture letter
Number.
Binary potential function described in the step (6b)
PijIt is distributed by gamma and calculates pixel i, the similitude of j, si, sjIt is i, the soft label of pixel of two points of j, NiIt is
Discrete Stochastic group,For binary potential function.
The present invention has the advantages that compared with prior art
1, due to present invention employs discrete window method, reducing the smooth phenomenon due to caused by the interference of noise.
2, present invention uses the methods that Linearization apparatus is combined with condition random field, during carrying out river extraction,
Since the addition of Linearization apparatus reduces the interference of pseudo- river and noise, extraction accuracy is improved.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention;
Fig. 2 (a), 2 (b) are the schematic diagram for establishing random group and multiple dimensioned Linearization apparatus respectively;
Fig. 3 is the classification results comparison diagram with three width SAR images in the present invention and existing method.
Specific embodiment
A kind of deep layer condition random field combination Linearization apparatus includes the following steps: river SAR image split plot design
(1) a width SAR picture is inputted: arbitrarily one river synthetic aperture radar SAR image to be split of input;
(2) rough Threshold segmentation:
Using threshold value rough cutting operation advanced to river SAR image, the image after dividing be divided into river region,
Pseudo- river region and non-river region stick mark for non-river region to river region and pseudo- river region all labelled 1
Label 0;Threshold decision formula is as follows:
Thrd=mean (S)-ε * std (S) (1)
The wherein soft tally set of S representative image, mean calculate the average value of the soft label of image, and std calculates the soft label of image
Standard deviation, ε are a coefficients, and experience value is 1.
(3) image normalization:
(3a) carries out image normalization to original image, and value range prevents the case where denominator is 0 between [1,2],
It is because avoiding the problem that subsequent operation is made troubles;
(3b) takes and pixel value identical after image normalization, referred to as soft label;
(4) sliding window of more size discretes is established:
The interference of (4a) due to SAR image by noise can not calculate single pixel point, still, directly take n*n
Window can cause smooth puzzlement to image, therefore the discrete window of various sizes of n*n is established above in rough segmented image
Mouthful;
(4b) passes through the label divided roughly that step (2) obtain, then by judging whether window interior label value is identical,
Then judge the pixel for participating in calculating;
(4c) reformulates a discrete sliding window by these discrete pixels;
(5) discrete group and the more dimension linearity devices of penalty factor are established
(5a) judges whether two center pixels are similar by Similarity measures, if the similar value between two pixels
It just include another pixel in the group centered on one of pixel, otherwise with one of them between [0,1]
Group centered on pixel does not then include that another pixel judges that the function of similitude is as follows:
dijIt indicates to judge the similitude between two pixels by calculating two discrete windows,Represent with i as
The average value of image pixel intensities in the discrete window of center pixel,Similarly, var { Wi+WjIt is pooled variance, calculation method is
Formula (3), WiRepresent the sum of the intensity of all pixels point in discrete window, WjSimilarly, niRepresent pixel in this discrete window
Total number, njSimilarly;
Formula (7) passes throughIt is similar to want to judge, discrete group is established, it is 0.1 that γ, which takes empirical value,Value range be [0,
1]。
(5b) is taking 24 different angles (between every antenna segment centered on a pixel i in the window of 15*15
Interval degree be 15 °), as large scale Linearization apparatus.The small size for establishing 24 different angles in the window of 9*9 again is linear
Device.The Linearization apparatus of two different scales is combined as penalty factor.The judgment formula of Linearization apparatus is as follows:
For the average value of all pixels in large scale Linearization apparatus,For " winner line " wired middle gray scale
Value is maximum, and W is large scale Linearization apparatus window size, and L is the size of small size Linearization apparatus window, passes throughEnd value, to i
Point pixel judge whether it being noise or pseudo- shade.
Referring to fig. 2 (a): wherein y0It is central pixel point, y1-y9It is y0Equidistant pixel, wherein y9、y5It is y0It is related
Pixel, y10、y11Distance y0Although distance it is remoter than other pixels, and y0But there is bigger similitude.In order to
It is enough more accurately to find pixel relevant to center pixel, so the discrete of group the inside is found in the space for taking different range
Point.
Referring to fig. 2 (b): what is wherein formed in green box is the Linearization apparatus of small size, what the endpoint of red lines was formed
Box is large-sized Linearization apparatus.
(6) judge similitude:
(6a) promotes the continuity between soft label and pixel value by unitary potential function
Wherein, xiRepresent the pixel value of i point, siThe soft label of i point is represented, L is equivalent number,For unitary potential function.
(6b) judges the similitude between the soft label an of pixel and the soft label of other pixels by binary potential function.
Wherein, PijIt is distributed by gamma and calculates pixel i, the similitude of j, si, sjIt is i, the soft mark of pixel of two points of j
Label, NiIt is that Discrete Stochastic group is judged by formula (7),For binary potential function.
(7) the soft label of iteration: the condition random field being made of continuous iteration unitary potential function and binary potential function
Soft label in formula reaches a stable value.
(8) it extracts: stablizing soft label by what last iteration obtained, compared with segmentation threshold, obtain segmentation knot to the end
Fruit figure.
(9) computational accuracy:
(9a) is calculated identical with ground truth class label in the river target of extraction by ground truth
Number, calculate its percentage in the object pixel number of extraction;Commission error (CE) misclassification error: CE is anti-
The ratio of the erroneous segmentation in all Target Segmentations is answered.
(9b) calculates ground truth identical number of class label in the target of extraction, then calculate again its
Percentage in ground truth.Omission error (OE) leaks point error: OE has reacted the omission divided and has really schemed
As the ratio of target.
(9c) averaged error (AE) overall accuracy: AE provides the balancing evaluation to segmentation ability.
Above-mentioned formula (1) is referring to bibliography 1, formula (2) referring to bibliography 2, formula (3) referring to bibliography 2, public affairs
Formula (4) referring to bibliography 3, formula (5) referring to bibliography 1, formula (6) referring to bibliography 1, formula (7) referring to reference
Document 1.
The present invention can be verified by following emulation experiment
1, simulated conditions
Effect of the invention can be further illustrated by following emulation experiment:
Emulation experiment hardware platform are as follows:
Intel (R) Core (TM) i7-6700CPU, 3.40GHz, 16GB RAM
Software platform: using Matlab R2016a, wherein
Matlab R2016a: it is the business mathematics software that MathWorks company, the U.S. produces, is used for algorithm development, data
Visualization, data analysis and numerical value calculate;
2, emulation content:
This method is applied in emulation 1., and the method that the condition random field that deep layer is rolled into a ball at random is combined with Linearization apparatus is respectively to a width
SAR image is split extraction, and wherein Fig. 3 (a) is SAR image to be split, and size 256*256, Fig. 3 (d) are threshold value point
The segmentation result of segmentation method, 3 (g) be the full method segmentation result for connecting discrete conditions random field, and Fig. 3 (j), 3 (m) are respectively
Segmentation result of the invention.
From Fig. 3 (j), 3 (m) as it can be seen that the present invention to the segmentation result of the first width SAR image are as follows:
Black region is that river extracts region, and white area is background area, shows this method to making an uproar by comparative test
The interference of sound, smooth phenomenon have more reductions, slightly better than full connection discrete conditions random field.
This method is applied in emulation 2., and full discrete conditions random field, the method for Threshold segmentation of connecting is respectively to a width SAR image
It is split extraction, wherein Fig. 3 (b) is SAR image to be split, and size 256*256, Fig. 3 (e) are threshold segmentation method
Segmentation result, 3 (h) be the full method segmentation result for connecting discrete conditions random field, and Fig. 3 (k), 3 (n) are segmentation knot of the invention
Fruit.
From Fig. 3 (k), 3 (n) as it can be seen that the present invention to the segmentation result of the second width SAR image are as follows: black region mentions for river
Region is taken, white area is background area, shows that this method has subtract to the interference of noise, smooth phenomenon more by comparative test
It is few, slightly better than full connection discrete conditions random field.
This method is applied in emulation 3., and full discrete conditions random field, the method for Threshold segmentation of connecting is respectively to a width SAR image
It is split extraction, wherein Fig. 3 (c) is SAR image to be split, and size 256*256, Fig. 3 (f) are threshold segmentation method
Segmentation result, 3 (i) be the full method segmentation result for connecting discrete conditions random field, and Fig. 3 (l), 3 (o) are segmentation knot of the invention
Fruit.
From Fig. 3 (l), 3 (o) as it can be seen that the present invention to the segmentation result of third width SAR image are as follows: black region mentions for river
Region is taken, white area is background area, shows that this method has subtract to the interference of noise, smooth phenomenon more by comparative test
It is few, slightly better than full connection discrete conditions random field.
Table 1 is to three kinds of methods to the comprehensive evaluation form of three width SAR image segmentation results
As seen from Table 1, the present invention to the segmentation result of three width SAR images by OE (leakage point error), (mistake divides mistake to CE
Difference), the comparison of AE (overall accuracy), be superior to comparative test, illustrate that this method overcomes the noise jamming of SAR image and smooth
Window bring crosses smoothing problasm.
Find out from three emulation experiments and data comparison, the present invention rolls into a ball and Linearization apparatus at random due to using deep layer, compares
The segmentation precision of classical Threshold segmentation and SFCCRF method is higher, so, using the present invention, synthetic aperture radar SAR is schemed
When picture is split, segmentation effect effect increases, and experiment further demonstrates effect of the invention.
Bibliography:
[1].Xu L,Shafiee M J,Wong A,et al.Fully connected continuous
conditional random field with stochastic cliques for dark-spot detection in
SAR imagery[J].IEEE Journal of Selected Topics in Applied Earth Observations
and Remote Sensing,2016,9(7):2882-2890.
[2].Yang W,Chen L,Dai D,et al.Semantic labelling of SAR images with
conditional random fields on region adjacency graph[J].IET radar,sonar &
navigation,2011,5(8):835-841.
[3].Nguyen U T V,Bhuiyan A,Park L A F,et al.An effective retinal
blood vessel segmentation method using multi-scale line detection[J]. Pattern
recognition,2013,46(3):703-715。
Claims (6)
1. a kind of deep layer condition random field combination Linearization apparatus is to river SAR image split plot design, it is characterised in that: including walking as follows
It is rapid:
(1) picture is inputted: one river synthetic aperture radar sar image to be split of input;
(2) rough Threshold segmentation: rough cutting operation is first carried out to river sar image using threshold value, the figure after dividing
As being divided into river region, pseudo- river region and non-river region, to river region and pseudo- river region all labelled 1, for
Non- river region labelled 0;
(3) image is normalized and establishes soft label:
(3a) carries out image normalization processing to original image, and value range is between [1,2];
(3b) takes with the pixel value after image normalization as soft label;
(4) the discrete sliding window of more sizes is established:
(4a) discrete window of various sizes of n*n is established above in rough segmented image;
(4b) passes through the label divided roughly that step (2) obtain, then by judging whether window interior label value is identical, then
The discrete pixel calculated is participated in required for judging;
(4c) reformulates a discrete sliding window by these discrete pixels;
(5) discrete group and penalty factor Linearization apparatus are established:
(5a) judges that the similitude between pixel, determination range are not limited solely to adjacent pixel by similar function,
But the similitude between pixel is judged in image different range by adjusting window size, these similar pixels are formed
Random group;
(5b) takes 24 different angles as large scale Linearization apparatus centered on a pixel in the window of 15*15, then
The small size Linearization apparatus of 24 different angles is established in the window of 9*9, the Linearization apparatus of two different scales is combined as punishment
The factor;
(6) judge similitude:
(6a) judges the continuity between soft label and pixel value by unitary potential function;
(6b) judges the similitude between the soft label an of pixel and the soft label of other pixels by binary potential function;
(6c), as penalty factor, judges noise and pseudo- target by Linearization apparatus;
(7) the soft label of iteration:
The soft label in condition random field formula being made of continuous iteration unitary potential function and binary potential function, makes it
Reach a stable value;
(8) it extracts: stablizing soft label by what last iteration obtained, compared with segmentation threshold, obtain segmentation result figure to the end.
(9) computational accuracy:
(9a) calculates in the river target of extraction identical with ground truth class label by ground truth
Number, calculates its percentage in the object pixel number of extraction;
(9b) calculates ground truth identical number of class label in the target of extraction, then calculate again its
Percentage in ground truth.
2. deep layer condition random field combination Linearization apparatus according to claim 1 exists to river SAR image split plot design, feature
In: Threshold segmentation formula described in the step (2) is as follows:
Thrd=mean (S)-ε * std (S)
The wherein soft tally set of S representative image, mean calculate the average value of the soft label of image, and std calculates the standard of the soft label of image
Deviation, ε are a coefficients, and experience value is that 1, thrd is the threshold value obtained.
3. deep layer condition random field combination Linearization apparatus according to claim 1 exists to river SAR image split plot design, feature
In: the formula of the pixel similarity of two points of calculating described in the step (5a):
dijIt indicates to judge the similitude between two pixels by calculating two discrete windows,Represent the picture centered on i
The average value of image pixel intensities in the discrete window of element,Similarly, var { Wi+WjIt is pooled variance, calculation method is formula
(3),WiRepresent the sum of the intensity of all pixels point in discrete window, WjSimilarly, niRepresent the total of pixel in this discrete window
Number, njSimilarly.
4. deep layer condition random field combination Linearization apparatus according to claim 1 exists to river SAR image split plot design, feature
In: the judgment formula of step (5b) Linearization apparatus are as follows:
For the average value of all pixels in large scale Linearization apparatus,Most for " winner line " wired middle gray value
Greatly, wherein W represents the window size of large scale Linearization apparatus, and L represents the Linearization apparatus window size of small size.
5. deep layer condition random field combination Linearization apparatus according to claim 1 exists to river SAR image split plot design, feature
In: unitary potential function described in the step (6a) are as follows:
In formula, xiRepresent the pixel value of i point, siThe soft label of i point is represented, L is equivalent number,For unitary potential function.
6. deep layer condition random field combination Linearization apparatus according to claim 1 exists to river SAR image split plot design, feature
In: binary potential function described in the step (6b) are as follows:
PijIt is distributed by gamma and calculates pixel i, the similitude of j, si, sjIt is i, the soft label of pixel of two points of j, NiIt is discrete
Random group,For binary potential function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910197015.4A CN110097565A (en) | 2019-03-15 | 2019-03-15 | Deep layer condition random field combination Linearization apparatus is to river SAR image split plot design |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910197015.4A CN110097565A (en) | 2019-03-15 | 2019-03-15 | Deep layer condition random field combination Linearization apparatus is to river SAR image split plot design |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110097565A true CN110097565A (en) | 2019-08-06 |
Family
ID=67444041
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910197015.4A Pending CN110097565A (en) | 2019-03-15 | 2019-03-15 | Deep layer condition random field combination Linearization apparatus is to river SAR image split plot design |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110097565A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111191393A (en) * | 2019-12-20 | 2020-05-22 | 河海大学 | Method and device for evaluating non-determinacy of hydrodynamic landslide based on discrete cosine transform |
-
2019
- 2019-03-15 CN CN201910197015.4A patent/CN110097565A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111191393A (en) * | 2019-12-20 | 2020-05-22 | 河海大学 | Method and device for evaluating non-determinacy of hydrodynamic landslide based on discrete cosine transform |
CN111191393B (en) * | 2019-12-20 | 2022-02-11 | 河海大学 | Method and device for evaluating non-determinacy of hydrodynamic landslide based on discrete cosine transform |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Feng et al. | Multiphase SAR image segmentation with $ G^{0} $-statistical-model-based active contours | |
Shen et al. | Detection of stored-grain insects using deep learning | |
US20070223815A1 (en) | Feature Weighted Medical Object Contouring Using Distance Coordinates | |
US20150347871A1 (en) | Data fusion analysis for maritime automatic target recognition | |
CN109446894B (en) | Multispectral image change detection method based on probability segmentation and Gaussian mixture clustering | |
Choong et al. | Image segmentation via normalised cuts and clustering algorithm | |
CN102903102A (en) | Non-local-based triple Markov random field synthetic aperture radar (SAR) image segmentation method | |
CN106611422B (en) | Stochastic gradient Bayes's SAR image segmentation method based on sketch structure | |
CN108805057B (en) | SAR image reservoir area detection method based on joint significance analysis | |
CN106991686A (en) | A kind of level set contour tracing method based on super-pixel optical flow field | |
CN109712149A (en) | A kind of image partition method based on wavelet energy and fuzzy C-mean algorithm | |
US20210256717A1 (en) | Edge-guided ranking loss for monocular depth prediction | |
CN113362293A (en) | SAR image ship target rapid detection method based on significance | |
Chen et al. | Underwater sonar image segmentation combining pixel-level and region-level information | |
Wei et al. | Interval type-2 possibilistic fuzzy clustering noisy image segmentation algorithm with adaptive spatial constraints and local feature weighting & clustering weighting | |
Bryner et al. | Elastic shapes models for improving segmentation of object boundaries in synthetic aperture sonar images | |
CN110097565A (en) | Deep layer condition random field combination Linearization apparatus is to river SAR image split plot design | |
Fernández-Michelli et al. | Unsupervised classification algorithm based on EM method for polarimetric SAR images | |
US10832413B2 (en) | Curvilinear object segmentation with geometric priors | |
Chen et al. | Segmenting oil spills from blurry images based on alternating direction method of multipliers | |
Krylov et al. | Modeling the statistics of high resolution SAR images | |
Franchi et al. | Bagging stochastic watershed on natural color image segmentation | |
Al-Bayati et al. | Automatic thresholding techniques for SAR images | |
Wu et al. | Adaptive active contour model driven by image data field for image segmentation with flexible initialization | |
Chhabra et al. | Comparison of different edge detection techniques to improve quality of medical images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190806 |