CN107527350A - A kind of solid waste object segmentation methods towards visual signature degraded image - Google Patents
A kind of solid waste object segmentation methods towards visual signature degraded image Download PDFInfo
- Publication number
- CN107527350A CN107527350A CN201710559875.9A CN201710559875A CN107527350A CN 107527350 A CN107527350 A CN 107527350A CN 201710559875 A CN201710559875 A CN 201710559875A CN 107527350 A CN107527350 A CN 107527350A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- pixel
- super
- region
- 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.)
- Granted
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 41
- 239000002910 solid waste Substances 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 title claims abstract description 13
- 230000000007 visual effect Effects 0.000 title claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 239000012141 concentrate Substances 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 5
- 238000012804 iterative process Methods 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 5
- 241001269238 Data Species 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 239000003086 colorant Substances 0.000 claims description 2
- 238000003709 image segmentation Methods 0.000 abstract description 5
- 230000007850 degeneration Effects 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 2
- 238000000638 solvent extraction Methods 0.000 abstract 1
- 239000000428 dust Substances 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000008187 granular material Substances 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
- G06T5/30—Erosion or dilatation, e.g. thinning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20116—Active contour; Active surface; Snakes
-
- 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
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A kind of solid waste object segmentation methods suitable for visual signature degraded image, relate generally in the field such as robot vision and image segmentation.Because visual signature degeneration and solid waste object have adhesion and circumstance of occlusion, traditional images partitioning algorithm hardly results in high-precision segmentation result.The present invention obtains background model by depth background modeling, compares background model and solid waste point cloud to extract prospect mask.Local mask in extraction prospect mask, whole image segmentation problem is converted to multiple local mask segmentation problem.For local mask, split adhesion by fuzzy region extraction and block object, finally perform confusion region heavy label to obtain high-precision segmentation result.Segmentation precision of the present invention is high, can effectively split the solid waste object of severe color degeneration, and also very preferable for adhesion and the solid waste object blocked, segmentation effect.
Description
Technical field
The present invention relates to the solid waste of the technical fields, especially visual signature degraded image such as robot vision, image segmentation
Object Segmentation.
Background technology
Traditional image segmentation algorithm, use color and the feature of profile.But because industrial environment is more complicated:Transmission
Belt surface is covered by dust, and the dust granule of solid waste body surface result in serious visual signature and degenerate, and solid waste object is present
Adhesion and circumstance of occlusion.These can all split to two dimensional image has a great impact, so traditional image partition method is not
Suitable for industrial scene.
The content of the invention
Degenerated to solve existing visual signature, object adhesion and object block the problem of causing to split difficulty.The present invention
Provide the solid waste object segmentation methods under a kind of visual signature is degenerated seriously.The present invention obtains background by depth background modeling
Model, compare background model and solid waste point cloud to extract prospect mask.Mask is binary map, and interested pixel value is set to 255, its
After image element is set to 0.For the local mask connected in prospect mask, split by extracting fuzzy region, finally perform mould
Area's heavy label is pasted to obtain high-precision segmentation result.
The technical solution adopted for the present invention to solve the technical problems is:
A kind of solid waste object segmentation methods towards visual signature degraded image, the dividing method comprise the following steps:
1) background depth gauss hybrid models, are established by a series of depth information of background cloud datas, for image
In each pixel, modeled by Gaussian Mixture distribution, the probability distribution of the depth of a pixel equal to d is with formula (1.1)
To represent,
Wherein, wjIt is the weight of j-th of Gaussian Profile,K is the sum of Gaussian Profile, η (d;Θj) it is
J Gaussian Profile, represented with formula (1.2),
Wherein, μkIt is the average of k-th of Gaussian Profile, ∑kIt is the covariance matrix of k-th of Gaussian Profile, ∑k=σk 2I, I
It is unit matrix, σkIt is the standard deviation of k-th of Gaussian Profile, K Gaussian Profile is according to wk/σkSequence, B Gauss point before sequence
Cloth, obtained as background model, B by formula (1.3),
Wherein, T is minimum threshold value, for any one pixel in solid waste point cloud, finds the background model of correspondence position,
If the mean μ of all Gaussian Profiles in its depth value and background modelkPoor absolute value be more than standard deviation sigmakSetting times
Number, the pixel is by as foreground pixel;
2), by comparing background model and pending solid waste point cloud, in binary map, foreground pixel correspondence position is set to
255, background pixel is set to 0, obtains prospect mask, the local maskM connected in extraction prospect masklocal, and extract corresponding
Local RGB figure, local mask outline figure FcWith local depth edge figure Em, the segmentation problem of whole image is converted into
Multiple local mask segmentation problem;
3), for each local mask, super-pixel segmentation is carried out on corresponding local RGB image, obtains super-pixel
Set S={ s1,s2,s3,…,sn-1,sn, siA single super-pixel is represented, while is also a point set, by multiple features
Similar pixel composition;
4) local mask outline figure F, is passed throughcWith local edge graph Em, internal edge is obtained according to formula (4.1)
EinnerFigure,
Wherein,Represent in FcIt is upper to perform the expansive working that fuzzy core size is (2*k+1), pass through formula
(4.2) edge pixel collection E is extractedp,
Ep=p (x, y) | Einner(x, y)=255 }, (4.2)
Wherein, p (x, y) is to meet the pixel of condition, Einner(x, y) is EinnerThe upper y rows of figure, the pixel of xth row
Value;
5) super-pixel set S and edge pixel collection E, are passed throughp, edge super-pixel collection B is extracted according to formula (5.1)sp,
Wherein, p is any pixel, s in imagekIt is the super-pixel for the condition that meets, by BspThe super-pixel extraction of middle adjoining
Out borderline region B is defined as adjacent super-pixel set, each adjoining super-pixel collectionregion;
6), based on borderline region Bregion, confusion region is generated by an iteration, iterative process such as formula (6.1),
Wherein,It is BregionBorderline region after x expansion, expansion passes through merging to borderline region every time
Adjacent super-pixel is completed,It is by formula (6.2) expansion
Wherein,It is borderline regionAdjoining super-pixel collection, x is initially 0, and iteration, x add 1 each time,
By once or after successive ignition, MobjMultiple independent blocks can be divided into, for an independent block, set if it contains
Fixed number amount or more super-pixel, then it is the live part for forming an object to think it, otherwise it is assumed that being invalid portion
Point, when x is more than given threshold or MobjWhen possessing two pieces or more than two pieces separate live parts, iterative process is stopped
Only, the borderline region being calculated by formula (6.3)Confidence level,
Wherein,It is the borderline region ultimately generated, y is the number of ultimate bound zone broadening,RepresentThe number of pixels possessed, the ratio that borderline region accounts for local mask is bigger, and it turns into mould
It is smaller to paste the possibility in area, f=1 represents MobjContain two pieces or more than two pieces separate live parts;F=0 is represented
MobjWithout two pieces or more than two pieces separate live parts, ifMore than one threshold value C, this border
Region is just selected as confusion region, is adhesion and blocks the region that is difficult to differentiate between object, if a local mask does not have
Confusion region, then it is assumed that be single body;If there is confusion region, then Accurate Segmentation 7) is needed;
7) Accurate Segmentation, is realized by being allocated label to all pixels on local mask, carries out primary label
When, different label la are distributed to the pixel of different live parts, la={ 1,2,3 ... }, 0 is then distributed and gives confusion region and Mobj
The pixel of middle inactive portion;
8), for precise marking confusion region, the adjoining super-pixel collection of confusion region is extracted, according to the label of these super-pixel,
It is divided into two or more adjoining super-pixel collection, calculates the LAB colors and depth of each piece of super-pixel that super-pixel is concentrated
The average of degree, and the centre coordinate of super-pixel, for any one la=0 pixel, it and super picture are calculated by formula (8.1)
The diversity factor of element,
Wherein, dlabFor the Euclidean distance on LAB color spaces, ddepthFor the Euclidean distance of depth, dxyIt is image coordinate
Fasten the Euclidean distance of coordinate, wlab,wdepthAnd wxyIt is the weight of each distance, i is the sequence number that super-pixel concentrates super-pixel, is obtained
Into pixel and super-pixel set after the diversity factor of all super-pixel, pixel and adjacent super-pixel are calculated by formula (8.2)
Diversity factor between set,
D=min0 < i≤n(di), (8.2)
Wherein, i is the sequence number that super-pixel concentrates super-pixel, and n is the number that adjacent super-pixel concentrates super-pixel, for obtaining
D, d it is smaller represent pixel with abut super-pixel collection it is more similar, the label of most like super-pixel collection is distributed into the pixel, when
After la=0 all pixels heavy label terminates, local mask completes segmentation, and inspection result is with the presence or absence of isolated point or area
Domain, the optimization of segmentation result is realized by distributing label that most neighbor pixels possess.
The present invention technical concept be:Establish depth background model and carry out background subtraction, obtain prospect mask.Extraction prospect
Local mask in mask, whole image segmentation problem is converted to multiple local mask segmentation problem.Pass through confusion region
Extract to split local mask, and by confusion region heavy label, to obtain a high-precision segmentation result.
Beneficial effects of the present invention are mainly manifested in:Segmentation precision is high, can effectively split the solid waste of visual signature degeneration
Object, and it is also very preferable for adhesion and the solid waste object blocked, segmentation effect.It is finally based on the mark again of Pixel-level
Note, can obtain more accurate edge.
Brief description of the drawings
Fig. 1 is a local mask in prospect mask.
Fig. 2 is the border super-pixel of extraction.
Fig. 3 is a confusion region of extraction.
Fig. 4 is the result marked for the first time to local mask.
Fig. 5 is the adjoining super-pixel chosen, different according to the label of institute's band, can be divided into two set.Selection is adjacent super
Foundation of the set of pixels as heavy label.
Fig. 6 is the result of confusion region heavy label, and confusion region is divided into two parts, is belonging respectively to different objects, with different marks
Sign to represent.
Fig. 7 is the flow chart towards the solid waste object segmentation methods of visual signature degraded image.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings.
1~Fig. 7 of reference picture, a kind of solid waste object segmentation methods towards visual signature degraded image, comprises the following steps:
1) background depth gauss hybrid models, are established using a series of depth information of background cloud datas, for image
In each pixel, modeled by Gaussian Mixture distribution, the probability distribution of the depth of a pixel equal to d is with formula (1.1)
To represent,
Wherein, wjIt is the weight of j-th of Gaussian Profile,K is the sum of Gaussian Profile, and K is taken in the present invention
=5, and η (d;Θj) it is j-th of Gaussian Profile, represented with formula (1.2),
Wherein, μkIt is the average of k-th of Gaussian Profile, ∑kIt is the covariance matrix of k-th of Gaussian Profile, ∑k=σk 2I, I
It is unit matrix, σkIt is the standard deviation of k-th of Gaussian Profile, K Gaussian Profile is according to wk/σkSequence, B Gauss point before sequence
Cloth, obtained as background model, B by formula (1.3),
Wherein, T is minimum threshold value, for any one pixel in solid waste point cloud, finds the background model of correspondence position,
If the mean μ of all Gaussian Profiles in its depth value and background modelkPoor absolute value be more than standard deviation sigmakSetting times
Number (taking 2.5), the pixel is by as foreground pixel;
2), by comparing background model and pending solid waste point cloud, in binary map, foreground pixel correspondence position is set to
255, background pixel is set to 0, obtains prospect mask, the local maskM connected in extraction prospect masklocal, and extract corresponding
Local RGB figure, local mask outline figure FcWith local depth edge figure Em, the segmentation problem of whole image is converted into
Multiple local mask segmentation problem, figure one are example part mask;
3), for each local mask, super-pixel segmentation is carried out on corresponding local RGB image, obtains super-pixel
Set S={ s1,s2,s3,…,sn-1,sn, siA single super-pixel is represented, while is also a point set, by multiple features
Similar pixel composition;
4) local mask outline figure F, is passed throughcWith local edge graph Em, internal edge is obtained according to formula (4.1)
EinnerFigure,
Wherein,Represent in FcIt is upper to perform the expansive working that fuzzy core size is (2*k+1), pass through formula
(4.2) edge pixel collection E is extractedp,
Ep=p (x, y) | Einner(x, y)=255 }, (4.2)
Wherein, p (x, y) is to meet the pixel of condition, Einner(x, y) is EinnerThe upper y rows of figure, the pixel of xth row
Value;
5) it is, more mixed and disorderly due to obtaining the information that internal edge contains, and the missing at edge is there may be, it is necessary to enter one
Step processing, passes through super-pixel set S and edge point set Ep, edge super-pixel collection B is extracted according to formula (5.1)sp,
Wherein, p is any pixel, s in imagekIt is the super-pixel for the condition that meets, as shown in Fig. 2 BspMiddle super-pixel is
Internal edge is expanded, and remains the continuity of internal edge, shows as the adjoining of super-pixel, by BspThe super-pixel extraction of middle adjoining
Out borderline region B is defined as adjacent super-pixel set, each adjoining super-pixel collectionregion;
6), based on borderline region Bregion, confusion region is generated by an iteration, iterative process such as formula (6.1),
Wherein,It is BregionBorderline region after x expansion, expansion passes through merging to borderline region every time
Adjacent super-pixel is completed,It is by formula (6.2) expansion
Wherein,It is borderline regionAdjoining super-pixel collection, x is initially 0, and iteration, x add 1 each time,
By once or after successive ignition, MobjIndependent multiple pieces can be divided into, for an independent block, set if it contains
Fixed number amount (taking 7) is individual or more super-pixel, then it is the live part for forming an object to think it, otherwise it is assumed that being nothing
Part is imitated, when x is more than given threshold (taking 4) or MobjWhen possessing two pieces or more than two pieces separate live parts, repeatedly
Stop for process, the borderline region being calculated by formula (6.3)Confidence level,
Wherein,It is the borderline region ultimately generated, y is the number of ultimate bound zone broadening,RepresentThe number of pixels possessed, the ratio that borderline region accounts for local mask is bigger, and it turns into mould
It is smaller to paste the possibility in area, f=1 represents MobjContain two pieces or more than two pieces effective object parts;F=0 represents MobjNot yet
There are two pieces or more than two pieces effective object parts, ifMore than one threshold value C, the present invention in C=0.4, this
Individual borderline region is just selected as confusion region, is adhesion and blocks the region that is difficult to differentiate between object, if a part
Mask does not have confusion region, then it is assumed that is single body;If there is confusion region, as shown in figure 3, then needing Accurate Segmentation 7);
7), by being allocated label to all pixels on local mask to realize Accurate Segmentation, as shown in figure 4, just
During level mark, different label la are distributed to the pixel of different objects main body, la={ 1,2,3 ... }, 0 is then distributed and gives confusion region
And MobjIn inactive portion pixel;
8), for precise marking confusion region, the adjoining super-pixel collection of confusion region is extracted, as shown in figure 5, surpassing picture according to these
The label of element, it is divided into two or more adjoining super-pixel collection, calculates the LAB for each piece of super-pixel that super-pixel is concentrated
The average of color and depth, and the centre coordinate of super-pixel, for any one la=0 pixel, calculated by formula (8.1)
Its diversity factor with super-pixel,
Wherein, dlabFor the Euclidean distance on LAB color spaces, ddepthFor the Euclidean distance of depth, dxyIt is image coordinate
Fasten the Euclidean distance of coordinate, wlab,wdepthAnd wxyIt is the weight of each distance, W in the present inventionlab=4, wdepth=3, wxy=
3, i be the sequence number that super-pixel concentrates super-pixel, obtains pixel with after the diversity factor of all super-pixel in super-pixel set, passing through public affairs
Formula (8.2) calculates diversity factor between pixel and adjacent super-pixel set,
D=min0 < i≤n(di), (8.2)
Wherein, i is the sequence number that super-pixel concentrates super-pixel, and n is the number that adjacent super-pixel concentrates super-pixel, for obtaining
D, d it is smaller represent pixel with abut super-pixel collection it is more similar, so the label of most like super-pixel collection is distributed into the picture
Element, after la=0 all pixels heavy label terminates, as shown in fig. 6, confusion region is divided into two pieces, local mask completes segmentation,
Inspection result realizes segmentation result with the presence or absence of isolated point or region by distributing label that most neighbor pixels possess
Optimization.
In the present embodiment, using the method for confusion region extraction, there will be adhesion and the local mask blocked to separate, and obtains
Multiple effective object parts, carry out local mask first mark.Then by choosing the adjoining super-pixel of confusion region, according to
Label after first mark is different, is divided into two or more adjacent super-pixel set, according to pixel and adjacent super-pixel
The diversity factor of set, heavy label is carried out to non-classified pixel, to obtain high-precision segmentation result.
Claims (1)
1. a kind of solid waste object segmentation methods towards visual signature degraded image, the dividing method comprises the following steps:
1) background depth gauss hybrid models, are established by a series of depth information of background cloud datas, for every in image
One pixel, is modeled by Gaussian Mixture distribution, and the probability distribution of the depth of a pixel equal to d is with formula (1.1) come table
Show,
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>d</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</msubsup>
<msub>
<mi>w</mi>
<mi>j</mi>
</msub>
<mi>&eta;</mi>
<mrow>
<mo>(</mo>
<mi>d</mi>
<mo>;</mo>
<msub>
<mi>&Theta;</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1.1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, wjIt is the weight of j-th of Gaussian Profile,K is the sum of Gaussian Profile, η (d;Θj) it is j-th
Gaussian Profile, represented with formula (1.2),
<mrow>
<mi>&eta;</mi>
<mrow>
<mo>(</mo>
<mi>d</mi>
<mo>;</mo>
<msub>
<mi>&Theta;</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>&eta;</mi>
<mrow>
<mo>(</mo>
<mi>d</mi>
<mo>;</mo>
<msub>
<mi>&mu;</mi>
<mi>k</mi>
</msub>
<mo>,</mo>
<msub>
<mi>&Sigma;</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mrow>
<mn>2</mn>
<mi>&pi;</mi>
<mo>|</mo>
<msub>
<mi>&Sigma;</mi>
<mi>k</mi>
</msub>
<mo>|</mo>
</mrow>
</msqrt>
</mfrac>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<msup>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>-</mo>
<msub>
<mi>&mu;</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
<msup>
<msub>
<mi>&Sigma;</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>-</mo>
<msub>
<mi>&mu;</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msup>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1.2</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, μkIt is the average of k-th of Gaussian Profile, ∑kIt is the covariance matrix of k-th of Gaussian Profile, ∑k=σk 2I, I are single
Bit matrix, σkIt is the standard deviation of k-th of Gaussian Profile, K Gaussian Profile is according to wk/σkSequence, B Gaussian Profile, quilt before sequence
As background model, B is obtained by formula (1.3),
<mrow>
<mi>B</mi>
<mo>=</mo>
<mi>arg</mi>
<mi> </mi>
<msub>
<mi>min</mi>
<mi>b</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>b</mi>
</msubsup>
<msub>
<mi>w</mi>
<mi>j</mi>
</msub>
<mo>></mo>
<mi>T</mi>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1.3</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, T is minimum threshold value, for any one pixel in solid waste point cloud, finds the background model of correspondence position, if
The mean μ of all Gaussian Profiles in its depth value and background modelkPoor absolute value be more than standard deviation sigmakSetting multiple,
The pixel is by as foreground pixel;
2), by comparing background model and pending solid waste point cloud, in binary map, foreground pixel correspondence position is set to 255, the back of the body
Scene element is set to 0, obtains prospect mask, the local maskM connected in extraction prospect masklocal, and extract corresponding local
RGB schemes, local mask outline figure FcWith local depth edge figure Em, the segmentation problem of whole image is converted into multiple offices
Portion mask segmentation problem;
3), for each local mask, super-pixel segmentation is carried out on corresponding local RGB image, obtains super-pixel set S
={ s1,s2,s3,…,sn-1,sn, siA single super-pixel is represented, while is also a point set, as similar in multiple features
Pixel forms;
4) local mask outline figure F, is passed throughcWith local edge graph Em, internal edge E is obtained according to formula (4.1)innerFigure,
<mrow>
<msub>
<mi>E</mi>
<mrow>
<mi>i</mi>
<mi>n</mi>
<mi>n</mi>
<mi>e</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>E</mi>
<mi>m</mi>
</msub>
<mo>-</mo>
<msub>
<mi>F</mi>
<mi>c</mi>
</msub>
<mo>&CirclePlus;</mo>
<msub>
<mi>C</mi>
<mrow>
<mn>2</mn>
<mo>*</mo>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4.1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,Represent in FcIt is upper to perform the expansive working that fuzzy core size is (2*k+1), carried by formula (4.2)
Take edge pixel collection Ep,
Ep=p (x, y) | Einner(x, y)=255 }, (4.2)
Wherein, p (x, y) is to meet the pixel of condition, Einner(x, y) is EinnerThe upper y rows of figure, the pixel value of xth row;
5) super-pixel set S and edge pixel collection E, are passed throughp, edge super-pixel collection B is extracted according to formula (5.1)sp,
<mrow>
<msub>
<mi>B</mi>
<mrow>
<mi>s</mi>
<mi>p</mi>
</mrow>
</msub>
<mo>=</mo>
<mo>{</mo>
<mrow>
<msub>
<mi>s</mi>
<mi>k</mi>
</msub>
<mo>|</mo>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<msub>
<mi>s</mi>
<mi>k</mi>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>a</mi>
<mi>n</mi>
<mi>d</mi>
<mi> </mi>
<mi>p</mi>
<mo>&Element;</mo>
<msub>
<mi>E</mi>
<mi>p</mi>
</msub>
</mrow>
</mtd>
</mtr>
</mtable>
</mrow>
<mo>}</mo>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5.1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, p is any pixel, s in imagekIt is the super-pixel for the condition that meets, by BspThe super-pixel of middle adjoining extracts
As adjacent super-pixel set, each adjoining super-pixel collection is defined as borderline region Bregion;
6), based on borderline region Bregion, confusion region is generated by an iteration, iterative process such as formula (6.1),
<mrow>
<msub>
<mi>M</mi>
<mrow>
<mi>o</mi>
<mi>b</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>l</mi>
<mi>o</mi>
<mi>c</mi>
<mi>a</mi>
<mi>l</mi>
</mrow>
</msub>
<mo>-</mo>
<msup>
<msub>
<mi>B</mi>
<mrow>
<mi>r</mi>
<mi>e</mi>
<mi>g</mi>
<mi>i</mi>
<mi>o</mi>
<mi>n</mi>
</mrow>
</msub>
<msub>
<mi>x</mi>
<mrow>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
</msup>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6.1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,It is BregionBorderline region after x expansion, expansion passes through merging adjoining to borderline region every time
Super-pixel complete,It is by formula (6.2) expansion
<mrow>
<msup>
<msub>
<mi>B</mi>
<mrow>
<mi>r</mi>
<mi>e</mi>
<mi>g</mi>
<mi>i</mi>
<mi>o</mi>
<mi>n</mi>
</mrow>
</msub>
<msub>
<mrow>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
</msup>
<mo>=</mo>
<msup>
<msub>
<mi>B</mi>
<mrow>
<mi>r</mi>
<mi>e</mi>
<mi>g</mi>
<mi>i</mi>
<mi>o</mi>
<mi>n</mi>
</mrow>
</msub>
<msub>
<mi>x</mi>
<mrow>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
</msup>
<mo>&cup;</mo>
<msup>
<msub>
<mi>A</mi>
<mrow>
<mi>s</mi>
<mi>p</mi>
</mrow>
</msub>
<msub>
<mi>x</mi>
<mrow>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
</msup>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6.2</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,It is borderline regionAdjoining super-pixel collection, x is initially 0, each time iteration, and x adds 1, passes through
Once or after successive ignition, MobjMultiple independent blocks can be divided into, for an independent block, if it contains setting number
Amount or more super-pixel, then it is the live part for forming an object to think it, otherwise it is assumed that be inactive portion, when
X is more than given threshold or MobjWhen possessing two pieces or more than two pieces separate live parts, iterative process stops, and leads to
Cross the borderline region that formula (6.3) is calculatedConfidence level,
Wherein,It is the borderline region ultimately generated, y is the number of ultimate bound zone broadening,
RepresentThe number of pixels possessed, the ratio that borderline region accounts for local mask is bigger, and it turns into the possibility of confusion region
Smaller, f=1 represents MobjContain two pieces or more than two pieces separate live parts;F=0 represents MobjWithout two pieces or
More than two pieces separate live parts of person, ifMore than one threshold value C, this borderline region are just chosen
As confusion region, it is adhesion and blocks the region that is difficult to differentiate between object, if a local mask does not have confusion region, then it is assumed that
It is single body;If there is confusion region, then Accurate Segmentation 7) is needed;
7) Accurate Segmentation, is realized by being allocated label to all pixels on local mask, will when carrying out primary label
Different label la distribute to the pixel of different live parts, la={ 1,2,3 ... }, then distribute 0 and give confusion region and MobjIn it is invalid
Partial pixel;
8), for precise marking confusion region, the adjoining super-pixel collection of confusion region is extracted, according to the label of these super-pixel, is divided into
Two or more adjoining super-pixel collection, calculate the LAB colors and depth of each piece of super-pixel that super-pixel is concentrated
Average, and the centre coordinate of super-pixel, for any one la=0 pixel, it and super-pixel are calculated by formula (8.1)
Diversity factor,
<mrow>
<msub>
<mi>d</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>w</mi>
<mrow>
<mi>l</mi>
<mi>a</mi>
<mi>b</mi>
</mrow>
</msub>
<mo>*</mo>
<msub>
<mi>d</mi>
<mrow>
<mi>l</mi>
<mi>a</mi>
<mi>b</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>w</mi>
<mrow>
<mi>d</mi>
<mi>e</mi>
<mi>p</mi>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
<mo>*</mo>
<msub>
<mi>d</mi>
<mrow>
<mi>d</mi>
<mi>e</mi>
<mi>p</mi>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>w</mi>
<mrow>
<mi>x</mi>
<mi>y</mi>
</mrow>
</msub>
<mo>*</mo>
<msub>
<mi>d</mi>
<mrow>
<mi>x</mi>
<mi>y</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>w</mi>
<mrow>
<mi>l</mi>
<mi>a</mi>
<mi>b</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>w</mi>
<mrow>
<mi>d</mi>
<mi>e</mi>
<mi>p</mi>
<mi>t</mi>
<mi>h</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>w</mi>
<mrow>
<mi>x</mi>
<mi>y</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8.1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, dlabFor the Euclidean distance on LAB color spaces, ddepthFor the Euclidean distance of depth, dxyIt is that image coordinate fastens seat
Target Euclidean distance, wlab,wdepthAnd wxyIt is the weight of each distance, i is the sequence number that super-pixel concentrates super-pixel, obtains pixel
After the diversity factor of all super-pixel in super-pixel set, calculated by formula (8.2) between pixel and adjacent super-pixel set
Diversity factor,
< i≤n (the d of d=min 0i), (8.2)
Wherein, i is the sequence number that super-pixel concentrates super-pixel, and n is the number that adjacent super-pixel concentrates super-pixel, for obtained d,
D is smaller to represent that pixel is more similar to adjacent super-pixel collection, and the label of most like super-pixel collection is distributed into the pixel, works as la=
After 0 all pixels heavy label terminates, local mask completes segmentation, and inspection result is led to the presence or absence of isolated point or region
Cross and distribute label that most neighbor pixels possess to realize the optimization of segmentation result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710559875.9A CN107527350B (en) | 2017-07-11 | 2017-07-11 | A kind of solid waste object segmentation methods towards visual signature degraded image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710559875.9A CN107527350B (en) | 2017-07-11 | 2017-07-11 | A kind of solid waste object segmentation methods towards visual signature degraded image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107527350A true CN107527350A (en) | 2017-12-29 |
CN107527350B CN107527350B (en) | 2019-11-05 |
Family
ID=60748294
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710559875.9A Active CN107527350B (en) | 2017-07-11 | 2017-07-11 | A kind of solid waste object segmentation methods towards visual signature degraded image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107527350B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109146894A (en) * | 2018-08-07 | 2019-01-04 | 庄朝尹 | A kind of model area dividing method of three-dimensional modeling |
CN109409376A (en) * | 2018-11-05 | 2019-03-01 | 昆山紫东智能科技有限公司 | For the image partition method, terminal and storage medium of solid waste object |
CN109635809A (en) * | 2018-11-02 | 2019-04-16 | 浙江工业大学 | A kind of superpixel segmentation method towards vision degraded image |
CN110542908A (en) * | 2019-09-09 | 2019-12-06 | 阿尔法巴人工智能(深圳)有限公司 | laser radar dynamic object perception method applied to intelligent driving vehicle |
CN111105443A (en) * | 2019-12-26 | 2020-05-05 | 南京邮电大学 | Video group figure motion trajectory tracking method based on feature association |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150023575A1 (en) * | 2013-07-17 | 2015-01-22 | Siemens Medical Solutions Usa, Inc. | Anatomy Aware Articulated Registration for Image Segmentation |
CN104463870A (en) * | 2014-12-05 | 2015-03-25 | 中国科学院大学 | Image salient region detection method |
CN105957078A (en) * | 2016-04-27 | 2016-09-21 | 浙江万里学院 | Multi-view video segmentation method based on graph cut |
CN106056155A (en) * | 2016-05-30 | 2016-10-26 | 西安电子科技大学 | Super-pixel segmentation method based on boundary information fusion |
CN106886995A (en) * | 2017-01-13 | 2017-06-23 | 北京航空航天大学 | Polyteny example returns the notable object segmentation methods of image of device polymerization |
-
2017
- 2017-07-11 CN CN201710559875.9A patent/CN107527350B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150023575A1 (en) * | 2013-07-17 | 2015-01-22 | Siemens Medical Solutions Usa, Inc. | Anatomy Aware Articulated Registration for Image Segmentation |
CN104463870A (en) * | 2014-12-05 | 2015-03-25 | 中国科学院大学 | Image salient region detection method |
CN105957078A (en) * | 2016-04-27 | 2016-09-21 | 浙江万里学院 | Multi-view video segmentation method based on graph cut |
CN106056155A (en) * | 2016-05-30 | 2016-10-26 | 西安电子科技大学 | Super-pixel segmentation method based on boundary information fusion |
CN106886995A (en) * | 2017-01-13 | 2017-06-23 | 北京航空航天大学 | Polyteny example returns the notable object segmentation methods of image of device polymerization |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109146894A (en) * | 2018-08-07 | 2019-01-04 | 庄朝尹 | A kind of model area dividing method of three-dimensional modeling |
CN109635809A (en) * | 2018-11-02 | 2019-04-16 | 浙江工业大学 | A kind of superpixel segmentation method towards vision degraded image |
CN109635809B (en) * | 2018-11-02 | 2021-08-17 | 浙江工业大学 | Super-pixel segmentation method for visual degradation image |
CN109409376A (en) * | 2018-11-05 | 2019-03-01 | 昆山紫东智能科技有限公司 | For the image partition method, terminal and storage medium of solid waste object |
CN109409376B (en) * | 2018-11-05 | 2020-10-30 | 昆山紫东智能科技有限公司 | Image segmentation method for solid waste object, computer terminal and storage medium |
CN110542908A (en) * | 2019-09-09 | 2019-12-06 | 阿尔法巴人工智能(深圳)有限公司 | laser radar dynamic object perception method applied to intelligent driving vehicle |
CN110542908B (en) * | 2019-09-09 | 2023-04-25 | 深圳市海梁科技有限公司 | Laser radar dynamic object sensing method applied to intelligent driving vehicle |
CN111105443A (en) * | 2019-12-26 | 2020-05-05 | 南京邮电大学 | Video group figure motion trajectory tracking method based on feature association |
Also Published As
Publication number | Publication date |
---|---|
CN107527350B (en) | 2019-11-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107527350B (en) | A kind of solid waste object segmentation methods towards visual signature degraded image | |
CN103942794B (en) | A kind of image based on confidence level is collaborative scratches drawing method | |
CN104537676B (en) | Gradual image segmentation method based on online learning | |
CN104063702B (en) | Three-dimensional gait recognition based on shielding recovery and partial similarity matching | |
CN107610141A (en) | A kind of remote sensing images semantic segmentation method based on deep learning | |
CN102938066A (en) | Method for reconstructing outer outline polygon of building based on multivariate data | |
CN101770578B (en) | Image characteristic extraction method | |
CN105975974A (en) | ROI image extraction method in finger vein identification | |
CN104408733B (en) | Object random walk-based visual saliency detection method and system for remote sensing image | |
CN102147867B (en) | Method for identifying traditional Chinese painting images and calligraphy images based on subject | |
CN102622769A (en) | Multi-target tracking method by taking depth as leading clue under dynamic scene | |
CN103544685B (en) | A kind of image composition beautification method adjusted based on main body and system | |
CN102521597B (en) | Hierarchical strategy-based linear feature matching method for images | |
CN104913784B (en) | A kind of autonomous extracting method of planetary surface navigation characteristic | |
CN103793930A (en) | Pencil drawing image generation method and device | |
CN104517317A (en) | Three-dimensional reconstruction method of vehicle-borne infrared images | |
CN102799646B (en) | A kind of semantic object segmentation method towards multi-view point video | |
CN104299009A (en) | Plate number character recognition method based on multi-feature fusion | |
CN105374039A (en) | Monocular image depth information estimation method based on contour acuity | |
CN103971338A (en) | Variable-block image repair method based on saliency map | |
CN106937120A (en) | Object-based monitor video method for concentration | |
CN105469111A (en) | Small sample set object classification method on basis of improved MFA and transfer learning | |
CN103093470A (en) | Rapid multi-modal image synergy segmentation method with unrelated scale feature | |
CN104574307A (en) | Method for extracting primary colors of painting work image | |
CN104392462A (en) | SAR image registration method based on salient division sub-region pair |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20200611 Address after: Room 1504-2, Dikai International Center, Jianggan District, Hangzhou, Zhejiang Province Patentee after: HANGZHOU SHISHANG TECHNOLOGY Co.,Ltd. Address before: The city Zhaohui six districts Chao Wang Road Hangzhou City, Zhejiang province 310014 18 Patentee before: ZHEJIANG University OF TECHNOLOGY |