CN106570498B - Salient region detecting method and detection system - Google Patents
Salient region detecting method and detection system Download PDFInfo
- Publication number
- CN106570498B CN106570498B CN201610889100.3A CN201610889100A CN106570498B CN 106570498 B CN106570498 B CN 106570498B CN 201610889100 A CN201610889100 A CN 201610889100A CN 106570498 B CN106570498 B CN 106570498B
- Authority
- CN
- China
- Prior art keywords
- super
- pixel
- input picture
- image
- notable
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000001514 detection method Methods 0.000 title claims abstract description 25
- 230000011218 segmentation Effects 0.000 claims abstract description 18
- 238000009792 diffusion process Methods 0.000 claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims description 14
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 3
- 230000005764 inhibitory process Effects 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 10
- 230000000007 visual effect Effects 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000009432 framing Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The present invention relates to a kind of salient region detecting method and detection systems.Wherein, this method includes carrying out super-pixel segmentation to input picture, obtains image superpixel processing unit；Then, it is based on image superpixel processing unit, measures the topological background confidence level of input picture；Then, the topological background confidence level based on input picture calculates the contrast in color of image and spatial position feature, obtains image primary saliency value；It recycles compactedness method of diffusion to correct image primary saliency value, obtains compactedness notable figure；And topological background confidence level and compactedness notable figure using input picture, saliency value assignment is carried out, image single scale notable figure is obtained；Finally, the method using Multiscale Fusion handles image single scale notable figure, scale notable figure is obtained, to obtain image obvious object region.By using above-mentioned technical proposal, solve how fast and effeciently abstract image marking area the technical issues of.
Description
Technical field
The present embodiments relate to technical field of image processing, more particularly, to a kind of salient region detecting method and detection
System.
Background technique
Human visual system has attention function, can rapidly lock key area, and only selection enters the visual field
A portion information is for subsequent deeper information extraction and processing, to obtain efficient visual perception.Inspire in
The conspicuousness detection model of human visual attention mechanism can quickly and efficiently extract useful information for computer system realization
Solution route is provided.The cognitive theory and physiological models that view-based access control model pays attention to, the researcher of computer field propose very much
Conspicuousness detection model.Vision significance detects the pretreatment process as visual task, has boundless application prospect,
It can be applied to many different high-level complicated visual processes tasks, such as object detection, visual target tracking, object are known
Not, compression of images and robot cognition etc..
In recent years, the obvious object detection method much based on frame background priori is constantly proposed that they are obtained
Good detection effect.Such as:
Xiaohui Li,Huchuan Lu,Lihe Zhang,Xiang Ruan,and Ming-Hsuan
Yang.Saliency detection via dense and sparse reconstruction.In IEEE
International Conference on Computer Vision(ICCV),pages2976–2983,2013；
Wangjiang Zhu,Shuang Liang,YichenWei,and Jian Sun.Saliency
optimization from robust background detection.In IEEE Conference on Computer
Vision and Pattern Recognition(CVPR),pages 2814–2821,2014；
Chuan Yang,Lihe Zhang,Huchuan Lu,Xiang Ruan,and Ming-Hsuan
Yang.Saliency detection via graph-based manifold ranking.In IEEE Conference
on Computer Vision and Pattern Recognition(CVPR),pages 3166–3173,2013。
However, these methods are during inferring saliency value using the topological relation of image-region and frame region, it is past
It is past to be more heavily weighted toward global conspicuousness.By it is existing based on the conspicuousness detection method of frame background priori for, will lead to may deposit
In following problem: when obvious object region is similar to framing mask region, the local conspicuousness of obvious object is mistakenly pressed down
System；When image background regions and framing mask region are extremely dissimilar, the conspicuousness of background area may be exaggerated by mistake, thus
Generate the conspicuousness testing result of mistake.
In view of this, the present invention is specifically proposed.
Summary of the invention
How fast and effeciently the embodiment of the present invention provides a kind of salient region detecting method, significant to solve abstract image
The technical issues of region.In addition, the embodiment of the present invention also provides a kind of marking area detection system.
To achieve the goals above, according to an aspect of the present invention, the following technical schemes are provided:
A kind of salient region detecting method, the method may include:
Super-pixel segmentation is carried out to input picture, obtains image superpixel processing unit；
Based on described image super-pixel processing unit, the topological background confidence level of the input picture is measured；
The topological background confidence level based on the input picture, calculates pair in color of image and spatial position feature
Than degree, image primary saliency value is obtained；
Described image primary saliency value is corrected using compactedness method of diffusion, obtains compactedness notable figure；
Using the topological background confidence level of the input picture and the compactedness notable figure, saliency value tax is carried out
Value, obtains image single scale notable figure；
Described image single scale notable figure is handled using the method for Multiscale Fusion, obtains scale notable figure, from
And obtain image obvious object region.
Further, described to be based on described image super-pixel processing unit, the topological background for measuring the input picture is set
Reliability can specifically include:
Based on described image super-pixel processing unit, the super-pixel for calculating the input picture extends area；
Super-pixel is calculated in the input picture along the length of the input picture frame；
Extend area and the length based on the super-pixel, calculates the frame Connected degree of the input picture super-pixel；
Based on the frame Connected degree, the topological background confidence level of the input picture is calculated.
Further, the topological background confidence level based on the input picture, calculates color of image and space
Contrast on position feature obtains image primary saliency value, can specifically include:
The topological background confidence level based on the input picture, and described image color characteristic pair is calculated according to following formula
Than degree:
Wherein, the p and piIndicate super-pixel, the i=1 ..., N；The N indicates institute in the input picture
The super-pixel number for including；The biIndicate the topological background confidence level of the input picture；The dcolor(p,pi) table
Show the super-pixel p and piEuclidean distance on CIE L*a*b* color space；The contrastcolor(p) it indicates
Described image color characteristic contrast；
Wherein, the wpos(p,pi) determined by following formula:
Wherein, the dpos(p,pi) indicate the p and piThe Euclidean distance of center；The σposIndicate comparison
Spend opereating specification adjustment factor；
The topological background confidence level based on the input picture, and the spatial position feature pair is calculated according to following formula
Than degree:
Wherein, wcolor(p,pi) indicate the super-pixel p and piSimilitude on color characteristic；The contrastpos
(p) the spatial position Characteristic Contrast degree is indicated；
Wherein,
Wherein, the σcolorIndicate the spatial position Characteristic Contrast degree to the adjusting system of the color characteristic sensitivity
Number；The dcolor(p,pi) indicate the super-pixel p and piIn CIE L*a*b*Euclidean distance on color space；
Based on described image color characteristic contrast and the spatial position Characteristic Contrast degree, and according to described in following formula calculating
Image primary saliency value:
CS (p)=contrastcolor(p)·exp(-k·contrastpos(p))
Wherein, the k indicates described image color characteristic and the spatial position feature to the super-pixel saliency value
The weight coefficient of contribution.
Further, described to correct described image primary saliency value using compactedness method of diffusion, it is significant to obtain compactedness
Figure, can specifically include:
Construct k-regular closed loop figure, wherein respectively connect side by incidence matrix to measure in the closed loop figure；
Base is determined using the manifold smoothness constraint in graph theory based on described image primary saliency value and the incidence matrix
In the manifold ranking function of non-normalized Laplacian Matrix, to obtain the compactedness notable figure.
Further, the topological background confidence level and the compactedness notable figure using the input picture,
Saliency value assignment is carried out, image single scale notable figure is obtained, can specifically include:
Existed using the topological background confidence level of the input picture with the compactedness notable figure and the super-pixel
Euclidean distance on CIE L*a*b* color space carries out saliency value assignment, obtains the saliency value of the input picture super-pixel；
Assign the saliency value of the input picture super-pixel to all pictures in the input picture in the super-pixel
Element, as the saliency value of those pixels, to obtain described image single scale notable figure.
Further, the incidence matrix is determined by following formula:
Wherein, the wijIndicate bonding strength；The n takes positive integer；The dcolor(pi,pj) indicate super-pixel piAnd pj
In CIE L*a*b*Euclidean distance on color space；The σ indicates adjustment factor；The i and the j take 1 ... N, the N
Indicate super-pixel number.
Further, it is described using the topological background confidence level of the input picture and the compactedness notable figure and
The super-pixel is in CIE L*a*b*Euclidean distance on color space carries out saliency value assignment, obtains the super picture of the input picture
The saliency value of element, can specifically include:
The saliency value of the input picture super-pixel is obtained using following objective function:
Wherein, the siIndicate super-pixel piSaliency value；The sjIndicate super-pixel pjSaliency value；The i, j=
1,...,N；The N indicates the number of super-pixel；The biIndicate the topological background confidence level of the input picture；It is described
fiIndicate compactedness notable figure；The λ indicates regulatory factor；The dcolor(pi,pj) indicate the super-pixel piAnd pjIn CIE
L*a*b*Euclidean distance on color space；The σcolorIndicate adjustment factor；The μsmIndicate the adjustment factor of inhibition noise.
To achieve the goals above, according to another aspect of the present invention, a kind of marking area detection system, institute are also provided
The system of stating includes:
Divide module, for carrying out super-pixel segmentation to input picture, obtains image superpixel processing unit；
Metric module is connected with the segmentation module, for being based on described image super-pixel processing unit, measures described defeated
Enter the topological background confidence level of image；
Computing module is connected with the metric module, for the topological background confidence level based on the input picture,
The contrast in color of image and spatial position feature is calculated, image primary saliency value is obtained；
Correction module is connected with the computing module, for aobvious using compactedness method of diffusion amendment described image primary
Work value obtains compactedness notable figure；
Assignment module is connected with the metric module and the correction module respectively, for utilizing the input picture
The topology background confidence level and the compactedness notable figure carry out saliency value assignment, obtain image single scale notable figure；
Fusion Module is connected with the assignment module, for the method using Multiscale Fusion to described image single scale
Notable figure is handled, and scale notable figure is obtained, to obtain image obvious object region.
The embodiment of the present invention provides a kind of salient region detecting method and detection system.Wherein, this method includes to input
Image carries out super-pixel segmentation, obtains image superpixel processing unit；Then, it is based on image superpixel processing unit, is measured defeated
Enter the topological background confidence level of image；Then, the topological background confidence level based on input picture calculates color of image and space bit
The contrast in feature is set, image primary saliency value is obtained；It recycles compactedness method of diffusion to correct image primary saliency value, obtains
To compactedness notable figure；And topological background confidence level and compactedness notable figure using input picture, carry out saliency value tax
Value, obtains image single scale notable figure；Finally, the method using Multiscale Fusion handles image single scale notable figure,
Scale notable figure is obtained, to obtain image obvious object region.By using above-mentioned technical proposal, take into account in the detection process
Image overall conspicuousness and local conspicuousness, realize the beneficial effect of fast and effeciently abstract image marking area.
Detailed description of the invention
Fig. 1 is the flow diagram according to the salient region detecting method of the embodiment of the present invention；
Fig. 2 is the structural schematic diagram according to the marking area detection system of the embodiment of the present invention.
Specific embodiment
The preferred embodiment of the present invention described with reference to the accompanying drawings.It will be apparent to a skilled person that this
A little embodiments are used only for explaining technical principle of the invention, it is not intended that limit the scope of the invention.The present embodiment
Implemented under the premise of the technical scheme of the present invention, in conjunction with detailed embodiment and specific operating process, but this hair
Bright protection scope is not limited to following embodiments.
The prior art is during inferring saliency value using the topological relation of image-region and frame region, often more partially
Global conspicuousness is overweighted, because such methods utilize frame region approximate image global context, and frame region might not be located
In the local neighbor of image-region, so as to lead to the conspicuousness for mistakenly inhibiting obvious object region or mistakenly overstate
The conspicuousness in overall background region.
Basic thought of the embodiment of the present invention is: carrying out super-pixel segmentation to input picture, obtains image processing unit；It connects
, image background confidence level is calculated according to frame Connected degree, and be based on background confidence calculations image saliency value；Then, to institute
It obtains notable figure and carries out compactedness diffusion, correct saliency, obtain more acurrate more compact notable figure；Recycle background confidence
Degree carries out saliency value with compactedness notable figure and spatial neighbors information and optimizes assignment；Finally, the method using Multiscale Fusion obtains
Obtain multiple dimensioned notable figure.
For this purpose, the embodiment of the present invention provides a kind of salient region detecting method.As shown in Figure 1, this method can pass through step
Rapid S100 is realized to step S150.Wherein:
S100: super-pixel segmentation is carried out to input picture, obtains image superpixel processing unit.
Specifically, this step can use superpixel segmentation method and carry out super-pixel segmentation to input picture, to obtain figure
As super-pixel processing unit.
S110: it is based on image superpixel processing unit, measures the topological background confidence level of input picture.
Specifically, this step may include:
S111: being based on image superpixel processing unit, and the super-pixel of calculating input image extends area.
As an example, this step can according to the following formula calculating input image super-pixel extend area:
Wherein, p and piIndicate super-pixel, i=1 ..., N；N indicates super-pixel number included in input picture；σgeo
Indicate adjustment factor；dgeo(p,pi) indicate super-pixel p and piIn CIE L*a*b*Geodesic distance on color space；SpanArea
(p) indicate that the super-pixel of input picture extends area.
For sake of clarity, herein by dgeo(p,pi) it is expressed as dgeo(p, q) can be realized by following formula:
Wherein, dcolor(pi, pi+1) indicate super-pixel piAnd pi+1In CIE L*a*b*Euclidean distance on color space；N takes
Positive integer；dgeo(p, q) indicates super-pixel p and q in CIE L*a*b*Geodesic distance on color space.
S112: in calculating input image super-pixel along input picture frame length.
As an example, this step can carry out in calculating input image super-pixel along the length of input picture frame according to the following formula
Degree:
Wherein, piIndicate super-pixel, i=1 ..., N；N indicates super-pixel number included in input picture；δ(·)
Indicate indicator function, when super-pixel is on input picture frame, otherwise value 1 is 0；B is indicated on input picture frame
Super-pixel set；σgeoIndicate adjustment factor；dgeo(p,pi) indicate super-pixel p and piIn CIE L*a*b*Geodetic on color space
Distance；LB (p) indicates that super-pixel is along the length of input picture frame in input picture.
S113: super-pixel in area and input picture is extended based on super-pixel and is calculated along the length of input picture frame
The frame Connected degree of input picture super-pixel.
As an example, the embodiment of the present invention can be by following formula come the frame Connected degree of calculating input image super-pixel:
Wherein, LB (p) indicates that super-pixel is along the length of input picture frame in input picture；SpanArea (p) is indicated
The super-pixel of input picture extends area；BC (p) indicates input picture frame Connected degree.
S114: frame Connected degree, the topological background confidence level of calculating input image are based on.
As an example, the embodiment of the present invention can be by following formula come the topological background confidence level of calculating input image:
Wherein, BC (pi) indicate input picture frame Connected degree；σbcIndicate the frame for adjusting input picture super-pixel
Weighing factor of the Connected degree to the topological background confidence level of input picture；biIndicate the topological background confidence level of input picture.
S120: the topological background confidence level based on input picture calculates the comparison in color of image and spatial position feature
Degree obtains image primary saliency value.
Specifically, this step may include:
S121: the topological background confidence level based on input picture, and color of image Characteristic Contrast degree is calculated according to following formula:
Wherein, p and piIndicate super-pixel, i=1 ..., N；N indicates super-pixel number included in input picture；biTable
Show the topological background confidence level of input picture；dcolor(p,pi) indicate super-pixel p and piIn CIE L*a*b*Europe on color space
Family name's distance；contrastcolor(p) color of image Characteristic Contrast degree is indicated.
Wherein, wpos(p,pi) determined by following formula:
Wherein, dpos(p,pi) indicate p and piThe Euclidean distance of center；σposIndicate that contrast operation range adjusts system
Number.
S122: the topological background confidence level based on input picture, and spatial position Characteristic Contrast degree is calculated according to following formula:
Wherein, dpos(p,pi) indicate p and piThe Euclidean distance of center；biIndicate the topological background confidence of input picture
Degree；wcolor(p,pi) indicate super-pixel p and piSimilitude on color characteristic；contrastpos(p) representation space position is special
Levy contrast.
Wherein,
Wherein, σcolorAdjustment factor of the representation space position feature contrast to color characteristic sensitivity；dcolor(p,
pi) indicate super-pixel p and piIn CIE L*a*b*Euclidean distance on color space.
S123: it is based on color of image Characteristic Contrast degree and spatial position Characteristic Contrast degree, and according at the beginning of following formula calculating image
Grade saliency value:
CS (p)=contrastcolor(p)·exp(-k·contrastpos(p))；
Wherein, k indicates color of image feature and spatial position feature to the weight coefficient of the contribution of super-pixel saliency value；
contrastpos(p) representation space position feature contrast；contrastcolor(p) color of image Characteristic Contrast degree is indicated.
S130: image primary saliency value is corrected using compactedness method of diffusion, obtains compactedness notable figure.
Specifically, in some alternative embodiments, this step may further include:
S131: construction k-regular closed loop figure, wherein respectively connect side by incidence matrix to measure in the closed loop figure.
In some alternative embodiments, incidence matrix is determined by following formula:
Wherein, wijIndicate bonding strength；N takes positive integer；dcolor(pi, pj) indicate super-pixel piAnd pjIn CIE L*a*b*Face
Euclidean distance in the colour space；σ indicates adjustment factor；I and j take 1 ..., and N, N indicate super-pixel number.
The weight that σ is used to control connection side can take 1/ σ in the specific implementation process2=10.
S132: being based on image primary saliency value and incidence matrix, using the manifold smoothness constraint in graph theory, determines based on non-
The manifold ranking function for normalizing Laplacian Matrix, to obtain compactedness notable figure.
In practical applications, the diffusion process of the manifold ranking function based on non-normalized Laplacian Matrix can basis
Following formula calculates:
f*=(D- α W)-1y；
Wherein, D indicates connection matrix, D=diag { d11,...,dnn, dii=Σjwij, wijIndicate bonding strength；Y=
[y1,...,yn], yi=CS (pi)；CS(pi) indicate image primary saliency value, piIndicate super-pixel；α indicates adjustment factor；i,j,
N takes positive integer；f*Indicate compactedness notable figure,f*It can be resulting by manifold ranking diffusion process
Numerical value can be used to indicate that the saliency value of each super-pixel in the saliency value of each super-pixel in image namely image is fi=
fi *。
Preferably, above-mentioned α takes 0.99.
By step S130, enhance model using compactedness diffusion process to the descriptive power of image local conspicuousness,
So as to obtain more acurrate more compact notable figure.
S140: using the topological background confidence level and compactedness notable figure of input picture, saliency value assignment is carried out, figure is obtained
As single scale notable figure.
Specifically, this step may include:
S141: using the topological background confidence level and compactedness notable figure and super-pixel of input picture in CIE L*a*b*Face
Euclidean distance in the colour space carries out saliency value assignment, obtains the saliency value of input picture super-pixel.
In some alternative embodiments, this step can use following objective function and obtain input picture super-pixel
Saliency value:
Wherein, siIndicate super-pixel piSaliency value；sjIndicate super-pixel pjSaliency value；I, j=1 ..., N；N is indicated
The number of super-pixel；biIndicate the topological background confidence level of input picture；fiIndicate compactedness notable figure；λ indicates regulatory factor,
Preferably, λ takes 1-5.
Wherein,
Wherein, dcolor(pi, pj) indicate super-pixel piAnd pjIn CIE L*a*b*Euclidean distance on color space；σcolorTable
Show adjustment factor；μsmIndicate to inhibit the adjustment factor of noise to inhibit in scene for adjusting the optimization problem under complex scene
Noise.Preferably, μsm=0.1.
As an example, being obtained with input picture super-pixel by carrying out the operations such as derivation to above-mentioned objective function
Saliency value si。
S142: assigning the saliency value of input picture super-pixel to all pixels in input picture in the super-pixel, as
The saliency value of those pixels, to obtain image single scale notable figure.
S150: handling image single scale notable figure using the method for Multiscale Fusion, obtain scale notable figure, from
And obtain image obvious object region.
Specifically, in some alternative embodiments, multiple dimensioned notable figure can be obtained according to the following formula:
Wherein, r=1,2,3,4 correspond to the super-pixel segmentation generated according to the setting of different super-pixel sizes；SrIt indicates
The single scale notable figure generated under different scale super-pixel segmentation result；S indicates multiple dimensioned notable figure.
Above-described embodiment has taken into account image overall conspicuousness and local conspicuousness in the detection process, realizes quickly and effectively
The beneficial effect of ground abstract image marking area.
Although each step is described in the way of above-mentioned precedence in above-described embodiment, this field
Technical staff is appreciated that the effect in order to realize the present embodiment, executes between different steps not necessarily in such order,
It (parallel) execution simultaneously or can be executed with reverse order, these simple variations all protection scope of the present invention it
It is interior.
Based on technical concept identical with embodiment of the method, the embodiment of the present invention also provides a marking area detection system.
As shown in Fig. 2, the system 20 may include: segmentation module 21, metric module 22, computing module 23, correction module 24, assignment mould
Block 25 and Fusion Module 26.Wherein, segmentation module 21 is used to carry out super-pixel segmentation to input picture, obtains at image superpixel
Manage unit.Metric module 22 is connected with segmentation module 21, for being based on image superpixel processing unit, measures opening up for input picture
Flutter background confidence level.Computing module 23 is connected with metric module 22, for the topological background confidence level based on input picture, calculates
Contrast on color of image and spatial position feature obtains image primary saliency value.Correction module 24 and 23 phase of computing module
Even, for correcting image primary saliency value using compactedness method of diffusion, compactedness notable figure is obtained.Assignment module 25 respectively with
Metric module 22 is connected with correction module 24, for utilizing the topological background confidence level and compactedness notable figure of input picture, into
Row saliency value assignment obtains image single scale notable figure.Fusion Module 26 is connected with assignment module 25, for being melted using multiple dimensioned
The method of conjunction handles image single scale notable figure, obtains scale notable figure, to obtain image obvious object region.
It will be understood by those skilled in the art that above-mentioned marking area detection system can also include some other known knot
Structure, such as processor, controller, memory and bus etc., wherein memory include but is not limited to random access memory, flash memory, only
Read memory, programmable read only memory, volatile memory, nonvolatile memory, serial storage, parallel storage or
Register etc., processor include but is not limited to single core processor, multi-core processor, the processor based on X86-based, CPLD/
FPGA, DSP, arm processor, MIPS processor etc., bus may include data/address bus, address bus and control bus.In order to
Embodiment of the disclosure is unnecessarily obscured, these well known structures are not shown in FIG. 2.
It should be noted that marking area detection system provided by the above embodiment and detection method are carrying out marking area
When detection, only carried out with the division of above-mentioned each functional module or step for example, in practical applications, can according to need and
Above-mentioned function distribution is completed by different functional modules, i.e., by the embodiment of the present invention module or step decompose again or
Person's combination, to complete all or part of the functions described above.For module involved in the embodiment of the present invention or step
Title, it is only for distinguish modules or step, be not intended as the improper restriction to the scope of the present invention.
Also it should be noted is that, in order to avoid having obscured the present invention because of unnecessary details, in the accompanying drawings only
It shows the processing step closely related with technical solution of the present invention, therefore and is omitted and little other of relationship of the present invention
Details.
So far, preferred embodiment shown in the drawings is had been combined to the purpose of the present invention, technical scheme and beneficial effects
It has been further described, still, it will be readily appreciated by those skilled in the art that the obvious not office of protection scope of the present invention
It is limited to these specific embodiments.Under the premise of without departing from the principle of the present invention, all within the spirits and principles of the present invention,
Those skilled in the art can make equivalent change or replacement to the relevant technologies feature, the technology after these changes or replacement
Scheme will fall within the scope of protection of the present invention.
Claims (8)
1. a kind of salient region detecting method, which is characterized in that the described method includes:
Super-pixel segmentation is carried out to input picture, obtains image superpixel processing unit；
Based on described image super-pixel processing unit, the topological background confidence level of the input picture is measured；
The topological background confidence level based on the input picture, calculates the comparison in color of image and spatial position feature
Degree obtains image primary saliency value；
Described image primary saliency value is corrected using compactedness method of diffusion, obtains compactedness notable figure；
Using the topological background confidence level of the input picture and the compactedness notable figure, saliency value assignment is carried out, is obtained
To image single scale notable figure；
Described image single scale notable figure is handled using the method for Multiscale Fusion, scale notable figure is obtained, to obtain
Obtain image obvious object region.
2. the method according to claim 1, wherein described be based on described image super-pixel processing unit, measurement
The topological background confidence level of the input picture, specifically includes:
Based on described image super-pixel processing unit, the super-pixel for calculating the input picture extends area；
Super-pixel is calculated in the input picture along the length of the input picture frame；
Extend area and the length based on the super-pixel, calculates the frame Connected degree of the input picture super-pixel；
Based on the frame Connected degree, the topological background confidence level of the input picture is calculated.
3. the method according to claim 1, wherein the topological background based on the input picture is set
Reliability calculates the contrast in color of image and spatial position feature, obtains image primary saliency value, specifically includes:
The topological background confidence level based on the input picture, and the comparison of described image color characteristic is calculated according to following formula
Degree:
Wherein, p and piIndicate super-pixel, i=1 ..., N；N indicates super-pixel number included in the input picture；biTable
Show the topological background confidence level of the input picture；dcolor(p,pi) indicate the super-pixel p and piIn CIE L*a*
b*Euclidean distance on color space；contrastcolor(p) described image color characteristic contrast is indicated；
Wherein, wpos(p,pi) determined by following formula:
Wherein, dpos(p,pi) indicate the p and piThe Euclidean distance of center；σposIndicate contrast operation range tune
Save coefficient；
The topological background confidence level based on the input picture, and the spatial position Characteristic Contrast is calculated according to following formula
Degree:
Wherein, wcolor(p,pi) indicate the super-pixel p and piSimilitude on color characteristic；contrastpos(p) institute is indicated
State spatial position Characteristic Contrast degree；
Wherein,
Wherein, σcolorIndicate the spatial position Characteristic Contrast degree to the adjustment factor of the color characteristic sensitivity；
Described image is calculated based on described image color characteristic contrast and the spatial position Characteristic Contrast degree, and according to following formula
Primary saliency value:
CS (p)=contrastcolor(p)·exp(-k·contrastpos(p))
Wherein, k indicates described image color characteristic and the spatial position feature to the power of the contribution of the super-pixel saliency value
Weight coefficient.
4. the method according to claim 1, wherein at the beginning of the amendment described image using compactedness method of diffusion
Grade saliency value, obtains compactedness notable figure, specifically includes:
Construct k-regular closed loop figure, wherein respectively connect side by incidence matrix to measure in the closed loop figure；
It is determined using the manifold smoothness constraint in graph theory based on non-based on described image primary saliency value and the incidence matrix
The manifold ranking function for normalizing Laplacian Matrix, to obtain the compactedness notable figure.
5. the method according to claim 1, wherein described set using the topological background of the input picture
Reliability and the compactedness notable figure carry out saliency value assignment, obtain image single scale notable figure, specifically include:
Using the topological background confidence level of the input picture and the compactedness notable figure and the super-pixel in CIE L*
a*b*Euclidean distance on color space carries out saliency value assignment, obtains the saliency value of the input picture super-pixel；
It assigns the saliency value of the input picture super-pixel to all pixels in the input picture in the super-pixel, makees
For the saliency value of those pixels, to obtain described image single scale notable figure.
6. according to the method described in claim 4, it is characterized in that, the incidence matrix is determined by following formula:
W=[wij]n×n,
Wherein, wijIndicate bonding strength；N takes positive integer；dcolor(pi, pj) indicate super-pixel piAnd pjIn CIE L*a*b*Color is empty
Between on Euclidean distance；σ indicates adjustment factor；I and j take 1 ..., and N, N indicate super-pixel number.
7. according to the method described in claim 5, it is characterized in that, described set using the topological background of the input picture
Reliability and the compactedness notable figure and the super-pixel are in CIE L*a*b*Euclidean distance on color space carries out saliency value
Assignment obtains the saliency value of the input picture super-pixel, specifically includes:
The saliency value of the input picture super-pixel is obtained using following objective function:
Wherein, siIndicate super-pixel piSaliency value；sjIndicate super-pixel pjSaliency value；I, j=1 ..., N；N indicates super-pixel
Number；biIndicate the topological background confidence level of the input picture；fiIndicate each super-pixel in compactedness notable figure
Saliency value；λ indicates regulatory factor；dcolor(pi, pj) indicate the super-pixel piAnd pjIn CIE L*a*b*Europe on color space
Family name's distance；σcolorIndicate adjustment factor；μsmIndicate the adjustment factor of inhibition noise.
8. a kind of marking area detection system, which is characterized in that the system comprises:
Divide module, for carrying out super-pixel segmentation to input picture, obtains image superpixel processing unit；
Metric module is connected with the segmentation module, for being based on described image super-pixel processing unit, measures the input figure
The topological background confidence level of picture；
Computing module is connected with the metric module, for the topological background confidence level based on the input picture, calculates
Contrast on color of image and spatial position feature obtains image primary saliency value；
Correction module is connected with the computing module, for correcting described image primary saliency value using compactedness method of diffusion,
Obtain compactedness notable figure；
Assignment module is connected with the metric module and the correction module respectively, for using described in the input picture
Topological background confidence level and the compactedness notable figure carry out saliency value assignment, obtain image single scale notable figure；
Fusion Module is connected with the assignment module, for significant to described image single scale using the method for Multiscale Fusion
Figure is handled, and scale notable figure is obtained, to obtain image obvious object region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610889100.3A CN106570498B (en) | 2016-10-12 | 2016-10-12 | Salient region detecting method and detection system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610889100.3A CN106570498B (en) | 2016-10-12 | 2016-10-12 | Salient region detecting method and detection system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106570498A CN106570498A (en) | 2017-04-19 |
| CN106570498B true CN106570498B (en) | 2019-06-14 |
Family
ID=58532817
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610889100.3A Active CN106570498B (en) | 2016-10-12 | 2016-10-12 | Salient region detecting method and detection system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106570498B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805136B (en) * | 2018-03-26 | 2021-11-05 | 中国地质大学（武汉） | Significance detection method for water surface pollutant monitoring |
| CN109284753B (en) * | 2018-08-30 | 2020-07-07 | 深圳大学 | Positioning method and application of liquid transmission pipeline |
| CN109753957B (en) * | 2018-12-07 | 2020-11-27 | 东软集团股份有限公司 | Image significance detection method and device, storage medium and electronic equipment |
| CN110942095A (en) * | 2019-11-27 | 2020-03-31 | 中国科学院自动化研究所 | Method and system for detecting salient object area |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745468A (en) * | 2014-01-07 | 2014-04-23 | 上海交通大学 | Significant object detecting method based on graph structure and boundary apriority |
| CN103914834A (en) * | 2014-03-17 | 2014-07-09 | 上海交通大学 | Significant object detection method based on foreground priori and background priori |
| CN105046701A (en) * | 2015-07-08 | 2015-11-11 | 安徽大学 | Image composition line-based multi-scale salient target detection method |
-
2016
- 2016-10-12 CN CN201610889100.3A patent/CN106570498B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745468A (en) * | 2014-01-07 | 2014-04-23 | 上海交通大学 | Significant object detecting method based on graph structure and boundary apriority |
| CN103914834A (en) * | 2014-03-17 | 2014-07-09 | 上海交通大学 | Significant object detection method based on foreground priori and background priori |
| CN105046701A (en) * | 2015-07-08 | 2015-11-11 | 安徽大学 | Image composition line-based multi-scale salient target detection method |
Non-Patent Citations (2)
| Title |
|---|
| Salient region detection based on Local and Global Saliency;Peng Wang 等;《IEEE International Conference on Robotics & Automation (ICRA)》;20140929;全文 * |
| Sparse-Distinctive Saliency Detection;Yongkang Luo 等;《IEEE SIGNAL PROCESSING LETTERS》;20141218;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106570498A (en) | 2017-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570498B (en) | Salient region detecting method and detection system | |
| CN108717710B (en) | Positioning method, device and system in indoor environment | |
| US20180278823A1 (en) | Auto-exposure technologies using odometry | |
| CN108401457A (en) | A kind of control method of exposure, device and unmanned plane | |
| CN110807350A (en) | System and method for visual SLAM for scan matching | |
| EP2879080B1 (en) | Image processing device and method, and computer readable medium | |
| CN113674416B (en) | Three-dimensional map construction method and device, electronic equipment and storage medium | |
| CN108322724B (en) | Image solid matching method and binocular vision equipment | |
| CN110222686B (en) | Object detection method, object detection device, computer equipment and storage medium | |
| CN110147382A (en) | Lane line update method, device, equipment, system and readable storage medium storing program for executing | |
| US8416989B2 (en) | Image processing apparatus, image capture apparatus, image processing method, and program | |
| CN105144710A (en) | Technologies for increasing the accuracy of depth camera images | |
| CN108474658A (en) | Ground Morphology observation method and system, unmanned plane landing method and unmanned plane | |
| JP2020017173A (en) | Moving entity | |
| CN109948624A (en) | Method, apparatus, electronic equipment and the computer storage medium of feature extraction | |
| El Bouazzaoui et al. | Enhancing rgb-d slam performances considering sensor specifications for indoor localization | |
| US9958868B2 (en) | Moving object controller, moving object control method, and integrated circuit | |
| CN115655291A (en) | Laser SLAM closed-loop mapping method and device, mobile robot, equipment and medium | |
| CN114782496A (en) | Object tracking method and device, storage medium and electronic device | |
| CN113971687A (en) | Object tracking method, device electronic equipment and storage medium | |
| CN114972465A (en) | Image target depth detection method and device, electronic equipment and storage medium | |
| CN105323460B (en) | Image processing equipment and its control method | |
| US10447992B1 (en) | Image processing method and system | |
| JP2014203133A (en) | Image processing device and image processing method | |
| CN110288633A (en) | Target tracking method, device, readable storage medium storing program for executing and electronic equipment |
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 |