CN107016682A - A kind of notable object self-adapting division method of natural image - Google Patents
A kind of notable object self-adapting division method of natural image Download PDFInfo
- Publication number
- CN107016682A CN107016682A CN201710233611.4A CN201710233611A CN107016682A CN 107016682 A CN107016682 A CN 107016682A CN 201710233611 A CN201710233611 A CN 201710233611A CN 107016682 A CN107016682 A CN 107016682A
- Authority
- CN
- China
- Prior art keywords
- region
- image
- segmentation
- curve
- significance probability
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of notable object self-adapting division method of natural image, including：Divided the image into Random Walk Algorithm as different zones；Design section joint conspicuousness distribution, core salient region is chosen according to maximum significance probability；Using core salient region as seed, analysis calculates joint significance probability of the contiguous zone with respect to seed；The contiguous zone that the joint significance probability of relative seed is more than or equal to a certain numerical value is merged, notable object prime area S is obtained；The notable Object Segmentation of image is realized with level set algorithm.The present invention is according to the edge attributes of object, using the relative different of neighborhood territory pixel, is divided the image into Random Walk Algorithm as different zones, the influence of weak edge and noise to segmentation is made up to a certain extent.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of notable object adaptivenon-uniform sampling side of natural image
Method.
Background technology
Image, as most common information carrier in daily life, is a kind of objective description to real world.The mankind obtain
In the information taken, have and greatly come from vision.For human visual system, no matter many complicated images of scene, all
The different objects in image can that accurately and quickly soon be recognized into difference, and according to object and its different objects in space
Context relation to image carry out analysis and understanding.To the human eye, object is the elementary cell that graphical analysis understands, object point
The validity extracted and determine pictures subsequent analyzing and processing is cut, picture material, important semantic object is especially embodied, that is, shows
Write object.The notable object of image is to imitate the partial information that human vision concern mechanism extracts human eye concern from image.This pair
The extraction of elephant not only facilitates the visual cognition rule for finding out the mankind, and machine vision is laid the foundation.
With society, the development of science and technology, the information technology developed rapidly promotes image information increasingly to expand, accelerated simultaneously
The infiltration for various aspects of being lived to people, makes the mainstay of today's society.In order to solve the view data of magnanimity,
People handle and analyzed view data by machine vision, analysis and query image contain to oneself useful or available office
Portion's information.It is blindness and unpractical to analyze all view data comprehensively, it is necessary to re-recognize and describe vision significance
Connotation and its associated with information availability, to seeking to meet visually-perceptible and specific tasks using conspicuousness detection technique
Extraction approach.Human visual system itself just has very efficient data screening ability, unfortunately current on vision
The theoretical description noted is also generally hypothesis viewpoint.Also it is only to be proposed towards concrete application task in terms of computation vision
Limited computation model, is not yet formed than more complete theoretical frame.Therefore how image is fast and effeciently handled and analyze, from
It is middle to extract the basic assignment that the semantic object with important information is current graphical analysis understanding.
Notable object adaptivenon-uniform sampling technology is to utilize computer to divide significant semantic object from image
Cut out.The application of the technology it is wide oneself through being related to all many-sides of machine vision：1) wrapped towards still image processing
Include specific image region segmentation and profile evolution, the adapting to image compression and coding of object-oriented, object-based image mesh
Mark detection and scene analysis and images match, scene rendering and digital watermarking.2) notable object is included towards video analysis processing
Motion detection is supervised with tracking, object-based video image compression coding, object-based video content analysis, retrieval and video
Control and object abnormality detection etc..
Traditional images partitioning algorithm only divides image only in accordance with the uniformity of characteristics of the underlying image (brightness, color and texture)
Several semantic regions are cut into, the contribution that regional is understood graphical analysis, i.e. conspicuousness are not pointed out.Region significance detection according to
Its significance is analyzed according to region contrast and locus, but the region only uniformity with low-level image feature lacks the semanteme of object
Property.
The content of the invention
It is an object of the invention to solve defect that above-mentioned prior art is present there is provided a kind of notable object of natural image from
Adapt to dividing method.
A kind of notable object self-adapting division method of natural image, comprises the following steps：
Step one：According to the edge attributes of image, divided the image into Random Walk Algorithm as different zones；
Step 2：According to human eye notice Mechanism establishing image-region conspicuousness, according to sensitiveness point of the human eye to color
The color contrast in region, the significant spatial degree according to eye space notice mechanism analysis estimation region are analysed, with reference to the face
Color contrast and significant spatial degree, merge background priori, design section joint conspicuousness distribution, according to maximum significance probability
Choose core salient region；
Step 3：Using core salient region as seed, according to the spatial relation of regional, analysis calculating is adjoined
Joint significance probability of the region with respect to seed；
Step 4：The contiguous zone that the joint significance probability of relative seed is more than or equal to a certain numerical value is merged,
Obtain notable object prime area S；
Step 5：By initial curve of the prime area the notable Object Segmentation of image is realized with level set algorithm.
Wherein, the human eye notice mechanism refers to：Human eye note be the mankind obtain information an approach, when the external world letter
When breath stimulates human eye, vision attention can select useful information from substantial amounts of external information, refuse garbage.Cognitive psychological
Learn research to show, in the complicated scene of analysis, human eye uses Selective Attention Mechanism.From the point of view of picture material, by local feature
(color, brightness, orientation, motion and stereo disparity) selection scene specific region, while being primarily focused on local feature phase
To the region differed greatly, and more close-up and analysis are carried out to it；From the point of view of spatially, human eye at first view image when, note
Power is generally concentrated at image centre position.
The background priori refers to：Analyze according to statistics, 85% image from Internet, its image boundary belongs to
Background area, according to this priori, the present invention regard the region of 15 pixels of image boundary as initial background.
Further, method as described above, is divided the image into as not same district with Random Walk Algorithm described in step one
Domain method is：
From L*a*b*Color mode utilizes L as the feature space of natural image*Component calculates neighborhood territory pixel's
Relative different design segmentation weight：
With Random Walk Algorithm by L*Component is divided into M region, kth region RkArea nk
nk=card i | i ∈ Rk}
(4)。
Further, method as described above, is designed with reference to the color contrast and significant spatial degree described in step 2
Regional combination conspicuousness with respect to background is distributed, and choosing core salient region according to maximum significance probability includes：Determine region
RkJoint significance probability p (k), and according to maximum significance probability choose image in core salient region；
Wherein, the determination region RkJoint significance probability p (k) include：
Region RkJoint significance probability p (k) be：
P (k)=ωc(k)ωs(k) (10)
Wherein, ωc(k) it is region RkColor contrast, ωs(k) it is region RkSignificant spatial degree；
The region RkContrast ωc(k) it is：
Wherein, piRepresent region i area weight；μk, μiRegion k is represented respectively, and i is in L*a*b*Average.
The region RkSignificant spatial degree ωs(k) it is：
Wherein o represents the center of image, variances sigma2It is the normalization of picture size；ziThe space of representative image pixel
σ is determined by picture size in position, formula：
Further, method as described above, analysis described in step 3 calculates joint of the contiguous zone with respect to seed region
Significance probability includes：Using seed region as starting point, along clockwise with 80 ° of directions, 45 °, 90 °, 135 °, 180 °, 225 °,
270 °, 315 ° of nodes for research contiguous zones, statistical analysis obtains the regional from same target relative to seed region
Significance probability is all higher than being equal to 0.8.
Further, method as described above, is calculated by initial curve of the prime area with level set described in step 5
Method realizes that the notable Object Segmentation of image includes：
The Lipschitz function representations of notable object prime area are：
Image segmentation algorithm based on level set is according to the local edge of object outline, using edge as the pact of curve evolvement
Beam condition, image Ω → [0,1] segmentation curve evolvement energy function be：
WhereinInside or outside of curve region is represented,Segmentation curve is represented, g represents image border indicator function：
Using rapid decrease algorithm, initial curveIt is iterated, solves curve evolvement energy function, obtains object point
Cut curve.
Further, method as described above, by the initial curve cut zone confidence level, Region confidence Pr：
A in formulak,Ak-1Kth is represented during curve evolvement respectively and splits curve and its interior zone during k-1 iteration.
Beneficial effect：
1) present invention, will with Random Walk Algorithm using the relative different of neighborhood territory pixel according to the edge attributes of object
Image is divided into different zones, and the influence of weak edge and noise to segmentation is made up to a certain extent.
2) present invention has considered background priori in terms of region significance analysis, and pixel contrast and region are empty
Between the contribution of the feature to importance in human eye such as relativeness.
3) present invention chooses core salient region, using core salient region as seed, root according to maximum significance probability
The prime area of saliency object is constructed according to context relation and relative seed contrast.
4) present invention realizes the notable Object Segmentation of image with level set algorithm, and initial curve is adjacent to notable object wheel
Exterior feature, the curve evolvement time is shorter, compensate for the relatively low deficiency of traditional activity outline segmentation operational efficiency
5) present invention organically combines the technologies such as subject area feature, region significance and image segmentation, establishes image and shows
Write sex object segmentation framework.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below technical scheme in the present invention carry out it is clear
Chu, it is fully described by, it is clear that described embodiment is a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
In order to make up the shortcoming of two kinds of algorithms, the present invention organically combines two kinds of algorithms, establishes notable Object Segmentation mould
Type, the model according to the edge attributes of object, is divided the image into as different zones, algorithm profit with Random Walk Algorithm first
With the relative different of neighborhood territory pixel, influence of the weak edge to region segmentation compensate for.
Secondly the image boundary of statistical analysis 85% belongs to background this feature, belongs to background characteristics according to image boundary
Set up region significance analysis foundation.Simultaneously according to human eye to the relative contrast of the sensitivity analysis of color to region；Foundation
The significant spatial degree of eye space notice mechanism analysis estimation region, with reference to contrast and the relative background of significant spatial degree design
Regional combination conspicuousness distribution, according to maximum significance probability choose core salient region.
Finally using core salient region as seed, based on context relation and relative seed contrast construct conspicuousness pair
The prime area of elephant；The notable Object Segmentation of image is realized with level set algorithm by initial curve of the area, while according to segmentation
Region confidence, establish curve convergence condition, it is to avoid segmentation curve crosses convergence.
2.1 natural image region segmentations
The pixel of the same area should have identical attribute in feature space in natural image, in order to describe in nature
Any color and processing speed, the present invention selects L*a*b*Color mode as natural image feature space, within this space
L*,a*,b*Three-component is mutually perpendicular, L*Representation in components image luminance information, the component can reflect the main contents of image, a*,
b*Component represents two color components of natural image respectively.
Wherein, X, Y, Z are the three-component of CIEXYZ color modes, Xn,Yn,ZnFor reference white, respectively 95.047,100 Hes
108.083；F (t) is defined as：
Due to optical principle, there is weak edge, edge and do not connect in the object that natural image is likely to result in image in imaging
Continuous and ambiguous region, while inevitably by attacked by noise in image propagation, storing process.L*a*b*Color
The L of pattern*Component reflects the main contents and image border characteristic of image, is image border set with reference to object bounds set
Subset, while suppressing the influence of weak edge and noise on image region segmentation, the present invention utilizes L*Component neighborhood territory pixelRelative different design segmentation weight：
With Random Walk Algorithm to L*Component carries out region segmentation, and calculates k-th of cut zone RkArea nk。
nk=card i | i ∈ Rk}
(4)
(K herein represents k-th of cut zone, nkContribution be calculate regional area weight it is notable to region
The contribution of degree)
2.2 region significances are analyzed
Region significance analysis is similar to the attention stage in visually-perceptible, but it is not related to typically in specific semanteme
Hold, by the picture element feature related to semanteme, utilize its conspicuousness of the Disparity Analysis between region.The present invention from region office
Portion and global disparity set out contribution of the analysis regional to vision.Wherein local parallax are to represent region using picture contrast
Relative significance, contrast is the relative change size of adjacent pixel, and its value shows that more greatly contrast is higher, and image is more clear
Clear, color is sharper, conversely, image blurring.Area size is also to influence a factor of human eye attention rate simultaneously；Pay close attention to area
Large area and ignore zonule.According to this property, the present invention considers the area weight of regional to region significance
Contribution, size is region R in W × H imageskArea weight be：
The pixel of the same area should have identical attribute in feature space in natural image, using area pixel in spy
Levy space L*a*b*Mean μ represents the region, region RkMean μkIt is expressed as：
In combination with attention force-responsive of the human eye to area, region R is definedkContrast ωc(k)：
Definition region R of the present inventionkContrast be that, relative to the weighting contrast in all regions of image, it is local contrast
The extension of degree.The value size indicates the region relative to emphasis degree of other regions to human eye, and the value shows more greatly the area
Domain understands more important from pixel angle to graphical analysis.
In human visual system, more notices are attracted close to the region of picture centre, and image boundary region is usually
It is ignored.With the increase of the distance between object and picture centre, notice that gain is devalued.This is in saliency detection algorithm quilt
Referred to as " centre deviation rule ".The present invention devises the significant spatial degree in region according to " the centre deviation rule " of human eye, by area
Domain RkSignificant spatial degree ωs(k) define：
Wherein o represents the center of image, variances sigma2It is the normalization of picture size.According to σ in 3 σ principles (8) formulas by image
Size is determined：
Region contrast ωc(k) with significant spatial degree ωs(k) area pixel and relative tertiary location have been reacted respectively to people
The contribution of eye.Human eye usually combines both the importance for judging a region, and the present invention integrates contrast with multiplication
Region R is obtained with significant spatial degreekJoint significance probability p (k).
P (k)=ωc(k)ωs(k) (10)
The image boundary that statistical analysis of the present invention comes from Internet epigraph, the image set 85% belongs to background
Region.It is empty in feature with the region according to this feature using the region away from 15 pixels of image boundary as initial background region
Between joint significance probability based on, other regional combination significance probabilities are normalized.
Regional combination significance probability reflects the region in terms of area pixel and locus two and attracts human eye note
The degree for power of anticipating, joint significance probability is bigger, and the region is more easily taken seriously relative to other regions, conversely, being ignored.This
Invention chooses core salient region in image according to maximum significance probability.
The Random Walk Algorithm general character of pixel distribution and interregional otherness out of image-region are divided the image into
For different zones, the low-level image feature of regional has uniformity but without semantic meaning.Significance analysis Algorithm Analysis
Relative attraction of the regional to human eye.It is special that the core salient region chosen according to maximum probability only relies upon image bottom
Levy, the Semantic in region is not considered yet.
Natural image typically shows as diversity and variability in terms of content, and non-the one of region is shown as on pixel level
Cause property, pixel is non-constant in region, interregional there is gradually changeable.Meanwhile, the same object in natural image is at image bottom
Multiple regions are often appeared as in layer feature, regional is different to the attraction of human eye.Due to from same target
Regional abuts one another in space, and contiguous zone is little relative to the joint significant difference of seed region.In order to from certainly
The notable object with semantic meaning, the notice mechanism of the invention according to human eye, with core conspicuousness are partitioned into right image
Region is seed, and according to the spatial relation of regional, it is notable with respect to the joint of seed region that analysis calculates contiguous zone
Property probability.The present invention calculates seed contiguous zone according to the analysis of the context relation of same target regional, with seed region
For starting point, along clockwise with 80 ° of directions, 45 °, 90 °, 135 °, 180 °, 225 °, 270 °, 315 ° of nodes for research contiguous zones.
The present invention is to from Internet epigraph, objects in images shows as multiple section objects in terms of low-level image feature, statistics
Analyze the regional from same target to be all higher than being equal to 0.8 relative to the significance probability of seed region, according to this
Statistical property, the present invention merges the contiguous zone that the joint significance probability of relative seed region is more than or equal to 0.8, obtains
To notable object prime area, the regional ensemble is expressed as S.
2.3. saliency object is split
Random Walk Algorithm is divided the image into as different zones according to the low-level image feature of image, and region significance is analyzed only
Vision attention of the different zones to human eye is analyzed, specific semantic content is not related to yet.Provided to split from image
There is the object of semantic meaning, according to the local edge of object outline, with the partitioning algorithm segmentation figure picture based on active contour
Notable object, the algorithm is to realize Object Segmentation by the segmentation initial curve that develops.With the Lipschitz functions of three dimensionsLevel set representations curve C：Represent subject area,Region outside for object, the segmentation
The expression of curve is substantially to represent different zones using functional symbol.Calculate for convenience, introduce Heaviside functionsInside or outside of curve region is represented, curve C is represented byDerivative Dirac estimate WithDifference table
Show as follows：
The Lipschitz function representations of notable object prime area are：
(12) purpose of formula is to represent notable object prime area referencing function in 2.2, so as to realize with semantic right
The segmentation of elephant, function of region is represented so that the initial curve of notable Object Segmentation Algorithm is adjacent to object outline, is greatly reduced
The calculating cost of Object Segmentation.
According to the local edge of object outline, using edge as the constraints of curve evolvement, image Ω → [0,1] segmentation
Curve evolvement energy function be：
G represents image border indicator function：
If curve is located at smooth region, the gradient amplitude convergence zero, edge indicator function tends to 1,It is larger；If curve
Positioned at edge, its gradient amplitude is larger,Minimum, curve stops developing.The present invention uses rapid decrease Algorithm for Solving curve
The solution of evolution energy function (13) formula.
Discretization calculates (13) formula,Discretization computing：
According to cut zone confidence level, cut zone confidence level Pr is devised：
A represents Object Segmentation curve and its interior zone in formula.When the confidence level of cut zone meets following condition, table
Iterative segmentation result similarity reaches the degree required by segmentation twice before and after bright, then stops smooth iteration：
Pr≥T (17)
T is cut zone confidence threshold value.
In traditional activity profile algorithm, perpetual object generally has artificially given closing initial curve to represent, if initially
Curve is apart from each other with object outline, and the curve evolvement time is longer.Initial curve of the present invention is adjacent to notable object outline, curve
The evolution time is shorter.
Saliency object in image can be fast and accurately identified using high low-level feature in human visual system, from figure
From the point of view of the low-level image feature of picture, object forms main caused by the image self attributes such as edge, border, and object conspicuousness mainly depends on
Brightness, the relative different and relative tertiary location of color between object.The notable detection algorithm generation pixel of traditional images is " important
Property " probability notable figure, propose image binaryzation conspicuousness partitioning estimation model with random theory.According to human eye to color
Sensitiveness, region significance analysis model is established using color component statistical property.By human eye to different frequency signals
Response is different, devises the current conspicuousness detection algorithm based on signal frequency.Major part conspicuousness detection algorithm is at present
Set up human eye differentiate obscurity boundary object ability-CSF on the basis of, using Analysis of Contrast estimate image pixel or
Contribution of the region to vision, designed image region significance analysis model.Domain scope is analyzed according to region significance and is divided into office
Portion's contrast and global contrast model.Region contrast and relative position relation are organically combined, it is proposed that aobvious based on cluster
Work property detection algorithm.Traditional conspicuousness detection algorithm only highlights significance level of the region to graphical analysis, but region lacks
Weary Object Semanteme.
In order to extract the object with semantic concept from image, usually using image segmentation algorithm, based on active contour
Partitioning algorithm the object of complete priori can be effectively extracted from image, the evolution of algorithm priori curves simultaneously combine image-region
Or marginal information realizes Object Segmentation.In order to suppress the influence of texture and noise to segmentation, Mumford and Shah are carried out to image
The parted pattern of sectionally smooth and the curve that develops, but MS functionals are non-convex, are solved difficult.Chan and Vess is with regional average value table
Show that object establishes CV parted patterns, this method is preferable to cartoon image segmentation effect, while to insensitive for noise.In order to suppress
Texture Tsai and Yezzi et al. propose PS (Piece-Smooth) parted pattern of piecewise approximation, and the model is to a certain extent
Texture is inhibited, but amount of calculation is larger.Li et al. is according to the relation between image object profile and edge, it is proposed that based on edge
Active contour partitioning algorithm, but the algorithm make use of image border local message, and segmentation result is more sensitive to noise.In order to press down
Noise processed, usually treats segmentation figure picture and carries out Gaussian smoothing.But Gaussian smoothing has obscured object outline, weak edge is split and imitated
It is really undesirable.Conventional segmentation algorithm can effectively divide the image into significant region, but can not assess the region of segmentation
To the importance of visual analysis.
The present invention is divided the image into as different zones with Random Walk Algorithm, compensate for according to the edge attributes of object
Influence of the weak edge to region segmentation.Secondly belonging to background characteristics according to image boundary sets up region significance analysis foundation.Together
When according to human eye to the sensitivity analysis of color to region contrast；According to eye space notice mechanism analysis estimation region
Significant spatial degree, is distributed with reference to the regional combination conspicuousness of contrast and the relative background of significant spatial degree design, according to maximum notable
Property probability choose core salient region.The present invention considers background priori simultaneously in terms of region significance analysis, as
Contribution of the feature such as plain contrast and regional space relativeness to importance in human eye, using core salient region as seed,
Based on context relation and relative seed contrast construct the prime area of saliency object, compensate for traditional conspicuousness detection and lack
The deficiency of weary semantic meaning, the area is that initial curve realizes the notable Object Segmentation of image with level set algorithm, and the present invention is initial
Curve is adjacent to notable object outline, and the curve evolvement time is shorter.Simultaneously according to cut zone confidence level, curve convergence is established
Condition, it is to avoid segmentation curve crosses convergence.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
The present invention is described in detail with reference to the foregoing embodiments, it will be understood by those within the art that：It still may be used
To be modified to the technical scheme described in foregoing embodiments, or equivalent substitution is carried out to which part technical characteristic；
And these modification or replace, do not make appropriate technical solution essence depart from various embodiments of the present invention technical scheme spirit and
Scope.
Claims (6)
1. a kind of notable object self-adapting division method of natural image, it is characterised in that comprise the following steps：
Step one：According to the edge attributes of image, divided the image into Random Walk Algorithm as different zones；
Step 2：According to human eye notice Mechanism establishing image-region conspicuousness, according to sensitivity analysis area of the human eye to color
The color contrast in domain, the significant spatial degree according to eye space notice mechanism analysis estimation region, with reference to the color pair
Than degree and significant spatial degree, background priori is merged, design section joint conspicuousness distribution is chosen according to maximum significance probability
Core salient region；
Step 3：Using core salient region as seed, according to the spatial relation of regional, analysis calculates contiguous zone
With respect to the joint significance probability of seed；
Step 4：The contiguous zone that the joint significance probability of relative seed is more than or equal to a certain numerical value is merged, obtained
Notable object prime area S；
Step 5：By initial curve of the prime area the notable Object Segmentation of image is realized with level set algorithm.
2. according to the method described in claim 1, it is characterised in that divided the image into described in step one with Random Walk Algorithm
It is for different zones method：
From L*a*b*Color mode utilizes L as the feature space of natural image*Component calculates neighborhood territory pixel It is relative
Difference design segmentation weight：
With Random Walk Algorithm by L*Component is divided into M region, kth region RkArea nk
nk=card i | i ∈ Rk} (4)。
3. method according to claim 2, it is characterised in that aobvious with reference to the color contrast and space described in step 2
The regional combination conspicuousness of the relative background of work degree design is distributed, and choosing core salient region according to maximum significance probability includes：
Determine region RkJoint significance probability p (k), and according to maximum significance probability choose image in core salient region；
Wherein, the determination region RkJoint significance probability p (k) include：
Region RkJoint significance probability p (k) be：
P (k)=ωc(k)ωs(k) (10)
Wherein, ωc(k) it is region RkColor contrast, ωs(k) it is region RkSignificant spatial degree；
The region RkContrast ωc(k) it is：
Wherein, piRepresent region i area weight；μk, μiRegion k is represented respectively, and i is in L*a*b*Average；
The region RkSignificant spatial degree ωs(k) it is：
Wherein o represents the center of image, variances sigma2It is the normalization of picture size；ziThe locus of representative image pixel,
σ is determined by picture size in formula：
4. method according to claim 3, it is characterised in that analysis described in step 3 calculates contiguous zone with respect to seed zone
The joint significance probability in domain includes：Using seed region as starting point, along clockwise with 80 ° of directions, 45 °, 90 °, 135 °,
180 °, 225 °, 270 °, 315 ° of nodes for research contiguous zones, statistical analysis obtain the regional from same target relative to
The significance probability of seed region is all higher than being equal to 0.8.
5. method according to claim 4, it is characterised in that used described in step 5 by initial curve of the prime area
Level set algorithm realizes that the notable Object Segmentation of image includes：
The Lipschitz function representations of notable object prime area are：
Image segmentation algorithm based on level set is according to the local edge of object outline, the constraint bar by curve evolvement of edge
Part, image Ω → [0,1] segmentation curve evolvement energy function be：
WhereinInside or outside of curve region is represented,Segmentation curve is represented, g represents image border indicator function：
G=(1+ | ▽ Gσ*u|)-1
Using rapid decrease algorithm, initial curveIt is iterated, solves curve evolvement energy function, obtains Object Segmentation bent
Line.
6. method according to claim 5, it is characterised in that by the initial curve cut zone confidence level, the region
Confidence level Pr：
A in formulak,Ak-1Kth is represented during curve evolvement respectively and splits curve and its interior zone during k-1 iteration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710233611.4A CN107016682B (en) | 2017-04-11 | 2017-04-11 | Self-adaptive segmentation method for salient objects of natural images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710233611.4A CN107016682B (en) | 2017-04-11 | 2017-04-11 | Self-adaptive segmentation method for salient objects of natural images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107016682A true CN107016682A (en) | 2017-08-04 |
| CN107016682B CN107016682B (en) | 2020-03-31 |
Family
ID=59445269
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710233611.4A Expired - Fee Related CN107016682B (en) | 2017-04-11 | 2017-04-11 | Self-adaptive segmentation method for salient objects of natural images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107016682B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107680109A (en) * | 2017-09-15 | 2018-02-09 | 盐城禅图智能科技有限公司 | It is a kind of to quote inverse notice and the image, semantic dividing method of pixel similarity study |
| CN107818566A (en) * | 2017-10-27 | 2018-03-20 | 中山大学 | A kind of image partition method based on partial structurtes information around random walk combination pixel |
| CN108564559A (en) * | 2018-03-14 | 2018-09-21 | 北京理工大学 | A kind of multi-focus image fusing method based on two scale focused views |
| CN108648209A (en) * | 2018-04-08 | 2018-10-12 | 北京联合大学 | A kind of evaluating method of the centre deviation of saliency data collection |
| CN109753957A (en) * | 2018-12-07 | 2019-05-14 | 东软集团股份有限公司 | Image significance detection method, device, storage medium and electronic equipment |
| CN109872339A (en) * | 2019-01-21 | 2019-06-11 | 哈尔滨理工大学 | A kind of weighting symbiosis image partition method of local correlation |
| CN112200826A (en) * | 2020-10-15 | 2021-01-08 | 北京科技大学 | Industrial weak defect segmentation method |
| CN112785607A (en) * | 2021-01-28 | 2021-05-11 | 太原科技大学 | Method for cutting leaves of field plants |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101425182A (en) * | 2008-11-28 | 2009-05-06 | 华中科技大学 | Image object segmentation method |
| CN101520894A (en) * | 2009-02-18 | 2009-09-02 | 上海大学 | Method for extracting significant object based on region significance |
| CN101706780A (en) * | 2009-09-03 | 2010-05-12 | 北京交通大学 | Image semantic retrieving method based on visual attention model |
| CN105404888A (en) * | 2015-11-16 | 2016-03-16 | 浙江大学 | Saliency object detection method integrated with color and depth information |
| US9412175B2 (en) * | 2014-02-20 | 2016-08-09 | Nokia Technologies Oy | Method, apparatus and computer program product for image segmentation |
| CN106096615A (en) * | 2015-11-25 | 2016-11-09 | 北京邮电大学 | A kind of salient region of image extracting method based on random walk |
| KR20160134117A (en) * | 2015-05-14 | 2016-11-23 | 한국외국어대학교 연구산학협력단 | Semi-automatic segmentation apparatus and method for medical images using an organic seed |
| CN106327507A (en) * | 2016-08-10 | 2017-01-11 | 南京航空航天大学 | Color image significance detection method based on background and foreground information |
-
2017
- 2017-04-11 CN CN201710233611.4A patent/CN107016682B/en not_active Expired - Fee Related
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101425182A (en) * | 2008-11-28 | 2009-05-06 | 华中科技大学 | Image object segmentation method |
| CN101520894A (en) * | 2009-02-18 | 2009-09-02 | 上海大学 | Method for extracting significant object based on region significance |
| CN101706780A (en) * | 2009-09-03 | 2010-05-12 | 北京交通大学 | Image semantic retrieving method based on visual attention model |
| US9412175B2 (en) * | 2014-02-20 | 2016-08-09 | Nokia Technologies Oy | Method, apparatus and computer program product for image segmentation |
| KR20160134117A (en) * | 2015-05-14 | 2016-11-23 | 한국외국어대학교 연구산학협력단 | Semi-automatic segmentation apparatus and method for medical images using an organic seed |
| CN105404888A (en) * | 2015-11-16 | 2016-03-16 | 浙江大学 | Saliency object detection method integrated with color and depth information |
| CN106096615A (en) * | 2015-11-25 | 2016-11-09 | 北京邮电大学 | A kind of salient region of image extracting method based on random walk |
| CN106327507A (en) * | 2016-08-10 | 2017-01-11 | 南京航空航天大学 | Color image significance detection method based on background and foreground information |
Non-Patent Citations (1)
| Title |
|---|
| 赵宏伟 等: "视觉显著目标的自适应分割", 《光学 精密工程》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107680109A (en) * | 2017-09-15 | 2018-02-09 | 盐城禅图智能科技有限公司 | It is a kind of to quote inverse notice and the image, semantic dividing method of pixel similarity study |
| CN107818566A (en) * | 2017-10-27 | 2018-03-20 | 中山大学 | A kind of image partition method based on partial structurtes information around random walk combination pixel |
| CN108564559A (en) * | 2018-03-14 | 2018-09-21 | 北京理工大学 | A kind of multi-focus image fusing method based on two scale focused views |
| CN108564559B (en) * | 2018-03-14 | 2021-07-20 | 北京理工大学 | Multi-focus image fusion method based on two-scale focus image |
| CN108648209B (en) * | 2018-04-08 | 2021-06-29 | 北京联合大学 | Method for evaluating central deviation of significance data set |
| CN108648209A (en) * | 2018-04-08 | 2018-10-12 | 北京联合大学 | A kind of evaluating method of the centre deviation of saliency data collection |
| CN109753957A (en) * | 2018-12-07 | 2019-05-14 | 东软集团股份有限公司 | Image significance detection method, device, storage medium and electronic equipment |
| CN109872339A (en) * | 2019-01-21 | 2019-06-11 | 哈尔滨理工大学 | A kind of weighting symbiosis image partition method of local correlation |
| CN109872339B (en) * | 2019-01-21 | 2021-04-02 | 哈尔滨理工大学 | Locally-correlated weighted symbiotic image segmentation method |
| CN112200826A (en) * | 2020-10-15 | 2021-01-08 | 北京科技大学 | Industrial weak defect segmentation method |
| CN112200826B (en) * | 2020-10-15 | 2023-11-28 | 北京科技大学 | Industrial weak defect segmentation method |
| CN112785607A (en) * | 2021-01-28 | 2021-05-11 | 太原科技大学 | Method for cutting leaves of field plants |
| CN112785607B (en) * | 2021-01-28 | 2021-09-21 | 太原科技大学 | Method for cutting leaves of field plants |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107016682B (en) | 2020-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107016682A (en) | A kind of notable object self-adapting division method of natural image | |
| Wang et al. | RGB-D salient object detection via minimum barrier distance transform and saliency fusion | |
| Kim et al. | Optimized contrast enhancement for real-time image and video dehazing | |
| Xiao et al. | Fast image dehazing using guided joint bilateral filter | |
| Zhang et al. | Binary stereo matching | |
| WO2020165557A1 (en) | 3d face reconstruction system and method | |
| Zhang et al. | Stereo matching with segmentation-based cooperation | |
| CN111402170B (en) | Image enhancement method, device, terminal and computer readable storage medium | |
| CN111767920B (en) | Method and device for extracting region of interest, electronic equipment and storage medium | |
| CN106709964B (en) | Sketch generation method and device based on gradient correction and multidirectional texture extraction | |
| TW200834459A (en) | Video object segmentation method applied for rainy situations | |
| US11978216B2 (en) | Patch-based image matting using deep learning | |
| CN101789124A (en) | Segmentation method for space-time consistency of video sequence of parameter and depth information of known video camera | |
| Li et al. | A Simple Framework for Face Photo‐Sketch Synthesis | |
| Xu et al. | Depth-aided exemplar-based hole filling for DIBR view synthesis | |
| CN112101386A (en) | Text detection method and device, computer equipment and storage medium | |
| CN109255792A (en) | A kind of dividing method of video image, device, terminal device and storage medium | |
| Serna et al. | Attribute controlled reconstruction and adaptive mathematical morphology | |
| Mukherjee et al. | A hybrid algorithm for disparity calculation from sparse disparity estimates based on stereo vision | |
| CN106778789B (en) | A kind of fast target extracting method in multi-view image | |
| CN113570615A (en) | Image processing method based on deep learning, electronic equipment and storage medium | |
| CN113256484B (en) | Method and device for performing stylization processing on image | |
| CN108921852A (en) | Unstructured Terrain segmentation network outside double branch rooms based on parallax and plane fitting | |
| Guo et al. | Image saliency detection based on geodesic‐like and boundary contrast maps | |
| Jiang et al. | Gray‐Scale Image Dehazing Guided by Scene Depth Information |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200331 |