CN108921833A - A kind of the markov conspicuousness object detection method and device of two-way absorption - Google Patents
A kind of the markov conspicuousness object detection method and device of two-way absorption Download PDFInfo
- Publication number
- CN108921833A CN108921833A CN201810670321.0A CN201810670321A CN108921833A CN 108921833 A CN108921833 A CN 108921833A CN 201810670321 A CN201810670321 A CN 201810670321A CN 108921833 A CN108921833 A CN 108921833A
- Authority
- CN
- China
- Prior art keywords
- node
- closed loop
- graph model
- matrix
- super
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of markov conspicuousness object detection method of two-way absorption, the method includes：A：Image to be detected is split using SLIC algorithm, obtains the set of m super-pixel block；B：Using the borderline super-pixel block of described image to be detected as node is absorbed, the prospect probability of the transfering node in described image to be detected is obtained；C：The background probability for obtaining the transfering node in described image to be detected is absorbed based on prospect priori；D：According to the background probability of the prospect probability of the transfering node and the transfering node, the significance value of image to be detected is calculated.The embodiment of the invention also discloses a kind of markov conspicuousness object detecting devices of two-way absorption.Using the embodiment of the present invention, the accuracy rate of the conspicuousness detection of image to be detected can be improved.
Description
Technical field
The present invention relates to a kind of conspicuousness detection method and device of image, are more particularly to a kind of Ma Er of two-way absorption
It can husband's conspicuousness object detection method and device.
Background technique
With the fast development of the computer and networks communication technology, image data is more and more.Conspicuousness detection is as meter
The important preprocessing step that calculation machine visual field is used to reduce computation complexity is also widely used.The inspection of conspicuousness target
What is surveyed is that most significant foreground target is oriented from image to be detected.The application field of the technology is especially extensive, such as：Target inspection
Survey and detection, target following etc..
Currently, the existing well-marked target detection method based on Markov chain is using SLIC (simple linear
Iterative clustering, simple linear iteraction cluster) algorithm divides image to be detected and obtains super-pixel block structural map
Node, then by four boundaries carry out duplication obtain absorb node, calculate four boundary nodes soak time, absorption it is faster, inhale
Shorter between time receiving, then the corresponding super-pixel block of the node may be background；, whereas if soak time is longer, then more may be
Well-marked target.
But in addition to four boundaries in a picture, there are many more important informations, such as prior information, and therefore, application is existing
The conspicuousness for the picture that technology detects is not accurate enough.
Summary of the invention
Technical problem to be solved by the present invention lies in provide a kind of markov conspicuousness target inspection of two-way absorption
Method and device is surveyed, the not accurate enough technical problem of the conspicuousness to solve the picture detected in the prior art.
The present invention is to solve above-mentioned technical problem by the following technical programs：
The embodiment of the invention provides a kind of markov conspicuousness object detection method of two-way absorption, the method packets
It includes：
A：Image to be detected is split using SLIC algorithm, obtains the set of m super-pixel block；
B：Using the borderline super-pixel block of described image to be detected as node is absorbed, obtain in described image to be detected
Transfering node prospect probability；
C：The background probability for obtaining the transfering node in described image to be detected is absorbed based on prospect priori；
D：According to the background probability of the prospect probability of the transfering node and the transfering node, image to be detected is calculated
Significance value.
Optionally, the step B, including：
B1：According to the set of the m super-pixel block, the first closed loop graph model G is constructed1(V1,E1), wherein V1It is m
Super-pixel block node set corresponding with node is absorbed；E1The set on the side between each node；
B2：Utilize formula, z1=N1× c calculates the soak time of each transfering node in the first closed loop graph model, wherein
z1For the soak time of each transfering node；N1For the fundamental matrix of the first closed loop graph model；C be element be all 1 to
Amount；
B3：Using formula,Calculate the prospect probability of each transfering node, wherein zfIt is each described
The prospect probability of transfering node；For the normalization soak time of i-th of node in the first closed loop graph model；I is super-pixel
Block serial number.
Optionally, the calculating process of the fundamental matrix of the first closed loop graph model is：
Obtain the weight on each side in the first closed loop graph model；
Using formula,The pass of the first closed loop graph model is constructed according to the weight
Join matrix A1, wherein
For the incidence matrix A of the first closed loop graph model1In element；For the weight on side in the first closed loop graph model；
M1It (i) is the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the first closed loop graph model, wherein
P1For the transfer matrix of the first closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model,
AndDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the first closed loop graph model, wherein Q1It is first
The probability transfer matrix of closed loop graph model；R1For first between transfering node in the first closed loop graph model and absorption node
Transition probability matrix；I1For the first unit matrix of the absorption node composition；0 is the matrix that element is zero；
Utilize formula, N1=(I1-Q1)-1, calculate the fundamental matrix of the first closed loop graph model, wherein N1For the first closed loop figure
The fundamental matrix of model.
Optionally, the step C, including：
C1：According to the set of the m super-pixel block, the second closed loop graph model G is constructed2(V2,E2), wherein V2It is m
Super-pixel block node set corresponding with node is absorbed；E2The set on the side between each node；And it obtains described second and closes
The weight on each side in ring graph model；
C2：Using formula,Calculate the second closed loop graph model
In each transfering node prospect prior information, wherein fiFor the prospect prior information of i-th of transfering node；K is super-pixel block
Number；∑ is summing function；J is the serial number of super-pixel block；I is the serial number of super-pixel block；BC is contour connection value, and For the weight on side in the second closed loop graph model；xiIt is i-th
The value of a node in cielab color space；xjFor the value of j-th of node in cielab color space；σ is constant parameter；
σbFor the first preset value；da(i, j) is the face between i-th of super-pixel block and j-th of super-pixel block in cielab color space
Color characteristic distance；Exp () is the exponential function using the natural truth of a matter bottom of as；ds(i, j) is i-th to surpass in cielab color space
Color space distance between block of pixels and j-th of super-pixel block；σsFor the second preset value；
C3：Utilize formula, D2=i | fi> avg (f) }, select prospect priori node, wherein D2To select prospect
The set of priori node；Avg () is mean function；Avg (f) is the average value of the prospect prior information of each transfering node；
C4：Utilize formula, z2=N2× c calculates the soak time of each transfering node, wherein z2It is each described
The soak time of transfering node；N2For the fundamental matrix of the second closed loop graph model；C is the vector that element is all 1；
C5：Using formula,Calculate the background probability of each transfering node, wherein zbIt is each described
The background probability of transfering node；For the normalization soak time of i-th of node in the second closed loop graph model.
Optionally, the calculating process of the fundamental matrix of the second closed loop graph model is：
Using formula,The pass of the second closed loop graph model is constructed according to the weight
Join matrix A2, wherein
For the incidence matrix A of the first closed loop graph model2In element；For the weight on side in the second closed loop graph model；
M2It (i) is the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the second closed loop graph model, wherein
P2For the transfer matrix of the second closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model,
AndDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the second closed loop graph model, wherein Q2It is second
The probability transfer matrix of closed loop graph model；R2For transfering node in the second closed loop graph model and absorb the second transfer between node
Probability matrix；I2For the unit matrix for absorbing node composition；0 is the matrix that element is zero；
Utilize formula, N2=(I2-Q2)-1, calculate the fundamental matrix of the second closed loop graph model, wherein N2For the second closed loop figure
The fundamental matrix of model.
Optionally, the acquisition process on the side between each node includes：
According to the incidence relation between each node, the side between each node is obtained, wherein the incidence relation
Including：It is associated between the corresponding node of neighbouring super pixels block；Section between the super-pixel block of neighbours' super-pixel block having the same
Point association；It is associated between the corresponding node of borderline super-pixel block；It absorbs and is not associated between node；It will be chosen as absorbing node
Super-pixel block duplication after obtained node, be associated between the super-pixel block；It is described to be chosen as absorbing the super picture of node
The node and be associated with the associated node of the super-pixel block that plain block obtains after replicating.
Optionally, the D step, including：
Using formula,Calculate the conspicuousness of image to be detected
Value, wherein
X is the significance value of image to be detected；siFor the significance value of the corresponding node of i-th of super-pixel block；sjFor jth
The significance value of the corresponding node of a super-pixel block；K is the quantity of super-pixel block.
The embodiment of the invention also provides a kind of markov conspicuousness object detecting device of two-way absorption, described devices
Including：
Divide module and obtains the set of m super-pixel block for being split using SLIC algorithm to image to be detected；
First obtains module, for obtaining using the borderline super-pixel block of described image to be detected as node is absorbed
The prospect probability of transfering node in described image to be detected；
Second obtains module, for absorbing the background for obtaining the transfering node in described image to be detected based on prospect priori
Probability；
Computing module, for calculating according to the prospect probability of the transfering node and the background probability of the transfering node
The significance value of described image to be detected.
Optionally, described first module is obtained, be also used to：
B1：According to the set of the m super-pixel block, the first closed loop graph model G is constructed1(V1,E1), wherein V1It is m
Super-pixel block node set corresponding with node is absorbed；E1The set on the side between each node；
B2：Utilize formula, z1=N1× c calculates the soak time of each transfering node, wherein z1For each transfering node
Soak time；N1For the fundamental matrix of the first closed loop graph model；C is the vector that element is all 1；
B3：Using formula,Calculate the prospect probability of each transfering node, wherein zfFor each transfering node
Prospect probability；For the normalization soak time of i-th of node in the first closed loop graph model；I is super-pixel block serial number.
Optionally, described first module is obtained, be also used to：
Obtain the weight on each side in the first closed loop graph model；
Using formula,The pass of the first closed loop graph model is constructed according to the weight
Join matrix A1, wherein
For the incidence matrix A of the first closed loop graph model1In element；For the weight on side in the first closed loop graph model；
M1It (i) is the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the first closed loop graph model, wherein
P1For the transfer matrix of the first closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model,
AndDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the first closed loop graph model, wherein Q1It is first
The probability transfer matrix of closed loop graph model；R1For transfering node and absorb the transition probability matrix between node；I1To absorb node
The unit matrix of composition；0 is the matrix that element is zero；
Utilize formula, N1=(I1-Q1)-1, calculate the fundamental matrix of the first closed loop graph model, wherein N1For the first closed loop figure
The fundamental matrix of model.
The present invention has the following advantages that compared with prior art：
It by the conspicuousness detection means for giving Markov chain and is based on using the embodiment of the present invention by majorized function
The conspicuousness monitoring means of prior information is organically combined, and the markov conspicuousness target detection side of two-way absorption is obtained
Method can make background and prospect more obvious, therefore improve the accuracy rate of the conspicuousness monitoring of image.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of the markov conspicuousness object detection method of two-way absorption；
Fig. 2 is a kind of schematic illustration of the markov conspicuousness object detection method of two-way absorption；
Fig. 3 is a kind of structural schematic diagram of the markov conspicuousness object detecting device of two-way absorption.
Specific embodiment
It elaborates below to the embodiment of the present invention, the present embodiment carries out under the premise of the technical scheme of the present invention
Implement, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to following implementation
Example.
The embodiment of the invention provides a kind of markov conspicuousness object detection method of two-way absorption and device, below
A kind of markov conspicuousness object detection method of the two-way absorption provided first with regard to first aspect of the embodiment of the present invention carries out
It introduces.
Fig. 1 is a kind of flow diagram of the markov conspicuousness object detection method of two-way absorption；Fig. 2 is a kind of double
To the schematic illustration of the markov conspicuousness object detection method of absorption；As depicted in figs. 1 and 2, the method includes：
S101：Image to be detected is split using SLIC algorithm, obtains the set of m super-pixel block.
In practical applications, using SLIC, (simple linear iterative clustering simply linearly changes
Generation cluster) algorithm is split image to be detected, the super-pixel block set that available m super-pixel block forms.
S102：Using the borderline super-pixel block of described image to be detected as node is absorbed, the mapping to be checked is obtained
The prospect probability of transfering node as in.
Illustratively, image to be detected is rectangle, and there are four sides for tool.It is borderline using 4 of image to be detected respectively
Super-pixel block is as absorption node.
B1：According to the set of the m super-pixel block, by the borderline k in image to be detected1A super-pixel block H1's
Replica node is as virtual absorption node D1, and by k1A super-pixel block replicated after D1The structure together with m super-pixel block
Build the first closed loop graph model G1(V1,E1), wherein V1For m super-pixel block with absorb the corresponding node set of node, and H1,D1
∈V1。E1The set on the side between each node.It is understood that having n in the first closed loop graph model1A node, and n1
=k1+ m, in addition, the first closed loop graph model characterization be image to be detected boundary absorb structural map.
In practical applications, the side between each node can be obtained according to the incidence relation between each node,
Wherein, the incidence relation includes：It is associated between the corresponding node of neighbouring super pixels block in image to be detected；Neighbour having the same
Occupy the node association between the super-pixel block of super-pixel block；It is associated between the corresponding node of borderline super-pixel block；Duplication
It is not associated with as between the node for absorbing node.
The node that will be obtained after the super-pixel block duplication for being chosen as absorbing node, is associated between the super-pixel block；Institute
It states the node obtained after the super-pixel block duplication for being chosen as absorbing node and is associated with the associated node of the super-pixel block.Example
Such as, it is assumed that i is the corresponding node of super-pixel block on image inner boundary, then replicate generate new node i ', it is associated with node i
All nodes are associated between node i ' be associated with, and node i and node i '.
B2：Utilize formula, z1=N1× c calculates the soak time of each transfering node in the first closed loop graph model, wherein
z1For the soak time of each transfering node；N1For the fundamental matrix of the first closed loop graph model；C be element be all 1 to
Amount.
B3：Using formula,Calculate the prospect probability of each transfering node, wherein zfIt is each described
The prospect probability of transfering node；For the normalization soak time of i-th of node in the first closed loop graph model；I is super-pixel
Block serial number.
Specifically, the calculating process of the fundamental matrix of the first closed loop graph model is：It obtains in the first closed loop graph model
The weight on each side；Using formula,The first closed loop artwork is constructed according to the weight
The incidence matrix A of type1, whereinFor the incidence matrix A of the first closed loop graph model1In element；For the first closed loop graph model
The weight on middle side；M1It (i) is the set of the point adjacent with node i；Using formula,Calculate the first closed loop artwork
The transfer matrix of type, wherein P1For the transfer matrix of the first closed loop graph model；For the incidence matrix of the first closed loop graph model
Diagonal matrix, andDiag () is diagonal matrix function；∑ is summing function；Using formula,Calculate the probability transfer matrix of the first closed loop graph model, wherein Q1Turn for the probability of the first closed loop graph model
Move matrix；R1For transfering node in the first closed loop graph model and absorb the first transition probability matrix between node；I1For institute
State the first unit matrix for absorbing node composition；0 is the matrix that element is zero, can be k1* m ties up matrix；Utilize formula, N1=
(I1-Q1)-1, calculate the fundamental matrix of the first closed loop graph model, wherein N1For the fundamental matrix of the first closed loop graph model.This is basic
In matrix, nijIt is the expection soak time from transfering node i to transfering node j for the element of fundamental matrix；For node i
Total soak time.
S103：The background probability for obtaining the transfering node in described image to be detected is absorbed based on prospect priori.
Specifically, S103 step may include：
C1：According to the set of the m super-pixel block, the second closed loop graph model G is constructed2(V2,E2), wherein V2It is m
Super-pixel block node set corresponding with node is absorbed；E2The set on the side between each node；And it obtains described second and closes
The weight on each side in ring graph model；
C2：Using formula,Calculate the second closed loop graph model
In each transfering node prospect prior information, wherein fiFor the prospect prior information of i-th of transfering node；K is super-pixel block
Number；∑ is summing function；J is the serial number of super-pixel block；I is the serial number of super-pixel block；BC is contour connection value, and For the weight on side in the second closed loop graph model；xiIt is i-th
The value of a node in cielab color space；xjFor the value of j-th of node in cielab color space；σ is constant parameter；
σbIt can be 1 for the first preset value；da(i, j) is i-th of super-pixel block and j-th of super-pixel block in cielab color space
Between color characteristic distance；Exp () is the exponential function using the natural truth of a matter bottom of as；ds(i, j) is in CIELAB color space
In color space distance between i-th of super-pixel block and j-th of super-pixel block；σsIt can be 0.25 for the second preset value；
C3：Utilize formula, D2=i | fi> avg (f) }, select prospect priori node, wherein D2To select prospect
The set of priori node；Avg () is mean function；Avg (f) is the average value of the prospect prior information of each transfering node.
Detailed process can be, by fiThe node of > avg (f) is as prospect priori region H to be copied2, then by prospect to be copied
Priori region is replicated, the set of the dummy node after being replicated, D2, and H2,D2∈V2。
C4：Utilize formula, z2=N2× c calculates the soak time of each transfering node, wherein z2It is each described
The soak time of transfering node；N2For the second closed loop graph model G2(V2,E2) fundamental matrix；C is the vector that element is all 1；V2
For m super-pixel block node set corresponding with node is absorbed；E2The set on the side between each node.
C4：Using formula,Calculate the background probability of each transfering node, wherein zbFor each transfer
The background probability of node；For the normalization soak time of i-th of node in the second closed loop graph model.
In practical applications, it can use formula,It is constructed according to the weight
The incidence matrix A of second closed loop graph model2, wherein
For the incidence matrix A of the first closed loop graph model2In element；For the weight on side in the second closed loop graph model；
M2It (i) is the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the second closed loop graph model, wherein
P2For the transfer matrix of the second closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model,
AndDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the second closed loop graph model, wherein Q2It is second
The probability transfer matrix of closed loop graph model；R2For transfering node in the second closed loop graph model and absorb the second transfer between node
Probability matrix；I2For the unit matrix for absorbing node composition；0 is the matrix that element is zero；
Utilize formula, N2=(I2-Q2)-1, calculate the fundamental matrix of the second closed loop graph model, wherein N2For the second closed loop figure
The fundamental matrix of model.
It is emphasized that according to the incidence relation between each node in S103 step, obtain each node it
Between side process and the embodiment of the present invention in S102 according to the incidence relation between each node, obtain each node
Between side process it is consistent；In addition, the calculating process of the fundamental matrix of the second closed loop graph model is closed in S102 step first
The calculating process of the fundamental matrix of ring graph model is consistent, and which is not described herein again.
S104：According to the background probability of the prospect probability of the transfering node and the transfering node, each transfer is calculated
The significance value of node.
Specifically, the D step, including：
Using formula,Calculate the conspicuousness of image to be detected
Value, wherein X is the significance value of image to be detected；siFor the significance value of the corresponding node of i-th of super-pixel block；sjFor jth
The significance value of the corresponding node of a super-pixel block；K is the quantity of super-pixel block.
It should be noted that first item in above-mentioned formula by background probability obtain close to 0 value si, Section 2 is
By prospect probability obtain close to 1 value si, Section 3 is the function of definition, for obtaining successional saliency value.It is two-way
The markov of absorption, i.e., in four boundary absorption processes, absorption it is faster, soak time is shorter, then the node is corresponding super
Block of pixels may be background, more may be well-marked target if soak time is longer；Otherwise in prospect priori absorption process
In, absorption it is faster, soak time is shorter, then the corresponding super-pixel block of the node may be prospect, if soak time is longer,
It then more may be background.In practical applications, each node can be shown according to the saliency value of each node, obtains mesh
Mark notable figure.
The conspicuousness detection means of Markov chain will be given by majorized function using embodiment illustrated in fig. 1 of the present invention
Conspicuousness monitoring means based on prior information is organically combined, and the markov conspicuousness target of two-way absorption is obtained
Detection method can make background and prospect more obvious, therefore improve the accuracy rate of the conspicuousness monitoring of image.
Second aspect, corresponding with first aspect present invention embodiment, the embodiment of the invention also provides a kind of two-way suctions
The markov conspicuousness object detecting device of receipts.
Fig. 3 is a kind of structural schematic diagram of the markov conspicuousness object detecting device of two-way absorption, as shown in figure 3,
Described device includes：
Divide module 301 and obtains the collection of m super-pixel block for being split using SLIC algorithm to image to be detected
It closes；
First obtains module 302, for obtaining using the borderline super-pixel block of described image to be detected as node is absorbed
Take the prospect probability of the transfering node in described image to be detected；
Second obtains module 303, obtains the transfering node in described image to be detected for absorbing based on prospect priori
Background probability；
Computing module 304, for according to the prospect probability of the transfering node and the background probability of the transfering node, meter
Calculate the significance value of each transfering node.
The conspicuousness detection means of Markov chain will be given by majorized function using embodiment illustrated in fig. 3 of the present invention
Conspicuousness monitoring means based on prior information is organically combined, and the markov conspicuousness target of two-way absorption is obtained
Detection method can make background and prospect more obvious, therefore improve the accuracy rate of the conspicuousness monitoring of image.
In a kind of specific embodiment of the embodiment of the present invention, described first obtains module 302, is also used to：
B1：According to the set of the m super-pixel block, the first closed loop graph model G is constructed1(V1,E1), wherein V1It is m
Super-pixel block node set corresponding with node is absorbed；E1The set on the side between each node；
B2：Utilize formula, z1=N1× c calculates the soak time of each transfering node, wherein z1For each transfering node
Soak time；N1For the fundamental matrix of the first closed loop graph model；C is the vector that element is all 1；
B3：Using formula,Calculate the prospect probability of each transfering node, wherein zfFor each transfering node
Prospect probability；For the normalization soak time of i-th of node in the first closed loop graph model；I is super-pixel block serial number.
In a kind of specific embodiment of the embodiment of the present invention, described first obtains module 302, is also used to：
Obtain the weight on each side in the first closed loop graph model；
Using formula,The pass of the first closed loop graph model is constructed according to the weight
Join matrix A1, wherein
For the incidence matrix A of the first closed loop graph model1In element；For the weight on side in the first closed loop graph model；M1
It (i) is the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the first closed loop graph model, wherein
P1For the transfer matrix of the first closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model,
AndDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the first closed loop graph model, wherein Q1It is first
The probability transfer matrix of closed loop graph model；R1For transfering node and absorb the transition probability matrix between node；I1To absorb node
The unit matrix of composition；0 is the matrix that element is zero；
Utilize formula, N1=(I1-Q1)-1, calculate the fundamental matrix of the first closed loop graph model, wherein N1For the first closed loop figure
The fundamental matrix of model.
In a kind of specific embodiment of the embodiment of the present invention, described second obtains module 303, is specifically used for：
C1：According to the set of the m super-pixel block, the second closed loop graph model G is constructed2(V2,E2), wherein V2It is m
Super-pixel block node set corresponding with node is absorbed；E2The set on the side between each node；And it obtains described second and closes
The weight on each side in ring graph model；
C2：Using formula,Calculate the second closed loop graph model
In each transfering node prospect prior information, wherein fiFor the prospect prior information of i-th of transfering node；K is super-pixel block
Number；∑ is summing function；J is the serial number of super-pixel block；I is the serial number of super-pixel block；BC is contour connection value, and For the weight on side in the second closed loop graph model；xiIt is i-th
The value of a node in cielab color space；xjFor the value of j-th of node in cielab color space；σ is constant parameter；
σbFor the first preset value；da(i, j) is the face between i-th of super-pixel block and j-th of super-pixel block in cielab color space
Color characteristic distance；Exp () is the exponential function using the natural truth of a matter bottom of as；ds(i, j) is i-th to surpass in cielab color space
Color space distance between block of pixels and j-th of super-pixel block；σsFor the second preset value；
C3：Utilize formula, D2=i | fi> avg (f) }, select prospect priori node, wherein D2To select prospect
The set of priori node；Avg () is mean function；Avg (f) is the flat of the prospect prior information of each transfering node
Mean value；
C4：Utilize formula, z2=N2× c calculates the soak time of each transfering node, wherein z2It is each described
The soak time of transfering node；N2For the fundamental matrix of the second closed loop graph model；C is the vector that element is all 1；
C5：Using formula,Calculate the background probability of each transfering node, wherein zbIt is each described
The background probability of transfering node；For the normalization soak time of i-th of node in the second closed loop graph model.
In a kind of specific embodiment of the embodiment of the present invention, described second obtains module 303, is specifically used for：
Using formula,The pass of the second closed loop graph model is constructed according to the weight
Join matrix A2, wherein
For the incidence matrix A of the first closed loop graph model2In element；For the weight on side in the second closed loop graph model；M2
It (i) is the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the second closed loop graph model, wherein
P2For the transfer matrix of the second closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model,
AndDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the second closed loop graph model, wherein Q2It is second
The probability transfer matrix of closed loop graph model；R2For transfering node in the second closed loop graph model and absorb the second transfer between node
Probability matrix；I2For the unit matrix for absorbing node composition；0 is the matrix that element is zero；
Utilize formula, N2=(I2-Q2)-1, calculate the fundamental matrix of the second closed loop graph model, wherein N2For the second closed loop figure
The fundamental matrix of model.
In a kind of specific embodiment of the embodiment of the present invention, the first acquisition module 302 or described second obtain
Modulus block 303, is specifically used for：
According to the incidence relation between each node, the side between each node is obtained, wherein the incidence relation
Including：It is associated between the corresponding node of neighbouring super pixels block；Section between the super-pixel block of neighbours' super-pixel block having the same
Point association；It is associated between the corresponding node of borderline super-pixel block；It absorbs and is not associated between node；It will be chosen as absorbing node
Super-pixel block duplication after obtained node, be associated between the super-pixel block；It is described to be chosen as absorbing the super picture of node
The node and be associated with the associated node of the super-pixel block that plain block obtains after replicating.
In a kind of specific embodiment of the embodiment of the present invention, the computing module 304 is specifically used for：
Using formula,Calculate the conspicuousness of image to be detected
Value, wherein
X is the significance value of image to be detected；siFor the significance value of the corresponding node of i-th of super-pixel block；sjFor jth
The significance value of the corresponding node of a super-pixel block；K is the quantity of super-pixel block.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of markov conspicuousness object detection method of two-way absorption, which is characterized in that the method includes：
A：Image to be detected is split using SLIC algorithm, obtains the set of m super-pixel block；
B：Using the borderline super-pixel block of described image to be detected as node is absorbed, obtains and turn in described image to be detected
Move the prospect probability of node；
C：The background probability for obtaining the transfering node in described image to be detected is absorbed based on prospect priori；
D：According to the background probability of the prospect probability of the transfering node and the transfering node, the significant of image to be detected is calculated
Property value.
2. a kind of markov conspicuousness object detection method of two-way absorption according to claim 1, which is characterized in that
The step B, including：
B1：According to the set of the m super-pixel block, the first closed loop graph model G is constructed1(V1,E1), wherein V1For m super-pixel
Block node set corresponding with node is absorbed；E1The set on the side between each node；
B2：Utilize formula, z1=N1× c calculates the soak time of each transfering node in the first closed loop graph model, wherein z1For
The soak time of each transfering node；N1For the fundamental matrix of the first closed loop graph model；C is the vector that element is all 1；
B3：Using formula,Calculate the prospect probability of each transfering node, wherein zfFor each transfer
The prospect probability of node；For the normalization soak time of i-th of node in the first closed loop graph model；I is super-pixel block sequence
Number.
3. a kind of markov conspicuousness object detection method of two-way absorption according to claim 2, which is characterized in that
The calculating process of the fundamental matrix of first closed loop graph model is：
Obtain the weight on each side in the first closed loop graph model；
Using formula,The association square of the first closed loop graph model is constructed according to the weight
Battle array A1, wherein
For the incidence matrix A of the first closed loop graph model1In element；For the weight on side in the first closed loop graph model；M1(i)
For the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the first closed loop graph model, wherein
P1For the transfer matrix of the first closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model, andDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the first closed loop graph model, wherein Q1For the first closed loop figure
The probability transfer matrix of model；R1The first transfer between transfering node in the first closed loop graph model and absorption node is general
Rate matrix；I1For the first unit matrix of the absorption node composition；0 is the matrix that element is zero；
Utilize formula, N1=(I1-Q1)-1, calculate the fundamental matrix of the first closed loop graph model, wherein N1For the first closed loop graph model
Fundamental matrix.
4. a kind of markov conspicuousness object detection method of two-way absorption according to claim 1, which is characterized in that
The step C, including：
C1：According to the set of the m super-pixel block, the second closed loop graph model G is constructed2(V2,E2), wherein V2For m super-pixel
Block node set corresponding with node is absorbed；E2The set on the side between each node；And obtain the second closed loop artwork
The weight on each side in type；
C2：Using formula,It calculates each in the second closed loop graph model
The prospect prior information of a transfering node, wherein fiFor the prospect prior information of i-th of transfering node；K is of super-pixel block
Number；∑ is summing function；J is the serial number of super-pixel block；I is the serial number of super-pixel block；BC is contour connection value, and For the weight on side in the second closed loop graph model；xiIt is i-th
The value of a node in cielab color space；xjFor the value of j-th of node in cielab color space；σ is constant parameter；
σbFor the first preset value；da(i, j) is the face between i-th of super-pixel block and j-th of super-pixel block in cielab color space
Color characteristic distance；Exp () is the exponential function using the natural truth of a matter bottom of as；ds(i, j) is i-th to surpass in cielab color space
Color space distance between block of pixels and j-th of super-pixel block；σsFor the second preset value；
C3：Utilize formula, D2=i | fi> avg (f) }, select prospect priori node, wherein D2To select prospect priori
The set of node；Avg () is mean function；Avg (f) is the average value of the prospect prior information of each transfering node；
C4：Utilize formula, z2=N2× c calculates the soak time of each transfering node, wherein z2For each transfer
The soak time of node；N2For the fundamental matrix of the second closed loop graph model；C is the vector that element is all 1；
C5：Using formula,Calculate the background probability of each transfering node, wherein zbFor each transfer
The background probability of node；For the normalization soak time of i-th of node in the second closed loop graph model.
5. a kind of markov conspicuousness object detection method of two-way absorption according to claim 4, which is characterized in that
The calculating process of the fundamental matrix of second closed loop graph model is：
Using formula,The association square of the second closed loop graph model is constructed according to the weight
Battle array A2, wherein
For the incidence matrix A of the first closed loop graph model2In element；For the weight on side in the second closed loop graph model；M2(i)
For the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the second closed loop graph model, wherein
P2For the transfer matrix of the second closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model, andDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the second closed loop graph model, wherein Q2For the second closed loop
The probability transfer matrix of graph model；R2For transfering node in the second closed loop graph model and absorb the second transition probability between node
Matrix；I2For the unit matrix for absorbing node composition；0 is the matrix that element is zero；
Utilize formula, N2=(I2-Q2)-1, calculate the fundamental matrix of the second closed loop graph model, wherein N2For the second closed loop graph model
Fundamental matrix.
6. a kind of markov conspicuousness object detection method of two-way absorption according to claim 2 or 4, feature exist
In the acquisition process on the side between each node includes：
According to the incidence relation between each node, the side between each node is obtained, wherein the incidence relation packet
It includes：It is associated between the corresponding node of neighbouring super pixels block；Node between the super-pixel block of neighbours' super-pixel block having the same
Association；It is associated between the corresponding node of borderline super-pixel block；It absorbs and is not associated between node；It will be chosen as absorbing node
The node obtained after super-pixel block duplication, is associated between the super-pixel block；It is described to be chosen as absorbing the super-pixel of node
The node and be associated with the associated node of the super-pixel block that block obtains after replicating.
7. a kind of markov conspicuousness object detection method of two-way absorption according to claim 1, which is characterized in that
The D step, including：
Using formula,The significance value of image to be detected is calculated,
In,
X is the significance value of image to be detected；siFor the significance value of the corresponding node of i-th of super-pixel block；sjJ-th to surpass
The significance value of the corresponding node of block of pixels；K is the quantity of super-pixel block.
8. a kind of markov conspicuousness object detecting device of two-way absorption, which is characterized in that described device includes：
Divide module and obtains the set of m super-pixel block for being split using SLIC algorithm to image to be detected；
First obtains module, for using the borderline super-pixel block of described image to be detected as absorption node, described in acquisition
The prospect probability of transfering node in image to be detected；
Second obtains module, and the background for absorbing the transfering node in the described image to be detected of acquisition based on prospect priori is general
Rate；
Computing module, for according to the prospect probability of the transfering node and the background probability of the transfering node, described in calculating
The significance value of image to be detected.
9. a kind of markov conspicuousness object detecting device of two-way absorption according to claim 8, which is characterized in that
Described first obtains module, is also used to：
B1：According to the set of the m super-pixel block, the first closed loop graph model G is constructed1(V1,E1), wherein V1For m super-pixel
Block node set corresponding with node is absorbed；E1The set on the side between each node；
B2：Utilize formula, z1=N1× c calculates the soak time of each transfering node, wherein z1For the suction of each transfering node
Between time receiving；N1For the fundamental matrix of the first closed loop graph model；C is the vector that element is all 1；
B3：Using formula,Calculate the prospect probability of each transfering node, wherein zfBefore each transfering node
Scape probability；For the normalization soak time of i-th of node in the first closed loop graph model；I is super-pixel block serial number.
10. a kind of markov conspicuousness object detection method of two-way absorption according to claim 9, feature exist
In described first obtains module, is also used to：
Obtain the weight on each side in the first closed loop graph model；
Using formula,The association square of the first closed loop graph model is constructed according to the weight
Battle array A1, wherein
For the incidence matrix A of the first closed loop graph model1In element；For the weight on side in the first closed loop graph model；M1(i)
For the set of the point adjacent with node i；
Using formula,Calculate the transfer matrix of the first closed loop graph model, wherein
P1For the transfer matrix of the first closed loop graph model；For the diagonal matrix of the incidence matrix of the first closed loop graph model, andDiag () is diagonal matrix function；∑ is summing function；
Using formula,Calculate the probability transfer matrix of the first closed loop graph model, wherein Q1For the first closed loop figure
The probability transfer matrix of model；R1For transfering node and absorb the first transition probability matrix between node；I1To absorb node group
At unit matrix；0 is the matrix that element is zero；
Utilize formula, N1=(I-Q1)-1, calculate the fundamental matrix of the first closed loop graph model, wherein N1For the first closed loop graph model
Fundamental matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810670321.0A CN108921833B (en) | 2018-06-26 | 2018-06-26 | Two-way absorption Markov significance target detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810670321.0A CN108921833B (en) | 2018-06-26 | 2018-06-26 | Two-way absorption Markov significance target detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108921833A true CN108921833A (en) | 2018-11-30 |
| CN108921833B CN108921833B (en) | 2022-03-25 |
Family
ID=64421358
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810670321.0A Active CN108921833B (en) | 2018-06-26 | 2018-06-26 | Two-way absorption Markov significance target detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921833B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110097569A (en) * | 2019-04-04 | 2019-08-06 | 北京航空航天大学 | Oil tank object detection method based on color Markov Chain conspicuousness model |
| CN110111353A (en) * | 2019-04-29 | 2019-08-09 | 河海大学 | A kind of image significance detection method absorbing chain based on markov background and prospect |
| CN111310768A (en) * | 2020-01-20 | 2020-06-19 | 安徽大学 | Saliency target detection method based on robustness background prior and global information |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140270499A1 (en) * | 2011-12-08 | 2014-09-18 | Olympus Corporation | Image processing apparatus, image processing method, and computer-readable recording device |
| US20140365503A1 (en) * | 2013-06-11 | 2014-12-11 | International Business Machines Corporation | Estimation of closeness of topics based on graph analytics |
| CN105426895A (en) * | 2015-11-10 | 2016-03-23 | 河海大学 | Prominence detection method based on Markov model |
| CN106599668A (en) * | 2016-12-29 | 2017-04-26 | 中国科学院长春光学精密机械与物理研究所 | Target identity identification system |
| CN106780430A (en) * | 2016-11-17 | 2017-05-31 | 大连理工大学 | A kind of image significance detection method based on surroundedness and Markov model |
| CN106815843A (en) * | 2016-11-30 | 2017-06-09 | 江苏城乡建设职业学院 | A kind of fruit object acquisition methods based on convex closure center priori and absorbing Marcov chain |
| CN107609552A (en) * | 2017-08-23 | 2018-01-19 | 西安电子科技大学 | Salient region detection method based on markov absorbing model |
| CN107749053A (en) * | 2017-10-24 | 2018-03-02 | 郑州布恩科技有限公司 | A kind of binocular image collection and pretreatment unit and method for vision prosthesis |
| CN107862702A (en) * | 2017-11-24 | 2018-03-30 | 大连理工大学 | A kind of conspicuousness detection method of combination boundary connected and local contrast |
-
2018
- 2018-06-26 CN CN201810670321.0A patent/CN108921833B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140270499A1 (en) * | 2011-12-08 | 2014-09-18 | Olympus Corporation | Image processing apparatus, image processing method, and computer-readable recording device |
| US20140365503A1 (en) * | 2013-06-11 | 2014-12-11 | International Business Machines Corporation | Estimation of closeness of topics based on graph analytics |
| CN105426895A (en) * | 2015-11-10 | 2016-03-23 | 河海大学 | Prominence detection method based on Markov model |
| CN106780430A (en) * | 2016-11-17 | 2017-05-31 | 大连理工大学 | A kind of image significance detection method based on surroundedness and Markov model |
| CN106815843A (en) * | 2016-11-30 | 2017-06-09 | 江苏城乡建设职业学院 | A kind of fruit object acquisition methods based on convex closure center priori and absorbing Marcov chain |
| CN106599668A (en) * | 2016-12-29 | 2017-04-26 | 中国科学院长春光学精密机械与物理研究所 | Target identity identification system |
| CN107609552A (en) * | 2017-08-23 | 2018-01-19 | 西安电子科技大学 | Salient region detection method based on markov absorbing model |
| CN107749053A (en) * | 2017-10-24 | 2018-03-02 | 郑州布恩科技有限公司 | A kind of binocular image collection and pretreatment unit and method for vision prosthesis |
| CN107862702A (en) * | 2017-11-24 | 2018-03-30 | 大连理工大学 | A kind of conspicuousness detection method of combination boundary connected and local contrast |
Non-Patent Citations (7)
| Title |
|---|
| BOWEN JIANG等: "Saliency Detection via Absorbing Markov Chain", 《2013 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| RADHAKRISHNA ACHANTA等: "SLIC Superpixels Compared to State-of-the-Art Superpixel Methods", 《IEEE TRANSACTIONS ON SOFTWARE ENGINEERING》 * |
| SIMON H. TAUSCH等: "RAMBO-K: Rapid and Sensitive Removal of Background Sequences from Next Generation Sequencing Data", 《PLOS ONE | DOI:10.1371/JOURNAL.PONE.0137896》 * |
| WENJIE ZHANG等: "Region saliency detection via multi-feature on absorbing markov chain", 《SPRINGERLINK》 * |
| 张文杰: "基于图像配准与视觉显著性检测的指针仪表识别研究", 《中国优秀博硕士学位论文全文数据库（博士）_信息科技辑》 * |
| 蒋峰岭等: "背景吸收的马尔可夫显著性目标检测", 《中国图象图形学报》 * |
| 陶萍萍: "网格显著性检测中若干方法研究", 《中国优秀博硕士学位论文全文数据库（博士）_信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110097569A (en) * | 2019-04-04 | 2019-08-06 | 北京航空航天大学 | Oil tank object detection method based on color Markov Chain conspicuousness model |
| CN110111353A (en) * | 2019-04-29 | 2019-08-09 | 河海大学 | A kind of image significance detection method absorbing chain based on markov background and prospect |
| CN110111353B (en) * | 2019-04-29 | 2020-01-24 | 河海大学 | Image significance detection method based on Markov background and foreground absorption chain |
| CN111310768A (en) * | 2020-01-20 | 2020-06-19 | 安徽大学 | Saliency target detection method based on robustness background prior and global information |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108921833B (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022002039A1 (en) | Visual positioning method and device based on visual map | |
| CN106446815B (en) | A kind of simultaneous localization and mapping method | |
| Cui et al. | HSfM: Hybrid structure-from-motion | |
| Ahmed et al. | Constructing street networks from GPS trajectories | |
| CN108921833A (en) | A kind of the markov conspicuousness object detection method and device of two-way absorption | |
| Candelieri et al. | A graph based analysis of leak localization in urban water networks | |
| CN107369183A (en) | Towards the MAR Tracing Registration method and system based on figure optimization SLAM | |
| CN108596053A (en) | A kind of vehicle checking method and system based on SSD and vehicle attitude classification | |
| CN103679186B (en) | The method and apparatus of detect and track target | |
| CN105635963B (en) | Multiple agent co-located method | |
| CN103020989A (en) | Multi-view target tracking method based on on-line scene feature clustering | |
| CN104777469B (en) | A kind of radar node selecting method based on error in measurement covariance matrix norm | |
| Cui et al. | Efficient and robust large-scale structure-from-motion via track selection and camera prioritization | |
| CN112254656A (en) | Stereoscopic vision three-dimensional displacement measurement method based on structural surface point characteristics | |
| Pan et al. | Hyperbolic hierarchy-aware knowledge graph embedding for link prediction | |
| Li et al. | Dam surface crack detection based on deep learning | |
| Cui et al. | Tracks selection for robust, efficient and scalable large-scale structure from motion | |
| CN111860592A (en) | Solar cell defect classification detection method under condition of few samples | |
| He et al. | Indoor localization with spatial and temporal representations of signal sequences | |
| Gueguen et al. | Mapping human settlements and population at country scale from VHR images | |
| Zhu et al. | Accurate, scalable and parallel structure from motion | |
| CN114202632A (en) | Grid linear structure recovery method and device, electronic equipment and storage medium | |
| CN108830896A (en) | A kind of remote sensing image water body patch extracting method based on full convolutional network and architectural shadow geometrical model | |
| Zhang et al. | Dense 3d mapping for indoor environment based on feature-point slam method | |
| Haverkort et al. | Simple I/O-efficient flow accumulation on grid terrains |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221212 Address after: 230031 Shushan Lake Road, Shushan District, Hefei, Anhui 350 Patentee after: HEFEI INSTITUTES OF PHYSICAL SCIENCE, CHINESE ACADEMY OF SCIENCES Patentee after: HEFEI NORMAL University Address before: 230031 Shushan Lake Road, Shushan District, Hefei, Anhui 350 Patentee before: HEFEI INSTITUTES OF PHYSICAL SCIENCE, CHINESE ACADEMY OF SCIENCES |