CN109685806A - Image significance detection method and device - Google Patents
Image significance detection method and device Download PDFInfo
- Publication number
- CN109685806A CN109685806A CN201811355398.5A CN201811355398A CN109685806A CN 109685806 A CN109685806 A CN 109685806A CN 201811355398 A CN201811355398 A CN 201811355398A CN 109685806 A CN109685806 A CN 109685806A
- Authority
- CN
- China
- Prior art keywords
- super
- initial
- pixel block
- pixel
- notable
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
Abstract
The embodiment of the present invention provides a kind of image significance detection method and device, the described method includes: extracting multiple features from target image, generate characteristic pattern, carry out the segmentation of multiple scales respectively to the target image based on super-pixel segmentation algorithm, obtain initial super-pixel block, according to each characteristic pattern and each initial super-pixel block, the saliency value of each initial super-pixel block is obtained；The initial notable figure under each scale is generated according to the saliency value of each initial super-pixel block, each initial notable figure is optimized based on ant colony optimization algorithm, obtains the optimal notable figure under each scale；According to the optimal notable figure under each scale, the final notable figure of the target image is obtained.The embodiment of the present invention improves the accuracy that saliency detects under low illumination.
Description
Technical field
The embodiment of the present invention belongs to technical field of image processing, more particularly, to a kind of image significance detection method
And device.
Background technique
The vision significance information of image, which is able to reflect different zones in image, stimulates degree to human visual system.Image is aobvious
The basic research that target detection is computer vision field is write, it can be from the image data of magnanimity by conspicuousness detection
In find out the important area or object that may cause that human eye vision pays attention to.
Most of existing saliency detection model is only applicable to visible light environment, however in actual application scenarios
It often will appear some low photoenvironments, such as the ring very poor by the interference of the weather such as rain, snow, haze or night illumination condition
Border, the influence for the factors such as estimating for resulting in a feature that is easy to be changed by noise jamming and background, texture is fuzzy,
So that the reliability sharp fall of conspicuousness testing result.
Although it is existing research saliency detection on achieve some achievements, due under low illumination signal-to-noise ratio it is low and
Validity feature is limited, and the effect that saliency detects under low illumination is still poor.Therefore how low light image is shown
The detection of work property is still a challenging problem.
Summary of the invention
To overcome the problems, such as saliency detection effect difference under above-mentioned existing low illumination or at least being partially solved
The above problem, the embodiment of the present invention provide a kind of image significance detection method and device.
According to a first aspect of the embodiments of the present invention, a kind of image significance detection method is provided, comprising:
Multiple features are extracted from target image, generate characteristic pattern, based on super-pixel segmentation algorithm to the target image
The segmentation for carrying out multiple scales respectively, obtains initial super-pixel block, according to each characteristic pattern and each initial super-pixel block,
Obtain the saliency value of each initial super-pixel block；
The initial notable figure under each scale is generated according to the saliency value of each initial super-pixel block, it is excellent based on ant colony
Change algorithm to optimize each initial notable figure, obtains the optimal notable figure under each scale；
According to the optimal notable figure under each scale, the final notable figure of the target image is obtained.
Second aspect according to embodiments of the present invention provides a kind of saliency detection device, comprising:
Divide module, for extracting multiple features from target image, generates characteristic pattern, be based on super-pixel segmentation algorithm pair
The target image carries out the segmentation of multiple scales respectively, obtains initial super-pixel block, according to each characteristic pattern and each described
Initial super-pixel block obtains the saliency value of each initial super-pixel block；
Optimization module is initial significant under each scale for being generated according to the saliency value of each initial super-pixel block
Figure, optimizes each initial notable figure based on ant colony optimization algorithm, obtains the optimal notable figure under each scale；
Module is obtained, for obtaining the final significant of the target image according to the optimal notable figure under each scale
Figure.
In terms of third according to an embodiment of the present invention, a kind of electronic equipment is also provided, comprising:
At least one processor；And
At least one processor being connect with the processor communication, in which:
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to refer to
Order significant be able to carry out image provided by any possible implementation in the various possible implementations of first aspect
Property detection method.
4th aspect according to an embodiment of the present invention, also provides a kind of non-transient computer readable storage medium, described
Non-transient computer readable storage medium stores computer instruction, and the computer instruction makes the computer execute first aspect
Various possible implementations in image significance detection method provided by any possible implementation.
The embodiment of the present invention provides a kind of image significance detection method and device, and this method is by using super-pixel segmentation
Target image is divided into multiple initial super-pixel block by algorithm, obtains target image according to the multidimensional characteristic of each initial super-pixel block
Initial notable figure, regard each initial super-pixel block in initial notable figure as a node, graph model constructed, in graph model
On the basis of initial notable figure is optimized using ant colony optimization algorithm, target figure is detected according to the initial notable figure after optimization
The conspicuousness of picture improves the accuracy of low light image conspicuousness detection, fixed for night safe monitoring and complex environment target
The problems such as position, provides support.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is image significance detection method overall flow schematic diagram provided in an embodiment of the present invention；
Fig. 2 is the image significance detection method overall flow schematic diagram that further embodiment of this invention provides；
Fig. 3 is saliency detection device overall structure diagram provided in an embodiment of the present invention；
Fig. 4 is electronic equipment overall structure diagram provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
A kind of image significance detection method is provided in one embodiment of the invention, and Fig. 1 mentions for the embodiment of the present invention
The image significance detection method overall flow schematic diagram of confession, this method comprises: S101, extracts multiple spies from target image
Sign generates characteristic pattern, carries out the segmentation of multiple scales respectively to the target image based on super-pixel segmentation algorithm, obtains initial
Super-pixel block obtains the saliency value of each initial super-pixel block according to each characteristic pattern and each initial super-pixel block；
Wherein, multi-dimension feature extraction is carried out to target, the present embodiment is not limited to extract the type of feature.Super-pixel segmentation is calculated
Method refers to the irregular picture for having certain visual meaningaaa that will have many characteristics, such as similar grain, color, the adjacent pixel composition of brightness
Plain block replaces a large amount of pixel to express figure using the similitude of feature between pixel by group pixels with a small amount of super-pixel
As feature, to reduce the complexity of post processing of image.Target image is the image for needing to carry out conspicuousness detection.Initial super picture
Plain block is the super-pixel that target image is directly split to using super-pixel segmentation algorithm acquisition.Each initial super-pixel block tool
There is a class label.Target image is divided into multiple initial super-pixel block using super-pixel segmentation algorithm, the present embodiment is suitable
For various super-pixel segmentation algorithms.It is special from the multidimensional of target image according to the region of initial super-pixel block in the target image
The multidimensional characteristic that initial super-pixel block is obtained in sign uses global contrast side according to the multidimensional characteristic of each initial super-pixel block
Method obtains the saliency value of each initial super-pixel block.Multiple scales are carried out respectively to target image based on super-pixel segmentation algorithm to divide
It cuts and refers to repeated segmentation is carried out to target image using super-pixel segmentation algorithm, initial super-pixel number is not in each segmentation result
Together.
S102 generates the initial notable figure under each scale according to the saliency value of each initial super-pixel block, is based on
Ant colony optimization algorithm optimizes each initial notable figure, obtains the optimal notable figure under each scale；
The initial notable figure of target image is generated using each initial super-pixel block under each scale.Ant colony optimization algorithm is one
The probabilistic type algorithm for finding path optimizing is planted, which has the spy of distribution calculating, information positive feedback and heuristic search
Sign, is substantially the heuristic global optimization approach of one of evolution algorithm.Ant colony is used according to the feature of initial super-pixel block
Optimization algorithm is updated initial super-pixel block, obtains multiple super-pixel regions.Super-pixel region is to be calculated using ant group optimization
The super-pixel that method obtains after being updated to initial super-pixel block.
S103 obtains the final notable figure of the target image according to the optimal notable figure under each scale.
The multidimensional characteristic of each super-pixel region corresponding region in the target image is corresponding more as each super-pixel region
Dimensional feature.According to the corresponding multidimensional characteristic in each super-pixel region, the aobvious of each super-pixel region is calculated using global contrast method
Work value.The notable figure of target image is obtained using the saliency value in each super-pixel region.
Target image is divided into multiple initial super-pixel block by using super-pixel segmentation algorithm by the present embodiment, according to each
The multidimensional characteristic of initial super-pixel block obtains the initial notable figure of target image, by each initial super-pixel block in initial notable figure
Regard a node as, construct graph model, initial notable figure is optimized using ant colony optimization algorithm on the basis of graph model,
The conspicuousness that target image is detected according to the initial notable figure after optimization improves the accurate of low light image conspicuousness detection
Property, support is provided for the problems such as night safe monitoring and positioning of complex environment target.
On the basis of the above embodiments, the step of extracting multiple features in the present embodiment from target image is specifically wrapped
It includes: extracting color characteristic, textural characteristics, direction character and the Gradient Features of target image；Wherein, the color of the target image
Feature includes color characteristic of the target image in LAB, HSV and YCbCr space.
Specifically, 12 low-level visual features, including color characteristic, textural characteristics, direction spy are extracted from target image
It seeks peace Gradient Features.Wherein, color characteristic is by being respectively converted into LAB, HSV and YCbCr color for the target image of rgb format
Then the image in space extracts 9 features from 9 channels of three color spaces.Textural characteristics use the two of target image
Tieing up entropy indicates, has very strong resistivity to geometric deformation and ambient noise.It is extracted using the Gabor filter of different directions
Direction character with rotational invariance and global property, to reduce the influence of low-light.By to horizontal gradient and hang down
Vertical ladder degree is averaged, and can calculate Gradient Features to describe the variation of the amplitude of local gray level.
On the basis of the above embodiments, in the present embodiment based on super-pixel segmentation algorithm to the target image respectively into
The segmentation of the multiple scales of row, the step of obtaining initial super-pixel block, specifically include: according to multiple preset super-pixel numbers, in institute
It states and determines initial seed point in target image；According to the gradient value of each pixel in each initial seed neighborhood of a point, to institute
It states initial seed point to be updated, obtains new seed point；By the pixel in the preset range centered on each new seed point
As each new seed neighborhood of a point pixel, according between each new seed point and each new seed neighborhood of a point pixel
Distance, distribute class label for each new seed neighborhood of a point pixel, will make with the pixel of the identical class label
For an initial super-pixel block.
Specifically, using SLIC, (Simple Linear Iterative Clustering, simple linear change the present embodiment
Generation cluster) algorithm to target image carry out super-pixel segmentation.Initialization seed point, i.e. cluster centre first.According to presetting
Super-pixel number K seed point is uniformly distributed in target image.Assuming that a total of N number of pixel of Target Photo, pre-segmentation
For the super-pixel of K identical sizes, then the size of each super-pixel is N/K, then the distance of neighboring seeds point is approximately S=
sqrt(N/K).In order to avoid seed point is fallen in the biggish profile and border of gradient, subsequent Clustering Effect is influenced, the present embodiment exists
Seed point is reselected in the n*n neighborhood of seed point, generally takes n=3.Specific method is to calculate in each seed neighborhood of a point to own
Seed point is moved on to the smallest place of seed neighborhood of a point inside gradient by the gradient value of pixel.Then in each seed point week
Class label is distributed for each pixel in the neighborhood enclosed, that is, which cluster centre belonged to.Scheme with the k-means of standard at whole
Middle search is different, and the search range of SLIC is limited to 2S*2S, can accelerate convergence rate.
On the basis of the above embodiments, it is obtained in the present embodiment according to the multidimensional characteristic of each initial super-pixel block each
The step of saliency value of the initial super-pixel block, specifically includes: calculating the one-dimensional entropy of each characteristic pattern of the target image；It will
The feature of the maximum predetermined number of the one-dimensional entropy of target image is as target signature；According to each initial super-pixel block
Target signature obtains the saliency value of each initial super-pixel block based on global contrast method.
Specifically, every one-dimensional characteristic of target image extracts a characteristic pattern, calculates each characteristic pattern of target image
One-dimensional entropy, calculation formula is as follows:
Wherein, entropy is the one-dimensional entropy of any feature figure, piExpression gray scale is the pixel of i in this feature figure
Ratio.The characteristic pattern of the maximum predetermined number of one-dimensional entropy is selected to carry out conspicuousness inspection from all characteristic patterns of target image
It surveys.For example, extracting 12 dimensional features of target image, one-dimensional entropy maximum 9 are selected from 12 characteristic patterns of target image
Characteristic pattern carries out conspicuousness detection, using the corresponding feature of select 9 characteristic patterns as target signature, i.e., detects as conspicuousness
Used feature, so that the optimum visual feature of various low-light images be adaptive selected.Due to characteristic pattern size and
The size of target image is identical, according to each initial super-pixel block in the target image where region, by each initial super-pixel block
9 kind target signatures of the corresponding region as each initial super-pixel block in 9 characteristic patterns.According to each initial super-pixel block
9 kinds of target signatures obtain the saliency value of each initial super-pixel block based on global contrast method.Similarly, according to each super picture
Plain region corresponding region in the target image, using each super-pixel region in 9 characteristic patterns corresponding region as each super picture
The corresponding 9 kinds of target signatures in plain region are based on global contrast side according to the corresponding 9 kinds of target signatures in each super-pixel region
Method obtains the saliency value in each super-pixel region.
On the basis of the above embodiments, the initial super-pixel block is carried out based on ant colony optimization algorithm in the present embodiment
The step of merging, obtaining multiple super-pixel regions specifically includes: calculating the distance between initial super-pixel block described in any two；
The distance between initial super-pixel block of any two is compared with preset threshold, any two is obtained and initially surpasses
The information content in path between block of pixels；According to the information content in path between the initial super-pixel block of any two, ant is calculated
Select the probability in the path；When the ant is mobile according to the probability, the information content on the path is updated, according to institute
Information computing ant after stating routing update selects the probability in the path, until reaching preset termination condition；If each time repeatedly
The probability in the selection of ant described in generation path is greater than the probability in the selection of ant described in first time iteration path, then by the ant
The corresponding initial super-pixel block in the path of selection merges, using the super-pixel obtained after merging as super-pixel region.
Specifically, regard each initial super-pixel block as a node building graph model, m ant is in randomly selected knot
It is extended on point, the process for the food source that the process for finding optimal super-pixel is found in search space as ant.
The initial super-pixel block s of any twoiAnd sjThe distance between define dijAre as follows:
Wherein, m indicates the dimension of ant or node, | | indicate Euclidean distance, wkWeighted factor is indicated, according to collection
The percentage contribution of each initial super-pixel block determines in cluster.Cluster radius is indicated with r, uses Iij(t) it indicates in t moment from siTo sjIt
Between information content on path, dijIt indicates from siTo sjDistance, then:
According to the information content on heuristic function and each path, ant selection is calculated from node siTo node sjPath probability.
In ant moving process, the residual information in each path will change, the information after the completion of each iteration, on path
Amount will be updated, and in next iteration, select each paths using information computing ant updated on each paths
Probability, until reaching preset termination condition, as the number of iterations reaches preset constant.When the ant is mobile according to probability, lead to
It crosses following formula and updates information content on the path:
Iij(t+1)=(1- ρ) Iij(t)+ρ·ΔIij；
Wherein, i and j indicates the number of the initial super-pixel block of any two, Iij(t) indicate t moment from initial super-pixel block
siTo initial super-pixel block sjInformation content, Iij(t+1) indicate the t+1 moment from initial super-pixel block siTo initial super-pixel block sj
Information content, ρ indicates the change rate that the information content disappears at any time,Indicate m ant at the end of iteration
Information delta on each path.After all m ants are mobile, information matrix can update are as follows:
Wherein,Indicate information attenuation parameter, Iij(0) indicate that the information matrix of circulation for the first time, n indicate iteration
Total degree.
If the probability of ant selection either path selects the paths greater than ant in first time iteration in each iteration
Probability, i.e.,Then by the corresponding initial super-pixel block s of the pathsjIt is merged into siNeighborhood, after merging
Each super-pixel is as a super-pixel region.Wherein,It selects for kth ant in the t times iteration from initial super-pixel
Block siTo initial super-pixel block sjBetween path probability, P (0) be in first time iteration ant selection from initial super-pixel block si
To initial super-pixel block sjBetween path probability.Obtained Optimal cluster centers are as follows:
Wherein, sjTo need the initial super-pixel block merged, niFor the initial super-pixel block for needing to merge
Number, OjAs food source, s can be used asiIdeal neighborhood, the optimal saliency value in each super-pixel region is then calculated, according to all
The saliency value in super-pixel region is come the notable figure that generates.
On the basis of the above embodiments, pass through following formula in the present embodiment according to the initial super-pixel of any two
The information content in path between block calculates the step of ant selects the probability in the path and specifically includes:
Wherein, k indicates ant number, and i and j indicate the number of the initial super-pixel block of any two, Iij(t) t moment is indicated
From initial super-pixel block siTo initial super-pixel block sjInformation content, ηij=1/dijIndicate heuristic function, dijFor siTo sjAway from
From Ω (si) indicate siNeighborhood, α indicate information heuristic greedy method, β indicate heuristic information influence
On the basis of the above embodiments, in the present embodiment according to the optimal notable figure under each scale, described in acquisition
The step of final notable figure of target image, specifically includes: the optimal notable figure under each scale is merged；To fusion
Notable figure afterwards is filtered, and obtains the final notable figure of the target image.
Specifically, repeated segmentation is carried out to target image using default super-pixel number different SLIC algorithm, such as Fig. 2 institute
Show, for example, being split respectively using the SLIC algorithm that default super-pixel number is 100,200 and 300 to target image, obtains
Three kinds of super-pixel segmentation results.12 kinds of features of target image, including target image three channels in the space LAB are extracted simultaneously
On value L, A and B, value H, S and the V of target image in HSV space on three channels, target image is three in YCbCr space
Value Y, Cb and Cr on a channel, the textural characteristics T of target image, the direction character O of target image and the ladder of target image
Spend feature G.According to the 12 of target image kinds of corresponding characteristic patterns of feature, the one-dimensional entropy of each characteristic pattern is calculated.Select one-dimensional entropy most
9 big characteristic patterns are as optimal characteristics.According to 9 kinds of features of initial super-pixel block each in every kind of super-pixel segmentation result, base
The corresponding three kinds of initial notable figures of three kinds of segmentation results are obtained in global contrast method, ant colony is used to every kind of initial notable figure
Optimization algorithm optimizes, and is then based on the optimum results that global contrast method obtains three kinds of initial notable figures.It is excellent by three kinds
Change result to be merged, fusion results are filtered, to obtain the final notable figure of target image.
A kind of saliency detection device is provided in another embodiment of the present invention, and the device is for realizing aforementioned
Method in each embodiment.Therefore, the description and definition in each embodiment of aforementioned image significance detection method, Ke Yiyong
The understanding of each execution module in the embodiment of the present invention.Fig. 3 is saliency detection device provided in an embodiment of the present invention
Overall structure diagram, the device include segmentation module 301, optimization module 302 and obtain module 303；Wherein:
Divide module 301, for extracting multiple features from target image, generates characteristic pattern, calculated based on super-pixel segmentation
Method carries out the segmentation of multiple scales to the target image respectively, obtains initial super-pixel block, according to each characteristic pattern and respectively
The initial super-pixel block obtains the saliency value of each initial super-pixel block；
Wherein, multi-dimension feature extraction is carried out to target, the present embodiment is not limited to extract the type of feature.Super-pixel segmentation is calculated
Method refers to the irregular picture for having certain visual meaningaaa that will have many characteristics, such as similar grain, color, the adjacent pixel composition of brightness
Plain block replaces a large amount of pixel to express figure using the similitude of feature between pixel by group pixels with a small amount of super-pixel
As feature, to reduce the complexity of post processing of image.Target image is the image for needing to carry out conspicuousness detection.Initial super picture
Plain block is the super-pixel that target image is directly split to using super-pixel segmentation algorithm acquisition.Each initial super-pixel block tool
There is a class label.Divide module 301 and target image be divided into multiple initial super-pixel block using super-pixel segmentation algorithm,
The present embodiment is suitable for various super-pixel segmentation algorithms.According to the region of initial super-pixel block in the target image, from target figure
The multidimensional characteristic that initial super-pixel block is obtained in the multidimensional characteristic of picture, according to the multidimensional characteristic of each initial super-pixel block, using complete
Office's contrast method obtains the saliency value of each initial super-pixel block.
Optimization module 302 is used to be generated according to the saliency value of each initial super-pixel block initial aobvious under each scale
Figure is write, each initial notable figure is optimized based on ant colony optimization algorithm, obtains the optimal notable figure under each scale；
The initial notable figure of target image is generated using the saliency value of each initial super-pixel block.Ant colony optimization algorithm is a kind of
For finding the probabilistic type algorithm of path optimizing, which has the feature of distribution calculating, information positive feedback and heuristic search,
It is substantially the heuristic global optimization approach of one of evolution algorithm.Optimization module 302 is according to the feature of initial super-pixel block
Initial super-pixel block is updated using ant colony optimization algorithm, obtains super-pixel region.Super-pixel region is excellent using ant colony
Change the super-pixel obtained after algorithm is updated initial super-pixel block.
Module 303 is obtained to be used to obtain the final aobvious of the target image according to the optimal notable figure under each scale
Write figure.
The multidimensional characteristic of each super-pixel region corresponding region in the target image is corresponding more as each super-pixel region
Dimensional feature.Module 303 is obtained according to the corresponding multidimensional characteristic in each super-pixel region, calculates each super picture using global contrast method
The saliency value in plain region.The final notable figure of target image is obtained using the saliency value in super-pixel region under each scale.
Target image is divided into multiple initial super-pixel by using super-pixel segmentation algorithm by the present embodiment the present embodiment
Block obtains the initial notable figure of target image according to the multidimensional characteristic of each initial super-pixel block, will be each first in initial notable figure
Beginning super-pixel block regards a node as, constructs graph model, using ant colony optimization algorithm to initial significant on the basis of graph model
Figure optimizes, and the conspicuousness of target image is detected according to the initial notable figure after optimization, it is significant to improve low light image
Property detection accuracy, for night safe monitoring and complex environment target position the problems such as support is provided.
On the basis of the above embodiments, module is divided in the present embodiment to be further used for: extracts the color of target image
Feature, textural characteristics, direction character and Gradient Features；Wherein, the color characteristic of the target image includes the target image
In the color characteristic of LAB, HSV and YCbCr space.
On the basis of the above embodiments, module is divided in the present embodiment to be specifically used for: according to multiple default super-pixel
Number determines initial seed point in the target image respectively；According to each pixel in each initial seed neighborhood of a point
Gradient value is updated the initial seed point, obtains new seed point；By the default model centered on each new seed point
Pixel in enclosing is as each new seed neighborhood of a point pixel, according to the neighbour of each new seed point and each new seed point
The distance between domain pixel distributes class label for each new seed neighborhood of a point pixel, will have the identical classification mark
The pixel of label is as an initial super-pixel block.
On the basis of the above embodiments, module is divided in the present embodiment to be further used for: calculates the target image
The one-dimensional entropy of each characteristic pattern；Using the characteristic pattern of the maximum predetermined number of the one-dimensional entropy of the target image as target signature；Root
According to the target signature of each initial super-pixel block, the aobvious of each initial super-pixel block is obtained based on global contrast method
Work value.
On the basis of the above embodiments, optimization module is specifically used in the present embodiment: calculating initial described in any two
The distance between super-pixel block；The distance between initial super-pixel block of any two is compared with preset threshold, is obtained
Take the information content in path between the initial super-pixel block of any two；According to road between the initial super-pixel block of any two
The information content of diameter calculates the probability that ant selects the path；When the ant is mobile according to the probability, the road is updated
Information content on diameter selects the probability in the path according to the information computing ant after the routing update, pre- until reaching
If termination condition；If ant described in each secondary iteration selects the probability in path to be greater than ant described in first time iteration and selects path
Probability, then the corresponding initial super-pixel block in path that the ant selects is updated, the super-pixel that will be obtained after update
As super-pixel region.
On the basis of the above embodiments, in the present embodiment optimization module especially by following formula according to described any two
The information content in path between a initial super-pixel block calculates the step of ant selects the probability in the path and specifically includes:
Wherein, k indicates ant number, and i and j indicate the number of the initial super-pixel block of any two, Iij(t) t moment is indicated
From initial super-pixel block siTo initial super-pixel block sjInformation content, ηij=1/dijIndicate heuristic function, dijFor siTo sjAway from
From Ω (si) indicate siNeighborhood, α indicate information heuristic greedy method, β indicate heuristic information influence,
On the basis of the various embodiments described above, in the present embodiment obtain module be specifically used for: by under each scale most
Excellent notable figure is merged；Fused notable figure is filtered, the final notable figure of the target image is obtained.
The present embodiment provides a kind of electronic equipment, Fig. 4 is electronic equipment overall structure provided in an embodiment of the present invention signal
Figure, which includes: at least one processor 401, at least one processor 402 and bus 403；Wherein,
Processor 401 and memory 402 pass through bus 403 and complete mutual communication；
Memory 402 is stored with the program instruction that can be executed by processor 401, and the instruction of processor caller is able to carry out
Method provided by above-mentioned each method embodiment, for example, extract multiple features from target image, generate characteristic pattern, base
The segmentation for carrying out multiple scales respectively to the target image in super-pixel segmentation algorithm, obtains initial super-pixel block, according to each
The characteristic pattern and each initial super-pixel block obtain the saliency value of each initial super-pixel block；According to each described initial
The saliency value of super-pixel block generates the initial notable figure under each scale, based on ant colony optimization algorithm to each described initial significant
Figure optimizes, and obtains the optimal notable figure under each scale；According to the optimal notable figure under each scale, described in acquisition
The final notable figure of target image.
The present embodiment provides a kind of non-transient computer readable storage medium, non-transient computer readable storage medium storages
Computer instruction, computer instruction make computer execute method provided by above-mentioned each method embodiment, for example, from target
Multiple features are extracted in image, generate characteristic pattern, multiple rulers are carried out to the target image based on super-pixel segmentation algorithm respectively
The segmentation of degree obtains initial super-pixel block, according to each characteristic pattern and each initial super-pixel block, obtains each described initial
The saliency value of super-pixel block；The initial notable figure under each scale is generated according to the saliency value of each initial super-pixel block,
Each initial notable figure is optimized based on ant colony optimization algorithm, obtains the optimal notable figure under each scale；According to
Optimal notable figure under each scale obtains the final notable figure of the target image.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer readable storage medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: ROM, RAM, magnetic disk or light
The various media that can store program code such as disk.
The apparatus embodiments described above are merely exemplary, wherein described, unit can as illustrated by the separation member
It is physically separated with being or may not be, component shown as a unit may or may not be physics list
Member, it can it is in one place, or may be distributed over multiple network units.It can be selected according to the actual needs
In some or all of the modules achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying creativeness
Labour in the case where, it can understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
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
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of image significance detection method characterized by comprising
Multiple features are extracted from target image, generate characteristic pattern, and the target image is distinguished based on super-pixel segmentation algorithm
The segmentation for carrying out multiple scales obtains initial super-pixel block, according to each characteristic pattern and each initial super-pixel block, obtains
The saliency value of each initial super-pixel block；
The initial notable figure under each scale is generated according to the saliency value of each initial super-pixel block, is calculated based on ant group optimization
Method optimizes each initial notable figure, obtains the optimal notable figure under each scale；
According to the optimal notable figure under each scale, the final notable figure of the target image is obtained.
2. the method according to claim 1, wherein the step of extracting multiple features from target image is specifically wrapped
It includes:
Extract color characteristic, textural characteristics, direction character and the Gradient Features of target image；Wherein, the face of the target image
Color characteristic includes color characteristic of the target image in LAB, HSV and YCbCr space.
3. the method according to claim 1, wherein being distinguished based on super-pixel segmentation algorithm the target image
The segmentation for carrying out multiple scales, the step of obtaining initial super-pixel block, specifically include:
According to multiple default super-pixel numbers, initial seed point is determined in the target image respectively；
According to the gradient value of each pixel in each initial seed neighborhood of a point, the initial seed point is updated, is obtained
Take new seed point；
By the pixel in the preset range using centered on each new seed point as each new seed neighborhood of a point pixel, root
It is each new seed neighborhood of a point picture according to the distance between each new seed point and each new seed neighborhood of a point pixel
Element distribution class label, using the pixel with the identical class label as an initial super-pixel block.
4. the method according to claim 1, wherein according to each characteristic pattern and each initial super-pixel block
The step of obtaining the saliency value of each initial super-pixel block specifically includes:
Calculate the one-dimensional entropy of each characteristic pattern of the target image；
Using the characteristic pattern of the maximum predetermined number of the one-dimensional entropy of the target image as target signature；
According to the target signature of each initial super-pixel block, each initial super-pixel is obtained based on global contrast method
The saliency value of block.
5. being obtained each described the method according to claim 1, wherein each initial notable figure optimizes
The step of optimal notable figure under scale, specifically includes:
Calculate the distance between initial super-pixel block described in any two；
The distance between initial super-pixel block of any two is compared with preset threshold, at the beginning of obtaining any two
The information content in path between beginning super-pixel block；
According to the information content in path between the initial super-pixel block of any two, the probability that ant selects the path is calculated；
When the ant is mobile according to the probability, the information content on the path is updated, after the routing update
Information computing ant selects the probability in the path, until reaching preset termination condition；
If ant described in each secondary iteration selects the probability in path to be greater than the probability that ant described in first time iteration selects path,
Then the corresponding initial super-pixel block in path that the ant selects is updated.
6. according to the method described in claim 5, it is characterized in that, by following formula according to the initial super picture of any two
The information content in path between plain block calculates the step of ant selects the probability in the path and specifically includes:
Wherein, k indicates ant number, and i and j indicate the number of the initial super-pixel block of any two, Iij(t) indicate t moment from first
Beginning super-pixel block siTo initial super-pixel block sjInformation content, ηij=1/dijIndicate heuristic function, dijFor siTo sjDistance, Ω
(si) indicate siNeighborhood, α indicate information heuristic greedy method, β indicate heuristic information influence.
7. -6 any method according to claim 1, which is characterized in that according to the optimal notable figure under each scale,
The step of obtaining the final notable figure of the target image specifically includes:
Optimal notable figure under each scale is merged；
Fused notable figure is filtered, the final notable figure of the target image is obtained.
8. a kind of saliency detection device characterized by comprising
Divide module, for extracting multiple features from target image, characteristic pattern is generated, based on super-pixel segmentation algorithm to described
Target image carries out the segmentation of multiple scales respectively, obtains initial super-pixel block, according to each characteristic pattern and each described initial
Super-pixel block obtains the saliency value of each initial super-pixel block；
Optimization module, for generating the initial notable figure under each scale according to the saliency value of each initial super-pixel block,
Each initial notable figure is optimized based on ant colony optimization algorithm, obtains the optimal notable figure under each scale；
Module is obtained, for obtaining the final notable figure of the target image according to the optimal notable figure under each scale.
9. a kind of electronic equipment characterized by comprising
At least one processor, at least one processor and bus；Wherein,
The processor and memory complete mutual communication by the bus；
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to instruct energy
Enough methods executed as described in claim 1 to 7 is any.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method as described in claim 1 to 7 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811355398.5A CN109685806B (en) | 2018-11-14 | 2018-11-14 | Image significance detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811355398.5A CN109685806B (en) | 2018-11-14 | 2018-11-14 | Image significance detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109685806A true CN109685806A (en) | 2019-04-26 |
| CN109685806B CN109685806B (en) | 2020-06-30 |
Family
ID=66184682
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811355398.5A Expired - Fee Related CN109685806B (en) | 2018-11-14 | 2018-11-14 | Image significance detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685806B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070539A (en) * | 2019-04-28 | 2019-07-30 | 重庆大学 | Image quality evaluating method based on comentropy |
| CN111242941A (en) * | 2020-01-20 | 2020-06-05 | 南方科技大学 | Salient region detection method and device based on visual attention |
| CN111583279A (en) * | 2020-05-12 | 2020-08-25 | 重庆理工大学 | Super-pixel image segmentation method based on PCBA |
| CN114332633A (en) * | 2022-03-01 | 2022-04-12 | 北京化工大学 | Radar image target detection and identification method, equipment and storage medium |
| CN114581207A (en) * | 2022-05-09 | 2022-06-03 | 北京嘉沐安科技有限公司 | Commodity image big data accurate pushing method and system for E-commerce platform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930542A (en) * | 2012-10-31 | 2013-02-13 | 电子科技大学 | Detection method for vector saliency based on global contrast |
| CN103400129A (en) * | 2013-07-22 | 2013-11-20 | 中国科学院光电技术研究所 | Target tracking method based on frequency domain saliency |
| CN107527348A (en) * | 2017-07-11 | 2017-12-29 | 湖州师范学院 | Conspicuousness detection method based on multi-scale division |
| CN108345892A (en) * | 2018-01-03 | 2018-07-31 | 深圳大学 | A kind of detection method, device, equipment and the storage medium of stereo-picture conspicuousness |
-
2018
- 2018-11-14 CN CN201811355398.5A patent/CN109685806B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930542A (en) * | 2012-10-31 | 2013-02-13 | 电子科技大学 | Detection method for vector saliency based on global contrast |
| CN103400129A (en) * | 2013-07-22 | 2013-11-20 | 中国科学院光电技术研究所 | Target tracking method based on frequency domain saliency |
| CN107527348A (en) * | 2017-07-11 | 2017-12-29 | 湖州师范学院 | Conspicuousness detection method based on multi-scale division |
| CN108345892A (en) * | 2018-01-03 | 2018-07-31 | 深圳大学 | A kind of detection method, device, equipment and the storage medium of stereo-picture conspicuousness |
Non-Patent Citations (1)
| Title |
|---|
| NAN MU ET AL.: "Particle Swarm Optimization Based Salient Object Detection for Low Contrast Images", 《SPRINGER INTERNATIONAL PUBLISHING AG 2017》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070539A (en) * | 2019-04-28 | 2019-07-30 | 重庆大学 | Image quality evaluating method based on comentropy |
| CN111242941A (en) * | 2020-01-20 | 2020-06-05 | 南方科技大学 | Salient region detection method and device based on visual attention |
| CN111242941B (en) * | 2020-01-20 | 2023-05-30 | 南方科技大学 | Salient region detection method and device based on visual attention |
| CN111583279A (en) * | 2020-05-12 | 2020-08-25 | 重庆理工大学 | Super-pixel image segmentation method based on PCBA |
| CN114332633A (en) * | 2022-03-01 | 2022-04-12 | 北京化工大学 | Radar image target detection and identification method, equipment and storage medium |
| CN114332633B (en) * | 2022-03-01 | 2022-06-10 | 北京化工大学 | Radar image target detection and identification method and equipment and storage medium |
| CN114581207A (en) * | 2022-05-09 | 2022-06-03 | 北京嘉沐安科技有限公司 | Commodity image big data accurate pushing method and system for E-commerce platform |
| CN114581207B (en) * | 2022-05-09 | 2022-09-13 | 农夫铺子发展集团有限公司 | Commodity image big data accurate pushing method and system for E-commerce platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109685806B (en) | 2020-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685806A (en) | Image significance detection method and device | |
| Li et al. | Superpixel segmentation using linear spectral clustering | |
| Marchesi et al. | A context-sensitive technique robust to registration noise for change detection in VHR multispectral images | |
| Larsen et al. | Comparison of six individual tree crown detection algorithms evaluated under varying forest conditions | |
| CN109522908A (en) | Image significance detection method based on area label fusion | |
| US9147132B2 (en) | Classification of land based on analysis of remotely-sensed earth images | |
| CN107909015A (en) | Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion | |
| Aslan et al. | Benchmarking algorithms for food localization and semantic segmentation | |
| CN109558806A (en) | The detection method and system of high score Remote Sensing Imagery Change | |
| CN104834933A (en) | Method and device for detecting salient region of image | |
| CN109740721B (en) | Wheat ear counting method and device | |
| CN109583345A (en) | Roads recognition method, device, computer installation and computer readable storage medium | |
| CN109741293A (en) | Conspicuousness detection method and device | |
| CN102930279B (en) | For the image-recognizing method that product quantity detects | |
| CN110059677A (en) | Digital table recognition methods and equipment based on deep learning | |
| Zhang et al. | Detection of regions of interest in a high-spatial-resolution remote sensing image based on an adaptive spatial subsampling visual attention model | |
| CN110022674A (en) | For detecting method, medium and the system of Potyvirus in crop image | |
| CN107220647A (en) | Crop location of the core method and system under a kind of blade crossing condition | |
| Zhao et al. | Automatic blur region segmentation approach using image matting | |
| CN109919149A (en) | Object mask method and relevant device based on object detection model | |
| CN110276363A (en) | A kind of birds small target detecting method based on density map estimation | |
| CN105069459B (en) | One kind is directed to High Resolution SAR Images type of ground objects extracting method | |
| CN107274425B (en) | A kind of color image segmentation method and device based on Pulse Coupled Neural Network | |
| CN116645592A (en) | Crack detection method based on image processing and storage medium | |
| Wang et al. | An efficient attention module for instance segmentation network in pest monitoring |
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: 20200630 Termination date: 20201114 |