CN109543701A - Vision significance method for detecting area and device - Google Patents
Vision significance method for detecting area and device Download PDFInfo
- Publication number
- CN109543701A CN109543701A CN201811455470.1A CN201811455470A CN109543701A CN 109543701 A CN109543701 A CN 109543701A CN 201811455470 A CN201811455470 A CN 201811455470A CN 109543701 A CN109543701 A CN 109543701A
- Authority
- CN
- China
- Prior art keywords
- image
- value
- pixel
- processed
- saliency map
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of vision significance method for detecting area and devices.Wherein, it is that each pixel chooses multiple candidate image area blocks from image to be processed that method, which includes first with stochastic search methods, is then chosen and the highest object region block of pixel similarity from each candidate image area block;According to the similarity between each target image block and pixel, the vision significance value of each pixel is calculated;The vision significance value of each pixel based on image to be processed, generates the visual saliency map of each level of image to be processed;The weighted value that each layer visual saliency map is finally calculated according to the vision significance value of adjacent two layers visual saliency map merges each layer visual saliency map according to weighted value.The application improves the computational efficiency of pixel vision significance value, so that the detection efficiency in the vision significance region of image to be processed be effectively promoted, can also promote the accuracy in detection and precision in the vision significance region in image to be processed.
Description
Technical field
The present embodiments relate to technical field of machine vision, more particularly to a kind of vision significance method for detecting area
And device.
Background technique
With the fast development of machine vision technique, the vision significance region detection based on image procossing is as the technology
One ring of key during realization, obtains urgent development.
Vision noticing mechanism is the mankind and the important organoleptic feature that high iq animal possesses, this mechanism can be from magnanimity vision
Quickly found in information with vision significance and interested target object, and can rapidly " ignoring " other lose interest in
Target object, while reducing the calculation amount in information process.Therefore vision noticing mechanism is introduced into image segmentation process
In can significantly improve the working efficiency of image processing process.In image segmentation and image analysis treatment process, it will usually needle
Interest is kept to certain specific regions on image to be processed, these region parts are referred to as area-of-interest or target area.
Under normal circumstances, target is exactly the area-of-interest in image with special nature.It is just needed from image to extract target
Cutting operation is executed, is worked through can be just further analyzed after over-segmentation with relevant treatment.Image segmentation process is exactly will
Image is distinguished into several isolated areas with specific uniformity consistency, to reach area-of-interest from image complex background
In the image processing method that extracts.
How from image to be processed to detect area-of-interest, relate to the object detection method of image and video,
This method is by changing image sequence or video resolution, so that the perfect display of image sequence and video energy, and to the greatest extent
Amount saves the key content in image and video.And in the image of these image content-baseds and video object detection process,
How to be used for quickly detecting vision significance region is problem to be solved.
The fixed figure that human vision is noticed at first is found in vision significance Region detection algorithms concern in the prior art
As pixel or target object, can be applied for human vision concerned pixel point and related application, such as automatic focusing is understood.
The context of pixel present position is expressed according to the region unit, proposes that the vision significance of image slices vegetarian refreshments is by the picture
It is expressed the relevant range at vegetarian refreshments center.For example, if the associated picture region unit and original image of central pixel point
In other image-region blocks difference it is very big, then pixel can then regard the high pixel of vision significance as.
The method for detecting area of vision significance in the related technology is all using the system for calculating independent pixel every time
The conventional method of vision significance value.Pixel quantity scale will lead to computation complexity increase, or even also to construct higher-dimension to
The lookup tree construction of amount could execute, and the time complexity and space complexity of this method are all very high, be suitable for original image
Middle small volume target.
Summary of the invention
The embodiment of the present disclosure provides a kind of vision significance method for detecting area and device, can quickly, accurately detect
Vision significance region in image to be processed out.
In order to solve the above technical problems, the embodiment of the present invention the following technical schemes are provided:
On the one hand the embodiment of the present invention provides a kind of vision significance method for detecting area, comprising:
Multiple candidate image area blocks are chosen for current pixel point from image to be processed using stochastic search methods, from each
The object region block for meeting similarity condition is chosen in candidate image area block;
According to the similarity between each target image block and the current pixel point, the vision of the current pixel point is calculated
Significance value;
The vision significance value of each pixel based on the image to be processed generates each level of image to be processed
Visual saliency map;
According to the vision significance value of adjacent two layers visual saliency map, calculate the weighted value of each layer visual saliency map, with
Each layer visual saliency map is merged in the weighted value according to each layer visual saliency map.
Optionally, the vision significance value according to adjacent two layers visual saliency map, calculates each layer visual saliency map
After weighted value, further includes:
Multiple predeterminated position points are selected in the visual saliency map obtained after fusion;
To each predeterminated position point, according to the vision significance value of each pixel, be classified as strengthening pixel point set and
Pixel point set is weakened, the vision significance value for strengthening the pixel that pixel is concentrated is greater than what the reduction pixel was concentrated
The vision significance value of pixel;
Intensive treatment is carried out using each pixel that image enchancing method concentrates the reinforcing pixel;
Weakening treatment is carried out using each pixel that image method for weakening concentrates the reduction pixel.
Optionally, the vision significance value of each pixel based on the image to be processed generates described to be processed
The visual saliency map of each level of image includes:
The vision significance value of each pixel based on the image to be processed generates each level of image to be processed
Roughization visual saliency map;
Detailed-oriented operation is carried out to each roughization visual saliency map to obtain respective to remove picture noise signal
Detailed-oriented visual saliency map.
Optionally, described that the object region block packet for meeting similarity condition is chosen from each candidate image area block
It includes:
Calculate separately the similarity magnitude of the current pixel point and each candidate image area block;
It deletes similarity magnitude and is lower than the corresponding candidate image area block of similarity threshold, by remaining candidate image area
Block is as object region block.
Optionally, the similarity according between each target image block and the current pixel point calculates described current
The vision significance value of pixel includes:
The vision significance value S of the current pixel point is calculated using following formula:
Wherein,
In formula, dist (ri, rk) dissmilarity between the current pixel point position and k-th target image block
Property metric, K be target image block total number, distcolor(ri, rk) it is the object region block handled by vectorization
With Euclidean distance of the current pixel point position on hsv color space, distpos(ri, rk) it is the current pixel
Euclidean distance between point position and k-th target image block.
Optionally, the vision significance value of each pixel based on the image to be processed generates described to be processed
The visual saliency map of each level of image includes:
It is aobvious to generate local system for the local system significance value of each pixel based on image to be processed described in frequency-domain calculations
Write figure;
The global system significance value V of each pixel of the image to be processed is calculated using following formulaGlobal(x, y),
Generate global system notable figure:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and f (x, y) is the significance function for solving pixel (x, y),
faverage(x, y) is the arithmetic mean of instantaneous value of f (x, y), and the image size to be processed is M*N;
The rare significance value V of each pixel of the image to be processed is calculated using following formulaScarcity(x, y), it is raw
At rare notable figure:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, faverage(x, y) is the arithmetic mean of instantaneous value of f (x, y), h (faverage
(x, y)) it is the feature histogram that the image to be processed generates;
The local system notable figure, the global system notable figure and the rare notable figure are merged, obtained
The visual saliency map of each level of image to be processed.
Optionally, it is described by the local system notable figure, the global system notable figure and the rare notable figure into
Row fusion, the visual saliency map for obtaining each level of image to be processed include:
To each layer of the image to be processed, the vision of each pixel of the image to be processed is calculated using following formula
Significance value Vfinal, generate visual saliency map:
In formula, VLocal、VGlobal、VScarcityIt is followed successively by the local system conspicuousness for the pixel that transverse and longitudinal coordinate value is x, y
Value, global system significance value and rare significance value;v1For the weighted value of local system significance value, v2It is aobvious for global system
The weighted value of work property value, v3For the weighted value of rare significance value;I=1, Vi=V1For the local system significance value, i=
2, Vi=V2For whole system significance value, i=3, Vi=V3For the rare significance value, V1、V2、V3According toCalculate gained.
Optionally, the local system significance value packet of each pixel based on image to be processed described in frequency-domain calculations
It includes:
The local system significance value V of each pixel of the image to be processed is calculated using following formulaLocal(x, y):
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and FFT (u, v) is pixel characteristic value, | FFT (u, v) ejψ(u,v)|
For the gained picture amplitude spectrum after Fast Fourier Transform (FFT), ψ (u, v) is the image phase spectrum to be processed.
Optionally, the vision significance value according to adjacent two layers visual saliency map, calculates each layer visual saliency map
Weighted value, with for the weighted value according to each layer visual saliency map by each layer visual saliency map carry out fusion include:
The weighted value of each layer visual saliency map is calculated using following formula:
In formula, p is pixel position,For i-th layer of weighted value,For (i-1)-th layer of weighted value,It is i-th
The visualization significance value of layer,For (i-1)-th layer of vision significance value;
The visual saliency map of adjacent two layers is merged using following formula, obtains fusion visual saliency map;
In formula,For the pixel vision significance value of the fusion position visual saliency map p.
On the other hand the embodiment of the present invention provides a kind of vision significance regional detection device, comprising:
Random search module, for choosing multiple times from image to be processed using stochastic search methods for current pixel point
Image-region block is selected, the object region block for meeting similarity condition is chosen from each candidate image area block;
Vision significance value computing module, for according to similar between each target image block and the current pixel point
Degree, calculates the vision significance value of the current pixel point;
Multilayer visual saliency map generation module, the vision significance for each pixel based on the image to be processed
Value generates the visual saliency map of each level of image to be processed;
Visual saliency map Fusion Module calculates each layer for the vision significance value according to adjacent two layers visual saliency map
The weighted value of visual saliency map, to merge each layer visual saliency map for the weighted value according to each layer visual saliency map.
The embodiment of the invention also provides a kind of vision significance equipment for area detection equipment, including processor, the processors
Realizing the vision significance method for detecting area as described in preceding any one when for executing the computer program stored in memory
Step.
The embodiment of the present invention finally additionally provides a kind of computer readable storage medium, the computer readable storage medium
On be stored with vision significance region detector, when the vision significance region detector is executed by processor realize such as
The step of any one of preceding vision significance method for detecting area.
The advantages of technical solution provided by the present application, is, in the vision significance for calculating each pixel of image to be processed
When value, multiple candidate image area blocks are selected from image to be processed first with stochastic search methods, then from multiple candidate regions
Selection and the highest image-region block of current pixel point similarity, do not need each pixel from image to be processed in image block
It is found, shortens the lookup time of similarity highest image-region block, the scale of unrestricted image slices vegetarian refreshments, to solve
It is needed in the related technology from finding disadvantage existing for similarity highest image-region block in image to be processed in all pixels point
End, greatly improves pixel vision significance value computational efficiency, so that the vision in image to be processed be effectively promoted
The detection efficiency of salient region, the pixel vision significance value suitable for image to be processed under any scale calculate scene
In;In addition, not only contributing to eliminate the noise signal in visual saliency map by merging multi-level visual saliency map, can also incite somebody to action
Notable feature in image to be processed, which is accurately fused to, to be generated in vision notable feature figure, is conducive to be promoted in image to be processed
The accuracy in detection and precision in vision significance region.
It should be understood that the above general description and the following detailed description are merely exemplary, this can not be limited
It is open.
Detailed description of the invention
It, below will be to embodiment or correlation for the clearer technical solution for illustrating the embodiment of the present invention or the relevant technologies
Attached drawing needed in technical description is briefly described, it should be apparent that, the accompanying drawings in the following description is only 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 a kind of flow diagram of vision significance method for detecting area provided in an embodiment of the present invention;
Fig. 2 is the flow diagram of another vision significance method for detecting area provided in an embodiment of the present invention;
Fig. 3 is the ROC curve of the disclosure technical scheme shown according to an exemplary embodiment and the relevant technologies
Contrast schematic diagram;
Fig. 4 is that the ROC of technical scheme and the relevant technologies that the disclosure is shown according to another exemplary embodiment is bent
Line contrast schematic diagram;
Fig. 5 is that the objective contour in the illustrative image extracted using technical scheme that the disclosure provides is illustrated
Figure;
Fig. 6 is a kind of specific embodiment structure of vision significance regional detection device provided in an embodiment of the present invention
Figure;
Fig. 7 is another specific embodiment structure of vision significance regional detection device provided in an embodiment of the present invention
Figure.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.Obviously, described embodiments are only a part of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
The description and claims of this application and term " first ", " second ", " third " " in above-mentioned attached drawing
Four " etc. be for distinguishing different objects, rather than for describing specific sequence.Furthermore term " includes " and " having " and
Their any deformations, it is intended that cover and non-exclusive include.Such as contain a series of steps or units process, method,
System, product or equipment are not limited to listed step or unit, but may include the step of not listing or unit.
After describing the technical solution of the embodiment of the present invention, the various non-limiting realities of detailed description below the application
Apply mode.
Referring first to Fig. 1, Fig. 1 is that a kind of process of vision significance method for detecting area provided in an embodiment of the present invention is shown
It is intended to, the embodiment of the present invention may include the following contents:
S101: multiple candidate image areas are chosen for current pixel point from image to be processed using stochastic search methods
Block chooses the object region block for meeting similarity condition from each candidate image area block.
To each pixel in image to be processed, carry out choosing multiple candidate image areas using stochastic search methods
Block (such as 2K candidate image area block), current pixel point herein refer to current time using stochastic search methods to
Handle the pixel that candidate image block is chosen in image.
Similarity condition is to select from candidate image area block and those of the similarity highest of pixel image-region
Block, similarity condition can be according to practical application scene (such as number, candidate image area block and the pictures of candidate image area block
The similarity value of vegetarian refreshments) determine a similarity threshold, the image-region block that will be less than threshold value is deleted.
It is 2K with candidate image area block, such as K=32, for object region block is K, illustrates the reality of S101
Existing process.
The two dimensional coordinate map function of image to be processed is f (x): R → V, i.e., all pixels point coordinate in image to be processed
Upper definition two dimensional coordinate map function f (x), V represent the vision significance value for executing and obtaining after normalization process.It is located at wait locate
Manage the r pixel point on image R, the corresponding vision significance value on [0,1] of image R to be processed is v, mapping function be f (r)=
v.Function f value is the vision significance value being normalized in [0,1] range, and vision significance value is storable in scale identical as R
In two-dimensional array.
If v=f (r) is by significant to carry out vision with multiple associated picture region units being closely located to pixel r
Property value calculate, specific method is as shown in formula 1.
ri=r+ δ βiRi; (1)
In equation 1, RiIt is the stochastic variable being evenly distributed, its range value is limited in [- 1,1] × [- 1,1];δ
It is the 1/2 of picture size to be processed;β is window attenuation parameter, for i=1 to be increased to i=2K image-region block, until
Search radius ω RiNarrow down to single pixel point.If i < 2K, makes i=1, until test candidate region number of blocks meets 2K
Until, β=0.75 in specific implementation part.
It is carrying out calculating the dissimilarity measurement between candidate area blocks and pixel r in 2K candidate area blocks according to formula 1
Value mi.It only can retain the lesser image-region block of 1/2 dissimilarity value in 2K candidate image block in the application, give up residue 1/2
Image-region block.2K candidate image area block, these approximate image regions are selected according to the approximate any position pixel r
Block is the partial region of possibility candidate region all in image R to be processed, since this belongs to incompleteization sample collection procedure,
Therefore fragmentary sample sampling can introduce a degree of sample error to subsequent obtained visual saliency map.But with adopting
Sample quantity 2K becomes larger, and sample error is certain to be substantially reduced.
Single width or height can only be detected in the related technology as the rough vision significance image of 256 pixels by comparing
Administrative division map.Because each pixel requires to carry out vision significance calculating, operand is very big, and in order to most fast
Speed finds the most similar image-region block, needs to construct the lookup decision tree based on high dimension vector on sparse grid.Cause
This, if calculating the vision significance value of each pixel one by one, whole process can be very slow.The application is not in whole image institute
There is in pixel the vision significance value for finding the highest image-region block of similarity to calculate the pixel.Using searching at random
Suo Fangfa randomly selects 2K image-region block from image all pixels point, and guarantees wherein K image-region block and ri
The highest image-region block of similarity abandons other K image-region block.The application only needs to extract 2K image district at random
Domain block, and the high dimension vector that do not establish auxiliary searches decision tree structure to improve search efficiency.
S102: according to the similarity between each target image block and current pixel point, the vision for calculating current pixel point is aobvious
Work property value.
distcolor(ri, rj) it is the image-region block r handled by vectorizationiWith rjEuclidean on hsv color space
Distance, and execute normalization operation and correspond in [0,1] range;Work as distcolor(ri, rj) relative to arbitrary image region unit
rjWhen all very big, then assert that pixel i has vision significance.
distpos(ri, rj) it is image-region block riWith rjEuclidean distance between position executes this Euclidean distance
Process, normalization correspond in [0,1] range.The measure of dissimilarity between two corresponding image-region blocks can be measured
As shown in following formula:
In consideration of it, calculating the vision significance value of current pixel point using formula 2 in a kind of specific embodiment
S:
Wherein,
In formula, dist (ri, rk) dissimilarity degree between current pixel point position and k-th target image block
Magnitude, K are the total number of target image block, distcolor(ri, rk) it is the object region block that is handled by vectorization and work as
Euclidean distance of the preceding pixel point position on hsv color space, distpos(ri, rk) be current pixel point position with
Euclidean distance between k-th target image block.
S103: the vision significance value of each pixel based on image to be processed generates the view of each level of image to be processed
Feel notable figure.
The processing of vision significance measurement by calculating the vision significance value of each image slices vegetarian refreshments, with be originally inputted figure
Visual saliency map is indicated as big small-scale identical gray level image.Wherein, the vision significance value of each pixel represents original
The vision significance value of relevant position in beginning image.The vision significance value the big, indicates that the image slices vegetarian refreshments is being originally inputted figure
Its vision significance more can be protruded as upper, while being more easy to get the attention of human viewer.
To each level of image to be processed, it is significant corresponding vision can be generated for each level in the same manner
Figure.
S104: according to the vision significance value of adjacent two layers visual saliency map, calculating the weighted value of each layer visual saliency map,
Each layer visual saliency map to be merged for the weighted value according to each layer visual saliency map.
It, also can not be thorough even with certain noise reduction means in view of random search algorithm can generate much noise
Noise information red in each vision significance figure is removed, it is significant in order to which multiple scale vision significant characteristics are further incorporated vision
In the final result of figure, obtained multi-level vision significance figure can be merged, it is multi-level after final generation merging
Vision significance result figure, such as following formula can be used and merged:
Above pass through combined multiple scale vision notable figure for i-th layer,For i-th layer of more rulers
Visual saliency map is spent,For the detailed-oriented vision significance result figure with i-th layer immediate (i-1)-th layer.IfWithSize have any different, then needing handle when being merged at many levels
Size toSize dimension is close.
In technical solution provided in an embodiment of the present invention, in the vision significance for calculating each pixel of image to be processed
When value, multiple candidate image area blocks are selected from image to be processed first with stochastic search methods, then from multiple candidate regions
Selection and the highest image-region block of current pixel point similarity, do not need each pixel from image to be processed in image block
It is found, shortens the lookup time of similarity highest image-region block, the scale of unrestricted image slices vegetarian refreshments, to solve
It is needed in the related technology from finding disadvantage existing for similarity highest image-region block in image to be processed in all pixels point
End, greatly improves pixel vision significance value computational efficiency, so that the vision in image to be processed be effectively promoted
The detection efficiency of salient region, the pixel vision significance value suitable for image to be processed under any scale calculate scene
In;In addition, not only contributing to eliminate the noise signal in visual saliency map by merging multi-level visual saliency map, can also incite somebody to action
Notable feature in image to be processed, which is accurately fused to, to be generated in vision notable feature figure, is conducive to be promoted in image to be processed
The accuracy in detection and precision in vision significance region.
It is of less demanding for the detailed-oriented degree of vision significance result figure, but if requiring to reach compared with high detection speed
Application scenarios can quickly generate roughization visual saliency map using the S101-S104 of above-described embodiment.When the detailed-oriented degree of needs
The application scenarios of identical visual saliency map are exactly matched with original input picture, it can also be aobvious to the vision that above-described embodiment obtains
Work property figure carries out enhancing and merges to generate the vision significance result figure of high quality.If occurring higher vision at some position
Significance value indicates that higher image reliability is provided in the position, so can be by selectively for those noise signals
High vision significance region merges, to obtain final vision significance result figure.
In a kind of specific embodiment, the process of vision significance figure is generated in S103 according to vision significance value
Implementation can be as described below:
The region significance numerical value of image to be processed relies on the difference of unique characteristics and ambient enviroment.If some in image
Region is marking area, then the marking area is with peripheral region, there are one or more distinguishing characteristics.In different images, phase
With feature, for influence that vision significance generates, there is also differences.Brightness, color can be used as vision significance feature, because
This needs to extract different images visual signature during image preprocessing.Such as it can be used and extract the view such as brightness, direction, color
Feel that feature is measured.The application has found that role is unobvious in the picture for directional vision feature, simultaneously by many experiments
Algorithms T-cbmplexity is also added, in consideration of it, the application only extracts the color and brightness of image to be processed.
Three hsv color space application tone H (Hue), saturation degree S (Saturation), lightness V (Value) parameters are come
Specific color is described, HSV ratio RGB is more in line with human visual feature.Image to be processed is reflected from rgb space using formula 3
It penetrates and is converted into HSV space, to extract color and brightness.
In formula, h is H channel value in hsv color space, and l is the arithmetic mean of instantaneous value of H channel value, and s is S in hsv color space
Channel value, v are V channel value in hsv color space, and g is RGB three primary colors R channel value, and b is RGB three primary colors channel B value, and r is
RGB three primary colors R channel value, max are the maximum value of respective corresponding channel value, and min is the minimum value of respective corresponding channel value.
In order to promote the precision of vision significance region detection, when extracting the significant characteristics of image to be processed, can incite somebody to action
System notable figure is constituted in terms of three attribute of local system conspicuousness, global system conspicuousness and rare conspicuousness.
Each pixel vision significance not only relies only on the pixel characteristic value on image to be processed, but needs foundation
The difference degree of the pixel and its surrounding pixel point.Difference degree is bigger, then illustrates that the pixel more has vision significance.
The relevant technologies calculate its vision significance by the difference degree between each pixel of image and peripheral region, but region
Size is difficult to determine, while algorithm calculation amount is very huge.In order to solve these problems in the related technology, the application can be from frequency
The local system conspicuousness of analytical calculation image slices vegetarian refreshments is carried out in domain, and the effect of amplitude spectrum and phase spectrum in image frequency domain feature is deposited
In difference: amplitude spectrum includes the specific size of each pixel characteristic value;Phase spectrum includes image spatial feature, its energy
Enough reflect the situation of change of image pixel point feature value.
Present inventor has found that role during picture construction of phase spectrum and amplitude spectrum is different after study.
It is constructed using phase spectrum to be directed to result images merely, the result that there is approximate construction with original image can be obtained;And only
Only applies amplitude spectrum is constructed to be directed to result images, and the difference of acquired results image and original image is very big.
Therefore, formula 3 can be used to be directed to each characteristic image and carry out local system conspicuousness calculating in advance.First against
Input picture executes Fast Fourier Transfomation Process (Fast Fourier Transform, FFT) to extract phase spectrum and amplitude
Spectrum;Then it is constructed using phase spectrum for image, to obtain the local system notable figure of each characteristic image.
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and FFT (u, v) is pixel characteristic value, | FFT (u, v) eJ ψ (u, v)|
For the gained picture amplitude spectrum after Fast Fourier Transform (FFT), ψ (u, v) is image phase spectrum to be processed, VLocalIt is in image
The local system significance value of each image slices vegetarian refreshments.
If only considering local system significance value, it will cause to change violent edge or complex background area in image
The global system conspicuousness in domain is higher, while the local system conspicuousness of smooth region and target internal is then lower.In order to solve
The above problem can introduce global system conspicuousness.
The global system conspicuousness of pixel is for measuring the vision significance of the image slices vegetarian refreshments in entire image
Degree can generate the global system conspicuousness of image to be processed using formula 5.V in formula 5Global(x, y) is to be processed
The global system significance value of each pixel in image.
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and f (x, y) is the significance function for solving pixel (x, y),
faverage(x, y) is the arithmetic mean of instantaneous value of f (x, y), and image size to be processed is M*N.
The chance probability that rare conspicuousness indicates that certain characteristic value occurs in entire image is very low, then has this feature
The image slices vegetarian refreshments " unusual " of value, then the vision significance value of image slices vegetarian refreshments will be higher.It can be measured using formula 6
The rare significant characteristics of each pixel in input picture.H (the f of formula 6average(x, y)) it is what image to be processed generated
Feature histogram, VScarcity(x, y) is the rare significance value of pixel:
The local system significance value of each pixel of image to be processed, global system significance value and rare is calculated
It is corresponding after significance value, produce local system notable figure, global system Saliency maps and rare Saliency maps.And most
The vision significance figure obtained eventually is to merge these three vision significance figures, optionally, using formula 7 for calculating
The measurement results of the local system conspicuousness of some characteristic pattern (such as color characteristic), global system conspicuousness, rare conspicuousness
It is merged, to obtain final system notable figure.
In formula, VfinalFor the vision significance value of each pixel of image to be processed, VLocal(x,y)、VGlobal(x,y)、
VScarcity(x,y)(VLocal、VGlobal、VScarcity) it is followed successively by the local system conspicuousness for the pixel that transverse and longitudinal coordinate value is x, y
Value, global system significance value and rare significance value;v1For the weighted value of local system significance value, v2It is aobvious for global system
The weighted value of work property value, v3For the weighted value of rare significance value;I=1, Vi=V1For local system significance value, i=2, Vi
=V2For whole system significance value, i=3, Vi=V3For rare significance value, V1、V2、V3According toCalculate gained.
There is the contribution of image segmentation different in different characteristic remarkable pictures.Some visual saliency maps being capable of effective expression view
Feel marking area, and some visual saliency maps cannot then reach ideal effect.Therefore, it is necessary to reasonable Fusion Features strategy for
The multiple characteristic remarkable pictures of gained are merged and ultimately generate system visual saliency map.From vision significance point quantity, position, minute
Cloth situation come be dynamically selected feature and weighting treatment process.
It is calculated firstly the need of image threshold is executed for characteristic remarkable picture, extracts the vision saliency value pixel bigger than threshold value
As significant point.According to rare conspicuousness criterion, the conspicuousness point quantity on characteristic remarkable picture is more, then this characteristic remarkable picture pair
The percentage contribution of system notable figure is with regard to smaller.Define weight WregionIt is conspicuousness point quantity, as shown in formula 8.
Wregion=Nsaliency; (8)
The mankind are easy to concentrate on picture centre region part for the attention region of image.Picture centre region pixel
And peripheral region is easier to become salient region.Therefore, it can define weight WpixelFor in each significant pixel and image
The Euclidean distance arithmetic mean of instantaneous value of imago vegetarian refreshments, specific method are as shown in formula 9.
N in formula 9 is significant point quantity, sp in system features notable figureiIt is i-th of system significant point, center is
Picture centre pixel specific location.
If each Visual accent position in system features notable figure is scattered in each region of notable figure, assert
This system characteristic remarkable picture is inadequate to the percentage contribution of the final notable figure of system.Therefore, it can define WeffectWeight is each system
Euclidean distance arithmetic mean of instantaneous value between significant point, centeridIt is specific as shown in formula 10 for Visual accent place-centric.
According to system Visual accent distributing position and quantity situation, it is significant each system features can be calculated using formula 11
The related weight of figure merges each system features notable figure and obtains final system notable figure.
In formula 11, VfiIt is i-th of system features notable figure,It is that i-th of system is special
Levy notable figure affiliated area weight, pixel weight, effect weight, WfiIt is three kinds of weights in i-th of system features notable figure
Fusion, WiIt is final weight belonging to i-th of system features notable figure, V is final system notable figure.
From the foregoing, it will be observed that the embodiment of the present invention by local system conspicuousness, global system conspicuousness, rare conspicuousness come into
The specific vision significance of row illustrates description, would be even more beneficial to carry out Fast image segmentation process.By carrying out vision significance machine
It makes perfect, continuously improves and carry out image segmentation process with application vision significance mechanism, to reach better image point
Cut effect.
For S104, in another embodiment, can be used weighting vision significance result figure merging method with
The merging method of average visual significant result figure carries out the union operation of vision significance figure.By present inventor's
Test of many times proves, weights the merging method of vision significance result figure than average visual significant result figure merging method effect
More preferably.P coordinate position on i-th layer, vision significance value areWherein[0,1] region model is normalized
In enclosing.To calculate the merging vision significance value on i-th layer, the vision significance result figure on (i-1)-th layer can be adjusted to and
I-th layer of upper vision significance result figure same size.Use vision significance valueAfter indicating to operate on image sequence
Vision significance result figureThe vision significance value of upper p coordinate position.The detailed-oriented vision of adjacent two layers is aobvious
Work value merge after result as shown in formula 12, formula 13 and formula 14:
In formula, p is pixel position,For i-th layer of weighted value,For (i-1)-th layer of weighted value,It is i-th layer
Visualization significance value,For (i-1)-th layer of vision significance value;For the pixel for merging the position visual saliency map p
Point vision significance value.
According to formula 13 and formula 14 it is found that in i-th layer of detailed-oriented vision significance valueWith (i-1)-th layer of detailed-oriented view
Feel significance valueIn the case where having determined, so that it may calculateWithWhenWhen,It can basis
Formula formula 12 can calculate the detailed-oriented vision significance value after merging.
By the result that present inventor is tested using the method for above-described embodiment it is found that after union operation
Vision significance figure the respective vision significance Analysis On Multi-scale Features of two levels can be merged, by fusion after
The vision significance result figure arrived is more smooth and clear.
Using random search detection method selection candidate image area block during, due to sample collection quantity not enough with
And the random causes such as sample error will lead to the random presence in the vision significance figure of generation comprising much noise.Random noise
Signal Producing reason is because random selection 2K image-region block, which executes formula 1, generates picture noise being calculated.For
The accuracy for promoting vision significance figure carries out denoising after S103.
In one embodiment, it can be used eight neighborhood vision significance value can be for the visual saliency map changed roughly
(visual saliency map that S103 is generated) executes detailed-oriented process.Eight neighborhood coordinate method can be from corresponding eight directions pixel r
Vicinity points selection is carried out, the candidate image area block acquired by eight neighborhood coordinate method and random selection candidate image
Region block method has very big difference.Because adjacent domain image similarity is high at p coordinate, eight neighborhood coordinate method may make
Obtaining vision significance value at p coordinate can be less than normal than real image vision significance value;If according to what is obtained for the method
Vision significance value is normalized, then will lead to generate on the vision significance figure that aforementioned gained is changed roughly and accordingly make an uproar
Sound.But the vision significance such as similarity of neighbouring coordinate is high and adjacent, it is therefore desirable to be directed to and neighborhood vision significance value
The pixel coordinate position that differs greatly carries out detailed-oriented operation.In view of the confidence level factor of vision significance, between eight neighborhood
The confidence level for distinguishing biggish vision significance value is lower, therefore may be selected that detailed-oriented that biggish vision is distinguished between eight neighborhood is significant
Property value.
In addition, present invention also provides another embodiment, referring to Fig. 2, Fig. 2 is provided in an embodiment of the present invention another
A kind of flow diagram of vision significance method for detecting area, the embodiment of the present invention for example can be applied to image partition method
In, specifically may include the following contents:
S201: multiple candidate image areas are chosen for current pixel point from image to be processed using stochastic search methods
Block.
S202: the similarity magnitude of current pixel point and each candidate image area block is calculated separately.
S203: deleting similarity magnitude and be lower than the corresponding candidate image area block of similarity threshold, by remaining candidate figure
As region unit is as object region block.
S204: according to the similarity between each target image block and current pixel point, the vision for calculating current pixel point is aobvious
Work property value.
S205: the vision significance value of each pixel based on image to be processed generates the thick of each level of image to be processed
Slightly change visual saliency map.
S206: detailed-oriented operation is carried out to each roughization visual saliency map and obtains respective phase to remove picture noise signal
The detailed-oriented visual saliency map answered.
S207: according to the vision significance value of adjacent two layers visual saliency map, the weighted value of each layer visual saliency map is calculated.
S208: each layer visual saliency map is merged according to the weighted value of each layer visual saliency map, obtains initial visual
Notable figure.
S209: selecting multiple predeterminated position points in initial visual notable figure, to each predeterminated position point, according to each picture
The vision significance value of vegetarian refreshments is classified as strengthening pixel point set and weakens pixel point set.
The number of location point can be selected according to the size of image to be processed, the actual conditions of visual saliency map, this Shen
Any restriction is not done to this please.
The vision significance value for strengthening the pixel that pixel is concentrated is greater than the vision for the pixel that reduction pixel is concentrated
Significance value.
High being divided to of vision significance value can be strengthened pixel point set according to experience by those skilled in the art, will be low
The corresponding pixel of vision significance value is divided to reduction pixel point set.Or the vision for the pixel that can also be gone out according to each position
Significance value and total number are arranged a vision significance and divide threshold value, and the pixel that will be above the threshold value is divided to reinforcing picture
Vegetarian refreshments collection, the pixel that will be less than the threshold value are divided to reduction pixel point set.
S210: intensive treatment is carried out to each pixel that pixel is concentrated is strengthened using image enchancing method.
Any image enchancing method that enhancing pixel point feature can be achieved can be used, those skilled in the art can basis
Actual demand and practical application scene are selected, and the application does not do any restriction to this.
As to how realizing the process of image enhancement, the implementation of the relevant technologies record is seen, it is herein, just no longer superfluous
It states.
S211: Weakening treatment is carried out using each pixel that image method for weakening concentrates reduction pixel.
Any image enchancing method that reduction pixel point feature can be achieved can be used, those skilled in the art can basis
Actual demand and practical application scene are selected, and the application does not do any restriction to this.
As to how realizing the process of image reduction, the implementation of the relevant technologies record is seen, it is herein, just no longer superfluous
It states.
Step or method same with the above-mentioned embodiment in the embodiment of the present invention see the reality of above-described embodiment description
Existing process repeats no more herein.
From the foregoing, it will be observed that 4 stages of invention point of the embodiment of the present invention realize the detection in vision significance region, the 1st stage
For the random detection stage, random search and detection processing are carried out according to image to be processed, obtained using each hierarchical information of original image
To the roughization visual saliency map of each level;2nd stage is the detailed-oriented stage, significant using the roughization vision on each level
Figure carries out detailed-oriented operation, removes in roughization visual saliency map the image generated because of random search and detection processing and makes an uproar
Sound;3rd stage is the multi-level merging phase of administrative division map, and by carrying out to multilayer visual saliency map, detailed-oriented merging is final to be produced
The visual saliency map of combination local feature and global characteristics after intercrescence simultaneously;4th stage is the more new vision saliency value stage, will
Selectively the vision significance region high for those noise signals merges, so that it is detailed-oriented to obtain final high quality
Vision significance result figure.It is identical careful with original input picture (image to be processed) size to realize quick, real-time generation
Change vision significance result figure, to improve the overall quality of object segmentation result in sequence of video images.
In order to confirm that technical solution provided by the present application (being generally referred to as multiscalization vision significance detection algorithm) can
Detailed-oriented vision significance result figure identical with original input picture size is quickly generated, can be applied to want real-time
The higher vision significance result figure based on image sequence is sought, to improve the total of object segmentation result in sequence of video images
Weight.The application has carried out a series of experiments using Matlab simulated environment and has verified.
Under Microsoft Windows10 operating system environment, this multiscalization vision significance detection algorithm is realized.
Using vision significance detection algorithm, corresponding vision significance result figure is generated according to original input picture.Obtain a large amount of visions
The bright multiscalization vision significance detection method of significant result chart can be compared with existing mainstream vision significance measure
The visual saliency map of excellent effect, the application and 8 kinds of the relevant technologies are obtained on original input picture, such as AIM (is based on information theory
Conspicuousness detection algorithm), GBVS (the conspicuousness detection algorithm based on graph theory), SR (based on spectrum surplus conspicuousness inspection
Method of determining and calculating), IS (the conspicuousness dimensioning algorithm based on sparse marking area), ICL (dynamic vision caution detection algorithm), ITTI
(most classical model algorithm), RC (the global contrast algorithm based on marking area detection), SUN are (using the conspicuousness of statistics naturally
Bayesian frame model) compare.
6 width test images have been selected to be used to compare the method for the present invention and eight from 120 width original images of Bruce database
Kind the relevant technologies algorithm.For effect accuracy, the execution accuracy highest of this multiscalization method.
In experimental result it is found that for the facial image for holding card, face and hand-held card are human eye focal points,
This situation only has the application that could preferably detect range.It is cycled in image for people, the people of cycling is salient region, RC
It can preferably detect that the salient region in image, context of methods have also detected that corresponding conspicuousness area with SUN method
Domain.For pine tree image, " cavity " that the sparse part of pine needle is indistinctly formed in image is the marking area that human eye is most paid close attention to,
The vision significance detection difficulty of this image is larger, and range, most methods are not detected in eight kinds of related art methods
Can be the stronger white high bright region misidentification of image lower left corner contrast at visual salient region, and the application then detects substantially
Empty general area is gone out.On from the contrast in visual salient region and the non-significant region of vision, the application is also optimal.For
One width includes the scenery picture of portrait, although IS and ICL can detect the personage in image, the application contrast highest.
In order to objectively evaluate the specific manifestation of the application, using Receiver operating curve (Receiver
Operating Characteristic, ROC) analysis of Lai Jinhang quantitative result.As shown in Figure 3 and Figure 4, the application curve is most
It is high.Be segmented into two classes using eight kinds of methods: the simple algorithm for using prior probability knowledge, they include AIM, GBVS, IS and
SR;The simple algorithm for using Current observation image information, they include ICL, ITTI, RC and SUN.So, this multiscalization regards
Feel that conspicuousness detection method is more preferable than the method effect that prior probability knowledge or Current observation image information is used only.
Assessing the current majority of case of Segmentation of Color Image is all to execute subjective judgement by human eye, by present techniques
Scheme is compared to human eye segmentation result in Berkeley graphics standard library to this calculation applied to the result after image segmentation
Method carries out qualitative evaluation, the image segmentation that the application has had.Such as the landscape figure for a width bird in the tree, Fig. 5 are
The image divided using technical scheme, by Fig. 5 segmentation result as it can be seen that technical scheme can be in the picture
Salient region is accurately positioned, the wheel of bird can be clearly cut into using the image partition method of technical scheme
Wide and tree profile, can obtain and divide almost consistent image segmentation result with human eye.
From the foregoing, it will be observed that the embodiment of the present invention introduces random search detection method for vision significance, view is increased substantially
Feel the efficiency of conspicuousness detection algorithm, and obtains aobvious with the big small-scale identical detailed-oriented vision of original input picture
Work property figure.
The embodiment of the present invention provides corresponding realization device also directed to vision significance method for detecting area, further makes
It obtains the method and has more practicability.It is situated between below to vision significance regional detection device provided in an embodiment of the present invention
It continues, vision significance regional detection device described below can be mutually right with above-described vision significance method for detecting area
It should refer to.
Referring to Fig. 6, Fig. 6 is vision significance regional detection device provided in an embodiment of the present invention in a kind of specific embodiment party
Structure chart under formula, the device can include:
Random search module 601 is more for being chosen using stochastic search methods from image to be processed for current pixel point
A candidate image area block chooses the object region block for meeting similarity condition from each candidate image area block.
Vision significance value computing module 602, for according to the similarity between each target image block and current pixel point,
Calculate the vision significance value of current pixel point.
Multilayer visual saliency map generation module 603, for the vision significance value of each pixel based on image to be processed,
Generate the visual saliency map of each level of image to be processed.
Visual saliency map Fusion Module 604 calculates each for the vision significance value according to adjacent two layers visual saliency map
The weighted value of layer visual saliency map, to melt each layer visual saliency map for the weighted value according to each layer visual saliency map
It closes.
Optionally, in some embodiments of the present embodiment, referring to Fig. 7, described device for example can also include view
Feel conspicuousness update module 605, is selected in the visual saliency map that the vision significance update module 605 is used to obtain after fusion
Select multiple predeterminated position points;To each predeterminated position point, according to the vision significance value of each pixel, it is classified as strengthening picture
Vegetarian refreshments collection and reduction pixel point set, the vision significance value for strengthening the pixel that pixel is concentrated are greater than what reduction pixel was concentrated
The vision significance value of pixel;Intensive treatment is carried out to each pixel that pixel is concentrated is strengthened using image enchancing method;
Weakening treatment is carried out using each pixel that image method for weakening concentrates reduction pixel.
In a kind of specific embodiment, the multilayer visual saliency map generation module 603 may also include that
Roughization visual saliency map generates submodule, the vision significance for each pixel based on image to be processed
Value, generates the roughization visual saliency map of each level of image to be processed;
Detailed-oriented visual saliency map generates submodule, for carrying out detailed-oriented operation to each roughization visual saliency map, with
Picture noise signal is removed, respective detailed-oriented visual saliency map is obtained.
In other some embodiments, the random search module 601 can also be calculate separately current pixel point and
The similarity magnitude of each candidate image area block;It deletes similarity magnitude and is lower than the corresponding candidate image area of similarity threshold
Block, using remaining candidate image area block as the module of object region block.
Optionally, in some specific embodiments, the vision significance value computing module 602 may be, for example, utilization
Following formula calculate the module of the vision significance value S of current pixel point:
Wherein,
In formula, dist (ri, rk) dissimilarity degree between current pixel point position and k-th target image block
Magnitude, K are the total number of target image block, distcolor(ri, rk) it is the object region block that is handled by vectorization and work as
Euclidean distance of the preceding pixel point position on hsv color space, distpos(ri, rk) be current pixel point position with
Euclidean distance between k-th target image block.
In addition, the multilayer visual saliency map generation module 603 may also include that
Local system notable figure generates submodule, the part system for each pixel based on frequency-domain calculations image to be processed
System significance value, generates local system notable figure;
Global system notable figure generate submodule, for calculated using following formula image to be processed each pixel it is complete
Office system significance value VGlobal(x, y) generates global system notable figure:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and f (x, y) is the significance function for solving pixel (x, y),
faverage(x, y) is the arithmetic mean of instantaneous value of f (x, y), and image size to be processed is M*N;
Rare notable figure generates submodule, and the rare of each pixel for being calculated image to be processed using following formula is shown
Work property value VScarcity(x, y) generates rare notable figure:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, faverage(x, y) is the arithmetic mean of instantaneous value of f (x, y), h (faverage
(x, y)) it is the feature histogram that image to be processed generates;
Fusion submodule is obtained for merging local system notable figure, global system notable figure and rare notable figure
To the visual saliency map of each level of image to be processed.
In embodiments of the present invention, the fusion submodule for example can also be each layer to image to be processed, and utilization is following
Formula calculates the vision significance value V of each pixel of image to be processedfinal, generate the module of visual saliency map:
In formula, VLocal、VGlobal、VScarcityIt is followed successively by the local system conspicuousness for the pixel that transverse and longitudinal coordinate value is x, y
Value, global system significance value and rare significance value;v1For the weighted value of local system significance value, v2It is aobvious for global system
The weighted value of work property value, v3For the weighted value of rare significance value, i=1, Vi=V1For local system significance value, i=2, Vi
=V2For whole system significance value, i=3, Vi=V3For rare significance value, V1、V2、V3According toCalculate gained.
Optionally, it can also be to calculate image to be processed using following formula that the local system notable figure, which generates submodule,
The local system significance value V of each pixelLocalThe module of (x, y):
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and FFT (u, v) is pixel characteristic value, | FFT (u, v) eJ ψ (u, v)|
For the gained picture amplitude spectrum after Fast Fourier Transform (FFT), ψ (u, v) is image phase spectrum to be processed.
In some embodiments, the visual saliency map Fusion Module 604 can also be to calculate each layer using following formula
The module of the weighted value of visual saliency map:
In formula, p is pixel position,For i-th layer of weighted value,For (i-1)-th layer of weighted value,It is i-th layer
Visualization significance value,For (i-1)-th layer of vision significance value;
The visual saliency map of adjacent two layers is merged using following formula, obtains fusion visual saliency map;
In formula,For the pixel vision significance value for merging the position visual saliency map p.
The function of each functional module of vision significance regional detection device described in the embodiment of the present invention can be according to above-mentioned side
Method specific implementation in method embodiment, specific implementation process are referred to the associated description of above method embodiment, herein
It repeats no more.
From the foregoing, it will be observed that the embodiment of the present invention improves the computational efficiency of pixel vision significance value, to effectively mention
The detection efficiency in the vision significance region of image to be processed has been risen, the vision significance region in image to be processed can be also promoted
Accuracy in detection and precision.
The embodiment of the invention also provides a kind of vision significance equipment for area detection equipment, specifically can include:
Memory, for storing computer program;
Processor realizes vision significance region detection described in any one embodiment as above for executing computer program
The step of method.
The function of each functional module of vision significance equipment for area detection equipment described in the embodiment of the present invention can be according to above-mentioned side
Method specific implementation in method embodiment, specific implementation process are referred to the associated description of above method embodiment, herein
It repeats no more.
From the foregoing, it will be observed that the embodiment of the present invention improves the computational efficiency of pixel vision significance value, to effectively mention
The detection efficiency in the vision significance region of image to be processed has been risen, the vision significance region in image to be processed can be also promoted
Accuracy in detection and precision.
The embodiment of the invention also provides a kind of computer readable storage mediums, are stored with vision significance region detection journey
Sequence, vision significance region described in any one embodiment as above when the vision significance region detector is executed by processor
The step of detection method.
The function of each functional module of computer readable storage medium described in the embodiment of the present invention can be according to above method reality
The method specific implementation in example is applied, specific implementation process is referred to the associated description of above method embodiment, herein no longer
It repeats.
From the foregoing, it will be observed that the embodiment of the present invention improves the computational efficiency of pixel vision significance value, to effectively mention
The detection efficiency in the vision significance region of image to be processed has been risen, the vision significance region in image to be processed can be also promoted
Accuracy in detection and precision.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with it is other
The difference of embodiment, same or similar part may refer to each other between each embodiment.For being filled disclosed in embodiment
For setting, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is referring to method part
Explanation.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
A kind of vision significance method for detecting area provided by the present invention and device are described in detail above.This
Apply that a specific example illustrates the principle and implementation of the invention in text, the explanation of above example is only intended to
It facilitates the understanding of the method and its core concept of the invention.It should be pointed out that for those skilled in the art,
Without departing from the principles of the invention, can be with several improvements and modifications are made to the present invention, these improvement and modification are also fallen
Enter in the protection scope of the claims in the present invention.
Claims (10)
1. a kind of vision significance method for detecting area characterized by comprising
Multiple candidate image area blocks are chosen for current pixel point from image to be processed using stochastic search methods, from each candidate
The object region block for meeting similarity condition is chosen in image-region block;
According to the similarity between each target image block and the current pixel point, the vision for calculating the current pixel point is significant
Property value;
The vision significance value of each pixel based on the image to be processed generates the vision of each level of image to be processed
Notable figure;
According to the vision significance value of adjacent two layers visual saliency map, the weighted value of each layer visual saliency map is calculated, for pressing
Each layer visual saliency map is merged according to the weighted value of each layer visual saliency map.
2. vision significance method for detecting area according to claim 1, which is characterized in that described to be regarded according to adjacent two layers
The vision significance value for feeling notable figure, after the weighted value for calculating each layer visual saliency map, further includes:
Multiple predeterminated position points are selected in the visual saliency map obtained after fusion;
To each predeterminated position point, according to the vision significance value of each pixel, it is classified as strengthening pixel point set and reduction
Pixel point set, the vision significance value for strengthening the pixel that pixel is concentrated are all larger than the picture that the reduction pixel is concentrated
The vision significance value of vegetarian refreshments;
Intensive treatment is carried out using each pixel that image enchancing method concentrates the reinforcing pixel;
Weakening treatment is carried out using each pixel that image method for weakening concentrates the reduction pixel.
3. vision significance method for detecting area according to claim 1, which is characterized in that described based on described to be processed
The vision significance value of each pixel of image, the visual saliency map for generating each level of image to be processed include:
The vision significance value of each pixel based on the image to be processed generates the rough of each level of image to be processed
Change visual saliency map;
Detailed-oriented operation is carried out to each roughization visual saliency map, to remove picture noise signal, it is respective careful to obtain
Change visual saliency map.
4. vision significance method for detecting area according to claim 1, which is characterized in that described from each candidate image area
It is chosen in the block of domain and meets the object region block of similarity condition and include:
Calculate separately the similarity magnitude of the current pixel point and each candidate image area block;
It deletes similarity magnitude and is lower than the corresponding candidate image area block of similarity threshold, remaining candidate image area block is made
For object region block.
5. vision significance method for detecting area according to any one of claims 1 to 4, which is characterized in that described
According to the similarity between each target image block and the current pixel point, the vision significance value packet of the current pixel point is calculated
It includes:
The vision significance value S of the current pixel point is calculated using following formula:
Wherein,
In formula, dist (ri, rk) dissimilarity degree between the current pixel point position and k-th target image block
Magnitude, K are the total number of target image block, distcolor(ri, rk) for the object region block handled by vectorization and institute
State Euclidean distance of the current pixel point position on hsv color space, distpos(ri,rk) for the current pixel point institute
Euclidean distance between position and k-th target image block.
6. vision significance method for detecting area according to any one of claims 1 to 4, which is characterized in that the base
In the vision significance value of each pixel of the image to be processed, the visual saliency map of each level of image to be processed is generated
Include:
It is significant to generate local system for the local system significance value of each pixel based on image to be processed described in frequency-domain calculations
Figure;
The global system significance value V of each pixel of the image to be processed is calculated using following formulaGlobal(x, y) is generated
Global system notable figure:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and f (x, y) is the significance function for solving pixel (x, y), faverage(x,
It y) is the arithmetic mean of instantaneous value of f (x, y), the image size to be processed is M*N;
The rare significance value V of each pixel of the image to be processed is calculated using following formulaScarcity(x, y) is generated dilute
Lack notable figure:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, faverage(x, y) is the arithmetic mean of instantaneous value of f (x, y), h (faverage(x,
Y)) the feature histogram generated for the image to be processed;
The local system notable figure, the global system notable figure and the rare notable figure are merged, obtained described
The visual saliency map of each level of image to be processed.
7. vision significance method for detecting area according to claim 6, which is characterized in that described by the local system
Notable figure, the global system notable figure and the rare notable figure are merged, and each level of image to be processed is obtained
Visual saliency map includes:
To each layer of the image to be processed, the vision for calculating each pixel of the image to be processed using following formula is significant
Property value Vfinal, generate visual saliency map:
In formula, VLocal、VGlobal、VScarcityIt is followed successively by the local system significance value, complete for the pixel that transverse and longitudinal coordinate value is x, y
Office system significance value and rare significance value;v1For the weighted value of local system significance value, v2For global system significance value
Weighted value, v3For the weighted value of rare significance value;I=1, Vi=V1For the local system significance value, i=2, Vi=V2
For whole system significance value, i=3, Vi=V3For the rare significance value, V1、V2、V3According toCalculate gained.
8. vision significance method for detecting area according to claim 6, which is characterized in that described to be based on frequency-domain calculations institute
The local system significance value for stating each pixel of image to be processed includes:
The local system significance value VLocal (x, y) of each pixel of the image to be processed is calculated using following formula:
In formula, x, y are the transverse and longitudinal coordinate value of pixel, and FFT (u, v) is pixel characteristic value, | FFT (u, v) ejψ(u,v)| to pass through
Gained picture amplitude spectrum after Fast Fourier Transform (FFT), ψ (u, v) are the image phase spectrum to be processed.
9. vision significance method for detecting area according to any one of claims 1 to 4, which is characterized in that described
According to the vision significance value of adjacent two layers visual saliency map, the weighted value of each layer visual saliency map is calculated, for according to each layer
Each layer visual saliency map is carried out fusion by the weighted value of visual saliency map
The weighted value of each layer visual saliency map is calculated using following formula:
In formula, p is pixel position,For i-th layer of weighted value,For (i-1)-th layer of weighted value,For i-th layer of view
Feelization significance value,For (i-1)-th layer of vision significance value;
The visual saliency map of adjacent two layers is merged using following formula, obtains fusion visual saliency map;
In formula,For the pixel vision significance value of the fusion position visual saliency map p.
10. a kind of vision significance regional detection device characterized by comprising
Random search module, for choosing multiple candidate figures from image to be processed using stochastic search methods for current pixel point
As region unit, the object region block for meeting similarity condition is chosen from each candidate image area block;
Vision significance value computing module, for according to the similarity between each target image block and the current pixel point, meter
Calculate the vision significance value of the current pixel point;
Multilayer visual saliency map generation module, it is raw for the vision significance value of each pixel based on the image to be processed
At the visual saliency map of each level of image to be processed;
Visual saliency map Fusion Module calculates each layer vision for the vision significance value according to adjacent two layers visual saliency map
The weighted value of notable figure, to merge each layer visual saliency map for the weighted value according to each layer visual saliency map.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811455470.1A CN109543701A (en) | 2018-11-30 | 2018-11-30 | Vision significance method for detecting area and device |
PCT/CN2019/075206 WO2020107717A1 (en) | 2018-11-30 | 2019-02-15 | Visual saliency region detection method and apparatus |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811455470.1A CN109543701A (en) | 2018-11-30 | 2018-11-30 | Vision significance method for detecting area and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109543701A true CN109543701A (en) | 2019-03-29 |
Family
ID=65851847
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811455470.1A Pending CN109543701A (en) | 2018-11-30 | 2018-11-30 | Vision significance method for detecting area and device |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109543701A (en) |
WO (1) | WO2020107717A1 (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110264545A (en) * | 2019-06-19 | 2019-09-20 | 北京字节跳动网络技术有限公司 | Picture Generation Method, device, electronic equipment and storage medium |
CN110910830A (en) * | 2019-11-29 | 2020-03-24 | 京东方科技集团股份有限公司 | Display brightness adjusting method, display system, computer device and medium |
CN111079730A (en) * | 2019-11-20 | 2020-04-28 | 北京云聚智慧科技有限公司 | Method for determining area of sample image in interface image and electronic equipment |
CN111083477A (en) * | 2019-12-11 | 2020-04-28 | 北京航空航天大学 | HEVC (high efficiency video coding) optimization algorithm based on visual saliency |
CN113140005A (en) * | 2021-04-29 | 2021-07-20 | 上海商汤科技开发有限公司 | Target object positioning method, device, equipment and storage medium |
WO2022033088A1 (en) * | 2020-08-14 | 2022-02-17 | 北京迈格威科技有限公司 | Image processing method, apparatus, electronic device, and computer-readable medium |
WO2022141476A1 (en) * | 2020-12-31 | 2022-07-07 | 华为技术有限公司 | Image processing method, data obtaining method, and device |
CN117058723A (en) * | 2023-10-11 | 2023-11-14 | 腾讯科技(深圳)有限公司 | Palmprint recognition method, palmprint recognition device and storage medium |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111860643B (en) * | 2020-07-20 | 2023-10-03 | 苏州大学 | Visual template matching robustness improving method based on frequency modulation model |
CN112418223A (en) * | 2020-12-11 | 2021-02-26 | 互助土族自治县北山林场 | Wild animal image significance target detection method based on improved optimization |
CN113256581B (en) * | 2021-05-21 | 2022-09-02 | 中国科学院自动化研究所 | Automatic defect sample labeling method and system based on visual attention modeling fusion |
CN114022778B (en) * | 2021-10-25 | 2023-04-07 | 电子科技大学 | SAR (synthetic Aperture Radar) berthing ship detection method based on significance CNN (CNN) |
CN114332491A (en) * | 2021-12-03 | 2022-04-12 | 安徽理工大学 | Saliency target detection algorithm based on feature reconstruction |
CN114758119B (en) * | 2022-04-20 | 2024-06-07 | 北京航空航天大学 | Sea surface recovery target detection method based on eagle eye imitating vision and similar physical properties |
CN114926657B (en) * | 2022-06-09 | 2023-12-19 | 山东财经大学 | Saliency target detection method and system |
CN115147409B (en) * | 2022-08-30 | 2022-11-29 | 深圳市欣冠精密技术有限公司 | Mobile phone shell production quality detection method based on machine vision |
CN115272335A (en) * | 2022-09-29 | 2022-11-01 | 江苏万森绿建装配式建筑有限公司 | Metallurgical metal surface defect detection method based on significance detection |
CN116645368B (en) * | 2023-07-27 | 2023-10-03 | 青岛伟东包装有限公司 | Online visual detection method for edge curl of casting film |
CN116740098B (en) * | 2023-08-11 | 2023-10-27 | 中色(天津)新材料科技有限公司 | Aluminum alloy argon arc welding image segmentation method and system |
CN116721107B (en) * | 2023-08-11 | 2023-11-03 | 青岛胶州电缆有限公司 | Intelligent monitoring system for cable production quality |
CN117419650B (en) * | 2023-12-18 | 2024-02-27 | 湖南西欧新材料有限公司 | Alumina ceramic surface glaze layer thickness measuring method based on visual analysis |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090324088A1 (en) * | 2008-06-30 | 2009-12-31 | Christel Chamaret | Method for detecting layout areas in a video image and method for generating an image of reduced size using the detection method |
CN106407927A (en) * | 2016-09-12 | 2017-02-15 | 河海大学常州校区 | Salient visual method based on polarization imaging and applicable to underwater target detection |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2431919A1 (en) * | 2010-09-16 | 2012-03-21 | Thomson Licensing | Method and device of determining a saliency map for an image |
CN103049893B (en) * | 2011-10-14 | 2015-12-16 | 深圳信息职业技术学院 | A kind of method of image fusion quality assessment and device |
CN104463907A (en) * | 2014-11-13 | 2015-03-25 | 南京航空航天大学 | Self-adaptation target tracking method based on vision saliency characteristics |
CN107103608B (en) * | 2017-04-17 | 2019-09-27 | 大连理工大学 | A kind of conspicuousness detection method based on region candidate samples selection |
-
2018
- 2018-11-30 CN CN201811455470.1A patent/CN109543701A/en active Pending
-
2019
- 2019-02-15 WO PCT/CN2019/075206 patent/WO2020107717A1/en active Application Filing
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090324088A1 (en) * | 2008-06-30 | 2009-12-31 | Christel Chamaret | Method for detecting layout areas in a video image and method for generating an image of reduced size using the detection method |
CN106407927A (en) * | 2016-09-12 | 2017-02-15 | 河海大学常州校区 | Salient visual method based on polarization imaging and applicable to underwater target detection |
Non-Patent Citations (2)
Title |
---|
徐军: "采用显著性分析与改进边缘方向直方图特征的红外与可见光图像配准", 《光学精密工程》 * |
陈沅涛: "图像视觉显著性和改进型SVM在图像分割中的研究", 《中国优秀博士学位论文全文数据库 信息科技辑》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110264545A (en) * | 2019-06-19 | 2019-09-20 | 北京字节跳动网络技术有限公司 | Picture Generation Method, device, electronic equipment and storage medium |
CN111079730B (en) * | 2019-11-20 | 2023-12-22 | 北京云聚智慧科技有限公司 | Method for determining area of sample graph in interface graph and electronic equipment |
CN111079730A (en) * | 2019-11-20 | 2020-04-28 | 北京云聚智慧科技有限公司 | Method for determining area of sample image in interface image and electronic equipment |
CN110910830A (en) * | 2019-11-29 | 2020-03-24 | 京东方科技集团股份有限公司 | Display brightness adjusting method, display system, computer device and medium |
CN110910830B (en) * | 2019-11-29 | 2021-02-12 | 京东方科技集团股份有限公司 | Display brightness adjusting method, display system, computer device and medium |
CN111083477A (en) * | 2019-12-11 | 2020-04-28 | 北京航空航天大学 | HEVC (high efficiency video coding) optimization algorithm based on visual saliency |
CN111083477B (en) * | 2019-12-11 | 2020-11-10 | 北京航空航天大学 | HEVC (high efficiency video coding) optimization algorithm based on visual saliency |
WO2022033088A1 (en) * | 2020-08-14 | 2022-02-17 | 北京迈格威科技有限公司 | Image processing method, apparatus, electronic device, and computer-readable medium |
WO2022141476A1 (en) * | 2020-12-31 | 2022-07-07 | 华为技术有限公司 | Image processing method, data obtaining method, and device |
CN113140005A (en) * | 2021-04-29 | 2021-07-20 | 上海商汤科技开发有限公司 | Target object positioning method, device, equipment and storage medium |
CN113140005B (en) * | 2021-04-29 | 2024-04-16 | 上海商汤科技开发有限公司 | Target object positioning method, device, equipment and storage medium |
CN117058723A (en) * | 2023-10-11 | 2023-11-14 | 腾讯科技(深圳)有限公司 | Palmprint recognition method, palmprint recognition device and storage medium |
CN117058723B (en) * | 2023-10-11 | 2024-01-19 | 腾讯科技(深圳)有限公司 | Palmprint recognition method, palmprint recognition device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
WO2020107717A1 (en) | 2020-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109543701A (en) | Vision significance method for detecting area and device | |
JP7413400B2 (en) | Skin quality measurement method, skin quality classification method, skin quality measurement device, electronic equipment and storage medium | |
Fang et al. | Bottom-up saliency detection model based on human visual sensitivity and amplitude spectrum | |
CN107730515B (en) | Increase the panoramic picture conspicuousness detection method with eye movement model based on region | |
JP3740065B2 (en) | Object extraction device and method based on region feature value matching of region-divided video | |
US7400761B2 (en) | Contrast-based image attention analysis framework | |
CN112528969B (en) | Face image authenticity detection method and system, computer equipment and storage medium | |
CN106296638A (en) | Significance information acquisition device and significance information acquisition method | |
CN109754007A (en) | Peplos intelligent measurement and method for early warning and system in operation on prostate | |
Lopez-Molina et al. | Generating fuzzy edge images from gradient magnitudes | |
CN106127782B (en) | A kind of image partition method and system | |
Casanova et al. | Texture analysis using fractal descriptors estimated by the mutual interference of color channels | |
CN112215217B (en) | Digital image recognition method and device for simulating doctor to read film | |
CN113159200B (en) | Object analysis method, device and storage medium | |
CN110766052A (en) | Image display method, evaluation model generation device and electronic equipment | |
JP4506409B2 (en) | Region dividing method and apparatus, image recognition processing apparatus, program, and recording medium | |
CN111563408A (en) | High-resolution image landslide automatic detection method with multi-level perception characteristics and progressive self-learning | |
Guo et al. | Visual complexity assessment of painting images | |
CN111275126A (en) | Sample data set generation method, device, equipment and storage medium | |
CN113516194B (en) | Semi-supervised classification method, device, equipment and storage medium for hyperspectral remote sensing images | |
CN114049503A (en) | Saliency region detection method based on non-end-to-end deep learning network | |
Amelio et al. | An evolutionary approach for image segmentation | |
CN113096080A (en) | Image analysis method and system | |
KR101821770B1 (en) | Techniques for feature extraction | |
Zhang et al. | An Improved Computational Approach for Salient Region Detection. |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190329 |
|
WD01 | Invention patent application deemed withdrawn after publication |