CN108256519A - A kind of notable detection method of infrared image based on global and local interaction - Google Patents
A kind of notable detection method of infrared image based on global and local interaction Download PDFInfo
- Publication number
- CN108256519A CN108256519A CN201711322406.1A CN201711322406A CN108256519A CN 108256519 A CN108256519 A CN 108256519A CN 201711322406 A CN201711322406 A CN 201711322406A CN 108256519 A CN108256519 A CN 108256519A
- Authority
- CN
- China
- Prior art keywords
- formula
- notable
- structural
- global
- represent
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
Present invention is disclosed a kind of notable detection methods of infrared image based on global and local interaction, and including calculating structural local auto-adaptive recursive kernel, structure affine matrix promotes notable figure performance, establish global restriction based on gauss hybrid models, noise jamming is filtered out using structural filtering method, integrate part and global model calculates final notable figure.The present invention can effectively estimate vision well-marked target information, target following and target identification for the later stage improve effective subsequent sections, reduce the search consumption of machine vision algorithm, improve the operational efficiency of algorithm, the operation power consumption of hardware can also be reduced, the resource utilization of picture signal is improved, the visual task for the later stage provides effective image preprocessing support.
Description
Technical field
The present invention relates to the notable detection method of infrared image more particularly to it is a kind of based on global and local interaction it is red
The outer notable detection method of image belongs to the technical field of image understanding processing.
Background technology
Conspicuousness detection understands that analysis method plays an important role in night vision image (including low-light, infrared image), it
Also it plays an important role in machine vision applications.
(X.Hou, L.Zhang, Dynamic the visual attention of document one:searching for coding
length increments,in:D.Koller,D.Schuurmans,Y.Bengio,L.Bottou(Eds.),Advances
In Neural Information Processing Systems 21,2009.) rarity proposition one kind of feature based moves
State vision mode, and the gain of the entropy of each feature is estimated in application (ICL).Document two (T.Liu, Z.Yuan, J.Sun,
J.Wang,N.Zheng,X.Tang,H.-Y.Shum,Learning to detect a salient object,IEEE
TPAMI 33 (2), 2011.) propose a kind of binary conspicuousness detection method by the conditional random field of training, which combines
One group of novel feature, such as multistage contrast, be distributed centered around histogram and color space.But the above method is
It is proposed according to natural image, the less effective applied on infrared image.Document three (C.N.XinWang, L.Xu,
Saliency detection using mutual consistency-guided spatial cues combination,
Infrared Physics＆Technology 72,2015.) utilize the luminance contrast and contour feature of infrared image, estimation
The conspicuousness of infrared image.But this method may lead to the estimation result of mistake, and marking area is made to include ambient noise.
Invention content
Present invention aim to address above-mentioned the deficiencies in the prior art, provide and a kind of interacted based on global and local
The notable detection method of infrared image.
In order to achieve the above object, the technical solution adopted in the present invention is：
A kind of notable detection method of infrared image based on global and local interaction, includes the following steps：
Step 1：Structural local auto-adaptive recursive kernel is calculated,
The characteristic image extracted by input picture I is represented with F, and formula (1) is：
F (x, y)=Γ (I, x, y) (1)
Wherein, Γ () represents a kind of multidimensional function for extracting characteristics of image,
By the use of the position of infrared image, gradient, brightness, LBP and HOG information as image feature, in characteristic image F,
Each pixel can be described as the vector of one 7 dimension, and formula (2) is：
Wherein, (x, y) represents the position of pixel,Represent the marginal information of pixel, HOG () represents that HOG is special
Sign, Lu () represent brightness, and LBP () represents LBP features,
The a certain region R of F in characteristic image can be described as the covariance matrix C of a multidimensionalR, formula (3) is：
Wherein, zi, (i=1 ..., k) represents all characteristic points in the R of region, and μ represents ziAverage value,
The computational methods of structural local auto-adaptive recursive kernel, formula (4) are：
Wherein, l ∈ [1 ..., P2], P2Represent the sum of all pixels in local window, △ X represent window center and surrounding picture
The coordinate relationship of element, s={ x1,x2,z(x1,x2), z (x1,x2) it is pixel (x1,x2) gray value；
Step 2：Affine matrix is built, promotes notable figure performance,
The similitude between the two regions is represented with the distance of the structural local auto-adaptive recursive kernel in two regions, certain
The relevance w of region m inner structure local auto-adaptive recursive kernels and another region n inner structures local auto-adaptive recursive kernelmn, it is public
Formula (5) is：
Wherein, slm,slnThe mean value of the structural local auto-adaptive recursive kernel of region m, n is represented respectively, and Ω (n) represents area
One group of neighborhood of domain n, σ1It is the parameter for controlling similarity degree, MCS () represents cosine similarity matrix,
Then, the affine matrix of a row standardization is constructed, formula (6) is：
A=D-1·W(6)
Wherein, affine matrix W=[wmn]N×NThe similitude being used to represent between any pair of node, angle matrix D=
diag{d1,d2,...,dN, wherein dn=∑nwmnRepresent region n and the degree of association summation of other all areas,
Based on given affine matrix, the aobvious of regional area is defined with description of structural local auto-adaptive recursive kernel
Work property, formula (7) is：
Wherein, amnThe degree of association calculated in representation formula (6), SSLARKIt represents based on structural local auto-adaptive recursive kernel
Low-quality notable figure, N represents the quantity of structural local auto-adaptive recursive kernel；
Step 3：Based on gauss hybrid models, global restriction is established,
First, a global conditions are defined, and minimize its cost, formula (8) is：
Wherein, b1,b2The gauss hybrid models of foreground and background in infrared image are represented respectively,
Description of each structural local auto-adaptive recursive kernel is considered as combining the weight of gauss hybrid models,
It belongs to the probability of neighborhood, and formula (9) is：
Wherein, PnA kind of linear operation of n-th of structural local recursion's core region, w are extracted in expression from infrared imagemn
It is calculated by formula (5), ΣmIt is covariance matrix, Φ represents Gaussian Profile；
Step 4：Using structural filtering method, noise jamming is filtered out,
Based on structural local auto-adaptive recursive kernel, a kind of filtering method is designed, is wrapped with further filtering out in Gauss model
The ambient noise contained, formula (10) are：
Wherein, SG(x2) represent SGMiddle pixel x2Conspicuousness numerical value,It is normalization factor, R (x2) table
Show with x1For the pixel in the neighborhood in the center of circle,
Step 5：Part and global model are integrated, calculates final notable figure,
Formula (11) is：
S*=α S1+(1-α)S2 (11)
Wherein, α is balance factor, S1Represent local notable figure, S2Represent global notable figure.
The beneficial effects are mainly as follows：
Using region covariance can effectively well-marked target structural information, can effectively distinguish the difference of background and target；
Consider global and local information, can effectively excavate well-marked target information, and pass through an effective conformable frame and count again
Vision significance is calculated, improves the accuracy of conspicuousness detection.While global notable figure is calculated, Gaussian Mixture mould is utilized
Type establishes global clue constraint, the interference of noise is reduced by structural filtering.This method can effectively estimate that vision is shown
Target information is write, is that the target following in later stage and target identification improve effective subsequent sections, reduces searching for machine vision algorithm
Rope consumes, and improves the operational efficiency of algorithm, can also reduce the operation power consumption of hardware, improves the utilization of resources of picture signal
Rate, the visual task for the later stage provide effective image preprocessing support.
Description of the drawings
Fig. 1 is the schematic diagram of the notable detection method of infrared image of the present invention.
Fig. 2 is the local notable figure that step 2 of the present invention is obtained.
Fig. 3 is in step 2 of the present inventionTake local notable figure during different value.
Fig. 4 is the notable figure of world model of the present invention.
Fig. 5 is in step 4 of the present inventionTake local notable figure during different value.
Fig. 6 is the conspicuousness testing result figure of the present invention.
Specific embodiment
The present invention provides a kind of notable detection method of infrared image based on global and local interaction.Below in conjunction with attached
Technical solution of the present invention is described in detail in figure, so that it is more readily understood and grasps.
A kind of notable detection method of infrared image based on global and local interaction, as shown in Figure 1, including following step
Suddenly：
Step 1：Structural local auto-adaptive recursive kernel is calculated,
The characteristic image extracted by input picture I is represented with F, and formula (1) is：
F (x, y)=Γ (I, x, y) (1)
Wherein, Γ () represents a kind of multidimensional function for extracting characteristics of image,
By the use of the position of infrared image, gradient, brightness, LBP and HOG information as image feature, in characteristic image F,
Each pixel can be described as the vector of one 7 dimension, and formula (2) is：
Wherein, (x, y) represents the position of pixel,Represent the marginal information of pixel, HOG () represents that HOG is special
Sign, Lu () represent brightness, and LBP () represents LBP features,
The a certain region R of F in characteristic image can be described as the covariance matrix C of a multidimensionalR, formula (3) is：
Wherein, zi, (i=1 ..., k) represents all characteristic points in the R of region, and μ represents ziAverage value,
The computational methods of structural local auto-adaptive recursive kernel, formula (4) are：
Wherein, l ∈ [1 ..., P2], P2Represent the sum of all pixels in local window, △ X represent window center and surrounding picture
The coordinate relationship of element, s={ x1,x2,z(x1,x2), z (x1,x2) it is pixel (x1,x2) gray value；
Step 2：Affine matrix is built, promotes notable figure performance,
The similitude between the two regions is represented with the distance of the structural local auto-adaptive recursive kernel in two regions, certain
The relevance w of region m inner structure local auto-adaptive recursive kernels and another region n inner structures local auto-adaptive recursive kernelmn, it is public
Formula (5) is：
Wherein, slm,slnThe mean value of the structural local auto-adaptive recursive kernel of region m, n is represented respectively, and Ω (n) represents area
One group of neighborhood of domain n, σ1It is the parameter for controlling similarity degree, MCS () represents cosine similarity matrix,
Then, the affine matrix of a row standardization is constructed, formula (6) is：
A=D-1·W (6)
Wherein, affine matrix W=[wmn]N×NThe similitude being used to represent between any pair of node, angle matrix D=
diag{d1,d2,...,dN, wherein dn=∑nwmnRepresent region n and the degree of association summation of other all areas,
Based on given affine matrix, the aobvious of regional area is defined with description of structural local auto-adaptive recursive kernel
Work property, formula (7) is：
Wherein, amnThe degree of association calculated in representation formula (6), SSLARKIt represents based on structural local auto-adaptive recursive kernel
Low-quality notable figure, N represents the quantity of structural local auto-adaptive recursive kernel；
Step 3：Based on gauss hybrid models, global restriction is established,
First, a global conditions are defined, and minimize its cost, formula (8) is：
Wherein, b1,b2The gauss hybrid models of foreground and background in infrared image are represented respectively,
Description of each structural local auto-adaptive recursive kernel is considered as combining the weight of gauss hybrid models,
It belongs to the probability of neighborhood, and formula (9) is：
Wherein, PnA kind of linear operation of n-th of structural local recursion's core region, w are extracted in expression from infrared imagemn
It is calculated by formula (5), ΣmIt is covariance matrix, Φ represents Gaussian Profile；
Step 4：Using structural filtering method, noise jamming is filtered out,
Based on structural local auto-adaptive recursive kernel, a kind of filtering method is designed, is wrapped with further filtering out in Gauss model
The ambient noise contained, formula (10) are：
Wherein, SG(x2) represent SGMiddle pixel x2Conspicuousness numerical value,It is normalization factor, R (x2) table
Show with x1For the pixel in the neighborhood in the center of circle,
Step 5：Part and global model are integrated, calculates final notable figure,
Formula (11) is：
S*=α S1+(1-α)S2 (11)
Wherein, α is balance factor, S1Represent local notable figure, S2Represent global notable figure.
As shown in Fig. 2, the local notable figure obtained by step 2 in the present invention, Fig. 2 (a) is input picture, Fig. 2 (b)
For local notable figure, Fig. 2 (c) is to add in the local notable figure that affine matrix calculates.The influence of affine matrix is increased, is made significantly
Figure is more accurate, and ambient noise is inhibited well.
As shown in figure 3, for formula (5) in step 2Take local notable figure during different value.Wherein, in Fig. 3 (a)In Fig. 3 (b)In Fig. 3 (c)In Fig. 3 (d)
As shown in figure 4, the notable figure of world model.Fig. 4 (a) is input picture, and Fig. 4 (b) is global notable figure, Fig. 4 (c)
The global notable figure calculated to add in structural filtering.The processing of structural filtering is have passed through, both highlights target, is also inhibited
Ambient noise.
As shown in figure 5, for formula (10) in step 4Take local notable figure during different value.In Fig. 5 (a)
In Fig. 5 (b)In Fig. 5 (c)In Fig. 5 (d)
As shown in fig. 6, the conspicuousness testing result figure for invention.
By above description it can be found that a kind of infrared image based on global and local interaction of the present invention is significantly examined
Survey method can effectively estimate vision well-marked target information, after the target following and target identification for the later stage improve effectively
The search consumption of machine vision algorithm is reduced in continuous region, improves the operational efficiency of algorithm, can also reduce the operation work(of hardware
Consumption, improves the resource utilization of picture signal, and the visual task for the later stage provides effective image preprocessing support.
Technical scheme of the present invention is fully described above, it should be noted that specific embodiment party of the invention
Formula is simultaneously not limited by the description set out above, the Spirit Essence of those of ordinary skill in the art according to the present invention structure, method or
All technical solutions that function etc. is formed using equivalents or equivalent transformation, all fall within protection scope of the present invention
Within.
Claims (1)
1. a kind of notable detection method of infrared image based on global and local interaction, includes the following steps：
Step 1：Structural local auto-adaptive recursive kernel is calculated,
The characteristic image extracted by input picture I is represented with F, and formula (1) is：
F (x, y)=Γ (I, x, y) (1)
Wherein, Γ () represents a kind of multidimensional function for extracting characteristics of image,
By the use of the position of infrared image, gradient, brightness, LBP and HOG information as image feature, in characteristic image F, each
Pixel can be described as the vector of one 7 dimension, and formula (2) is：
Wherein, (x, y) represents the position of pixel,Represent the marginal information of pixel, HOG () represents HOG features, Lu
() represents brightness, and LBP () represents LBP features,
The a certain region R of F in characteristic image can be described as the covariance matrix C of a multidimensionalR, formula (3) is：
Wherein, zi, (i=1 ..., k) represents all characteristic points in the R of region, and μ represents ziAverage value,
The computational methods of structural local auto-adaptive recursive kernel, formula (4) are：
Wherein, l ∈ [1 ..., P2], P2Sum of all pixels in expression local window, △ X expression window centers and surrounding pixel
Coordinate relationship, s={ x1,x2,z(x1,x2), z (x1,x2) it is pixel (x1,x2) gray value；
Step 2：Affine matrix is built, promotes notable figure performance,
The similitude between the two regions, certain region are represented with the distance of the structural local auto-adaptive recursive kernel in two regions
The relevance w of m inner structure local auto-adaptive recursive kernels and another region n inner structures local auto-adaptive recursive kernelmn, formula
(5) it is：
Wherein, slm,slnThe mean value of the structural local auto-adaptive recursive kernel of region m, n is represented respectively, and Ω (n) represents region n's
One group of neighborhood, σ1It is the parameter for controlling similarity degree, MCS () represents cosine similarity matrix,
Then, the affine matrix of a row standardization is constructed, formula (6) is：
A=D-1·W (6)
Wherein, affine matrix W=[wmn]N×NThe similitude being used to represent between any pair of node, angle matrix D=diag
{d1,d2,...,dN, wherein dn=∑nwmnRepresent region n and the degree of association summation of other all areas,
Based on given affine matrix, the notable of regional area is defined with description of structural local auto-adaptive recursive kernel
Property, formula (7) is：
Wherein, amnThe degree of association calculated in representation formula (6), SSLARKRepresent the low-quality based on structural local auto-adaptive recursive kernel
The notable figure of amount, N represent the quantity of structural local auto-adaptive recursive kernel；
Step 3：Based on gauss hybrid models, global restriction is established,
First, a global conditions are defined, and minimize its cost, formula (8) is：
Wherein, b1,b2The gauss hybrid models of foreground and background in infrared image are represented respectively,
Description of each structural local auto-adaptive recursive kernel is considered as combining the weight of gauss hybrid models, belongs to
In the probability of neighborhood, formula (9) is：
Wherein, PnA kind of linear operation of n-th of structural local recursion's core region, w are extracted in expression from infrared imagemnBy public affairs
Formula (5) is calculated, ΣmIt is covariance matrix, Φ represents Gaussian Profile；
Step 4：Using structural filtering method, noise jamming is filtered out,
Based on structural local auto-adaptive recursive kernel, a kind of filtering method is designed, to further filter out what is included in Gauss model
Ambient noise, formula (10) are：
Wherein, SG(x2) represent SGMiddle pixel x2Conspicuousness numerical value,It is normalization factor, R (x2) represent with
x1For the pixel in the neighborhood in the center of circle,
Step 5：Part and global model are integrated, calculates final notable figure,
Formula (11) is：
S*=α S1+(1-α)S2 (11)
Wherein, α is balance factor, S1Represent local notable figure, S2Represent global notable figure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711322406.1A CN108256519B (en) | 2017-12-13 | 2017-12-13 | Infrared image significance detection method based on global and local interaction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711322406.1A CN108256519B (en) | 2017-12-13 | 2017-12-13 | Infrared image significance detection method based on global and local interaction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108256519A true CN108256519A (en) | 2018-07-06 |
| CN108256519B CN108256519B (en) | 2022-06-17 |
Family
ID=62722584
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711322406.1A Active CN108256519B (en) | 2017-12-13 | 2017-12-13 | Infrared image significance detection method based on global and local interaction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108256519B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113822352A (en) * | 2021-09-15 | 2021-12-21 | 中北大学 | Infrared dim target detection method based on multi-feature fusion |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104599291A (en) * | 2015-01-21 | 2015-05-06 | 内蒙古科技大学 | Structural similarity and significance analysis based infrared motion target detection method |
| CN106295542A (en) * | 2016-08-03 | 2017-01-04 | 江苏大学 | A kind of road target extracting method of based on significance in night vision infrared image |
| CN106846331A (en) * | 2016-12-22 | 2017-06-13 | 中国科学院文献情报中心 | Joint vision significance and figure cut the image automatic segmentation method of optimization |
| CN107240096A (en) * | 2017-06-01 | 2017-10-10 | 陕西学前师范学院 | A kind of infrared and visual image fusion quality evaluating method |
-
2017
- 2017-12-13 CN CN201711322406.1A patent/CN108256519B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104599291A (en) * | 2015-01-21 | 2015-05-06 | 内蒙古科技大学 | Structural similarity and significance analysis based infrared motion target detection method |
| CN106295542A (en) * | 2016-08-03 | 2017-01-04 | 江苏大学 | A kind of road target extracting method of based on significance in night vision infrared image |
| CN106846331A (en) * | 2016-12-22 | 2017-06-13 | 中国科学院文献情报中心 | Joint vision significance and figure cut the image automatic segmentation method of optimization |
| CN107240096A (en) * | 2017-06-01 | 2017-10-10 | 陕西学前师范学院 | A kind of infrared and visual image fusion quality evaluating method |
Non-Patent Citations (1)
| Title |
|---|
| QIANG FAN 等: "《Saliency detection based on global and local short-term》", 《NEUROCOMPUTING》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113822352A (en) * | 2021-09-15 | 2021-12-21 | 中北大学 | Infrared dim target detection method based on multi-feature fusion |
| CN113822352B (en) * | 2021-09-15 | 2024-05-17 | 中北大学 | Infrared dim target detection method based on multi-feature fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108256519B (en) | 2022-06-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yeh et al. | Lightweight deep neural network for joint learning of underwater object detection and color conversion | |
| Xie et al. | Visual saliency detection based on Bayesian model | |
| CN108256562B (en) | Salient target detection method and system based on weak supervision time-space cascade neural network | |
| Chen et al. | Efficient hierarchical method for background subtraction | |
| CN103020992B (en) | A kind of video image conspicuousness detection method based on motion color-associations | |
| CN111639564B (en) | Video pedestrian re-identification method based on multi-attention heterogeneous network | |
| CN112819858B (en) | Target tracking method, device, equipment and storage medium based on video enhancement | |
| CN110135435B (en) | Saliency detection method and device based on breadth learning system | |
| CN104715476A (en) | Salient object detection method based on histogram power function fitting | |
| Xu et al. | Extended non-local feature for visual saliency detection in low contrast images | |
| Roy et al. | A comprehensive survey on computer vision based approaches for moving object detection | |
| Liu et al. | A shadow imaging bilinear model and three-branch residual network for shadow removal | |
| Li et al. | Gaussian-based codebook model for video background subtraction | |
| CN108256519A (en) | A kind of notable detection method of infrared image based on global and local interaction | |
| Zhou et al. | Dynamic background subtraction using spatial-color binary patterns | |
| Ouzounis et al. | Interactive collection of training samples from the max-tree structure | |
| Schulz et al. | Object-class segmentation using deep convolutional neural networks | |
| Chowdhury et al. | A background subtraction method using color information in the frame averaging process | |
| CN116310758A (en) | Indoor scene recognition method and system based on combined semantic region relation model | |
| You et al. | Db-net: dual attention network with bilinear pooling for fire-smoke image classification | |
| CN107016675A (en) | A kind of unsupervised methods of video segmentation learnt based on non local space-time characteristic | |
| Wang et al. | A brief survey of low-level saliency detection | |
| CN109033969B (en) | Infrared target detection method based on Bayesian saliency map calculation model | |
| Lin et al. | Infrared small target detection based on YOLO v4 | |
| Chengjun et al. | Background extraction and update method based on histogram in ycbcr color space |
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 |