CN106709512A - Infrared target detection method based on local sparse representation and contrast - Google Patents
Infrared target detection method based on local sparse representation and contrast Download PDFInfo
- Publication number
- CN106709512A CN106709512A CN201611127333.6A CN201611127333A CN106709512A CN 106709512 A CN106709512 A CN 106709512A CN 201611127333 A CN201611127333 A CN 201611127333A CN 106709512 A CN106709512 A CN 106709512A
- Authority
- CN
- China
- Prior art keywords
- image
- image block
- target
- residual
- region
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The invention discloses an infrared target detection method based on local sparse representation and contrast. The infrared target detection method comprises the following steps of 1, firstly, setting the original infrared image; on the basis of local sparse representation concept, designing a method based on an inner product, and building a non-similarity dictionary of each image block; according to the dictionary, calculating a residual image block of each image block; finally, calculating the residual image of the whole image; 2, predicting a candidate target area in the original infrared image by a gray scale contrast method, so as to obtain a target predicting image; 3, integrating the residual image and the target predicting image, precisely positioning the target, and performing binarization segmentation, so as to obtain the final target detection result. The infrared target detection method has the advantages that the residual image is calculated by the local sparse representation and inner product methods, and the accuracy in target detection is improved; the target predicting image is obtained by the gray scale contrast method, and the background noise is effectively inhibited; therefore, the robustness and precision in target detection are improved.
Description
Technical field
Infrared target detection method the present invention relates to be based on local rarefaction representation and contrast, it is more particularly to a kind of multiple
Infrared image object detection method under miscellaneous background clutter disturbed condition, belongs to technical field of computer vision.
Background technology
Along with the development of infrared imagery technique, infrared target detection is not only applicable to military field, also in industry, agriculture
The civil areas such as industry, medical science, traffic have a wide range of applications.Infrared imagery technique has good concealment, strong antijamming capability, fits
The advantages of answering environment capacity strong.There is challenge however, carrying out accurate infrared target detection under complex background and being still one
The problem of property because in complex background clutter and noise interference, cause that infrared target detection accuracy is high, robustness is not strong
The problems such as.
In recent years, sparse representation theory is applied to mesh by continuing to develop with sparse representation theory, many researchers
Mark detection field.
Publication number CN103440502A is based on the infrared small target detection method of mixed Gaussian rarefaction representation, and the method is used
K cluster singular value decomposition method K_SVD self adaptations build the super complete form dictionary of image, and the dictionary passes through the super complete dictionary of Gauss
Formed with the super complete dictionary of the ADAPTIVE MIXED Gauss of target morphology dictionary and background form dictionary, and ask for original image block and existed
Rarefaction representation coefficient under the dictionary, target detection is realized by comparing the relation between rarefaction representation coefficient and threshold value.But
It is that the algorithm needs target prior information, universality is not strong.
Publication number CN102842047A is based on the method for detecting infrared puniness target of multiple dimensioned sparse dictionary, and it is using original
Image configuration is based on the excessively complete dictionary of multi-scale self-adaptive of Quadtree, and on this basis to each original image sub-block
Rarefaction representation is carried out, sparse coefficient is obtained, these sparse systems is carried out combining row index fitting of distribution of going forward side by side, by threshold decision
The accurate location of target is determined with the multiple dimensioned directionality of sparse coefficient.But, the method is in the stronger situation of noise jamming
Under, target detection performance has declined.
The content of the invention
The technical problems to be solved by the invention are：Infrared target detection based on local rarefaction representation and contrast is provided
Method, the influence that background clutter and noise jamming can be overcome to be brought greatly improves target detection rate, reduces target false-alarm
Rate.
The present invention uses following technical scheme to solve above-mentioned technical problem：
Infrared target detection method based on local rarefaction representation and contrast, comprises the following steps：
Step 1, some image blocks are divided into by original infrared image, and each is built using local rarefaction representation and inner product approach
The non-similarity dictionary of image block, and based on non-similarity dictionary, the residual image block of each image block is calculated, according to residual plot
The residual image of original infrared image is obtained as block；
Step 2, for original infrared image, using the candidate target region in the method prognostic chart picture of grey-scale contrast,
By the gray value zero setting of non-candidate target area pixel in image, target prediction image is obtained；
Step 3, the target prediction image that the residual image and step 2 that step 1 is obtained are obtained is blended, and obtains target pre-
Residual image is surveyed, according to determination of target prediction residual error framing target area, and binarization segmentation is carried out to target area is obtained most
Whole object detection results.
As a preferred embodiment of the present invention, original infrared image is divided into some image blocks described in step 1, using office
Portion's rarefaction representation and inner product approach build the non-similarity dictionary of each image block, and detailed process is as follows：
Step 11, utilizesThe sliding window of size is by original infrared image Y by from top to bottom, from left to right
Mode travel through, obtain m size identical image blockK=1,2 ..., m, BsIt is image block YkSize；
Step 12, for each image block, expands the training window image of L × L sizes, using sliding window centered on it
MouthfulWill training window image by from top to bottom, from left to right in the way of travel through, obtain h size identical image blockG=1,2 ..., h, by each image block fgColumn vector is carried out, matrix is constituted
Step 13, calculates image block YkRelative to the inner product of each column vector in matrix F：ρg=<Yk,fg>, wherein, ρgTable
Show YkRelative to g-th inner product of column vector in matrix F,<·,·>Represent inner product operation；
Step 14, calculates image block YkWith the inner product β of its own：β=<Yk,Yk>；
Step 15, calculates β and ρgAbsolute difference And it is exhausted therefrom to find out c maximum difference
To value, c image block in c maximum absolute difference correspondence training window image will be arranged after c image block column vector
Into a matrix, image block Y is obtainedkNon-similarity dictionary
As a preferred embodiment of the present invention, non-similarity dictionary is based on described in step 1, calculates the residual of each image block
Difference image block, the residual image of original infrared image is obtained according to residual image block, and detailed process is as follows：
Step 16, image block Y is calculated using OMP algorithmskSparse coding coefficient X under non-similarity dictionaryk；
Step 17, calculates image block YkReconstructive residual error vector εk：εk=Yk-DkXk；
Step 18, reconstructive residual error vector is reduced toThe matrix of size, BsIt is image block YkSize, obtain
Image block YkResidual image block；
Step 19, to each image block obtained through step 11, all carries out the operation of step 12 to step 18, and will be all
The residual image block for obtaining generates new residual image according to its position of the corresponding image block in original infrared image.
Used as a preferred embodiment of the present invention, the detailed process of the step 2 is as follows：
Step 21, utilizesThe sliding window of size presses from top to bottom, from left to right original infrared image Y
Mode is traveled through, and obtains T size same levels image block Yt∈RU, t=1,2 ..., T, U are image block YtSize；
Step 22, by each image block YtAs region to be detected, will be with image block YtCentered on,Size
Search window in remove YtWhether other 8 image blocks judge region to be detected as background area according to below equation in addition
It is candidate target region：
Wherein, (x, y) represents pixel coordinate in region to be detected, MtIt is the average of grey scale pixel value in region to be detected, Mb
It is the average of grey scale pixel value in background area, StdbIt is the standard deviation of grey scale pixel value in background area, λ is parameter；
Step 23, when R (x, y)=1, represents that region to be detected is candidate target region, and the gray value in the region is protected
Stay, when R (x, y)=0, represent that region to be detected is not candidate target region, by the gray value zero setting in the region, obtain target
Prognostic chart picture.
Used as a preferred embodiment of the present invention, the detailed process of the step 3 is as follows：
Step 31, the gray value of respective regions in the gray value residual image of non-zero region in target prediction image is replaced
Change, obtain determination of target prediction residual error image；
Step 32, the grey scale pixel value average of the non-zero region of each non-interconnected in comparison object prediction residual image,
The maximum non-zero region of value is target area；
Step 33, target area is mapped in original infrared image, while to the non-target area in original infrared image
Domain zero setting, binarization segmentation is carried out to target area, obtains final target detection result.
The present invention uses above technical scheme compared with prior art, with following technique effect：
1st, method construct of the detection method based on local rarefaction representation and inner product has gone out the residual plot of original image
Picture, improves the accuracy of target detection.
2nd, detection method excavates the difference between target gray and background gray scale to predict target based on contrast
Region that may be present, forms target prediction image, can effectively suppress ambient noise.
3rd, detection method is merged residual image with target prediction image, not only increases target detection
Robustness, and improve the accuracy of target detection.
Brief description of the drawings
Fig. 1 is the implementing procedure figure of infrared target detection method of the present invention based on local rarefaction representation and contrast.
Fig. 2 is the non-similarity dictionary organigram of image block in detection method.
Fig. 3 is the search window schematic diagram for local detection in detection method.
Fig. 4 is the testing result figure of detection method embodiment.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the implementation method is shown in the drawings.Below by
The implementation method being described with reference to the drawings is exemplary, is only used for explaining the present invention, and is not construed as limiting the claims.
As shown in figure 1, infrared target detection method specific steps of the present invention based on local rarefaction representation and contrast are such as
Under.
Step one, on the basis of local rarefaction representation thought, designs a kind of method based on inner product, searches out each figure
As some image subblocks most unmatched therewith in block neighborhood, to build the non-similarity dictionary of the image block；Then, based on this
Dictionary, calculates the residual image block of each image block；Finally, the residual image of entire image is calculated, its specific operation process is such as
Under：
(1) acquisition of non-similarity dictionary
First, input size is the original infrared image Y of M × N, while generating a residual plot image set for skyWith one as original image Y sizes and element all for 0 residual image G=[0,0 ..., 0] ∈ RM ×N.Then, as shown in Fig. 2 use sliding window by image Y be divided into m size forThe image block that can be overlappedWherein k=1,2 ..., m, BsIt is the size of image block, image block YkBe exactly in Fig. 2 small square mark what is represented
Image block.
Secondly, as shown in Fig. 2 for each above-mentioned image block, the training window of L × L sizes is expanded centered on it
(being represented with big square), will train the image in window to be also divided into h size and beAnd the image block for overlapping each otherWherein g=1,2 ..., h, these blocks constitute setTo setIn each image block enter ranks to
Quantify, finally constitute matrix
3rd, calculate image block YkRelative to the inner product of all column vectors in matrix F, i.e.,：
ρg=<Yk,fg>
Wherein, ρgRepresent YkRelative to g-th inner product of column vector in matrix F,<·,·>Represent inner product operation.Therefore to square
Battle array F, Y is can be obtained by by above-mentioned computingkRelative to the inner product set ρ=[ρ of all column vectors in matrix F1,ρ2,…,ρh]。
4th, calculate YkWith the inner product of its own, i.e.,：
β=<Yk,Yk>
Wherein, β is YkThe result of inner product is carried out with its own.
Finally, for setFor, with YkMore similar image block, corresponding to the vector f in matrixgValue will be with
YkIt is more close；More dissimilar block, corresponding to the vector f in matrixgValue will be with YkThere is larger difference.It is former based on this
Reason, it is possible to distinguish set by the following methodSimilar block and dissimilar blocks, i.e.,：
Wherein,The absolute difference of g-th element in β and ρ is represented,Value it is bigger, represent setIn
G-th image block fgWith YkMore dissimilar, the difference that can be obtained by all elements in β and inner product set ρ by above-mentioned formula is exhausted
To value, the maximum image block of c absolute difference is taken, constitute setJust search out c individual least similar to original picture block
Image block.After by these image block column vectors, a matrix is arranged in, just constitutes image block YkNon-similarity dictionaryEach of which image block is all expressed as a row of matrix, the i.e. atom of dictionary.
(2) generation of the residual image block based on non-similarity dictionary
Specific obtaining step for the residual image block of each image block is as follows：
First, using OMP algorithms, image block Y is calculatedkIn dictionary DkUnder sparse coding coefficient Xk。
Then, image block Y is calculated by equation belowkReconstructive residual error vector εk：
εk=Yk-DkXk
Finally, reconstructive residual error vector is reduced intoThe matrix of size, is designated as residual image block repatch.Will
It is deposited into residual plot image set repatchset, i.e.,：
Repatchset=[repatchset, repatch]
(3) residual image is ultimately generated
To a series of image block in image Y, the operation of (1) and (2) is all carried out, and by residual plot image set repatchset
In each residual image block according to its position of the corresponding original picture block in image Y, be correspondingly put into residual image G
(lap is averaged), finally can be obtained by residual image G.
Step 2, for original image, will also be using the thought of grey-scale contrast come the candidate target area in prognostic chart picture
Domain, forms target prediction image.Comprise the following steps that：
(4) original infrared image Y is input into, is usedWindow original image is carried out from top to bottom, from left to right
Scanning, step-length is step, original image can be divided into a series of image blocks for overlapping each other.
(5) candidate target region in original image can be detected using search window as shown in Figure 3.In figure, should
Window is divided into nine size identical cells, and middle cell is represented with " 9 ", region referred to as to be detected, and representing target can
The region that can occur, other cells represent background area.Each original picture block in step (4) is regarded as one by one
Region to be detected, finds the search window of its corresponding local detection as shown in Figure 3, it is then possible to logical in original image Y
Cross equation below and filter out candidate target：
Wherein, MtIt is the average in region to be detected, MbIt is the average of background area, StdbIt is the standard deviation of background area, R
(x, y) is candidate target testing result, and λ represents parameter, general value 0.5-3.It is to be herein pointed out working as R (x, y)=1
When, representing that the region is candidate target region, we retain the numerical value in the region.When R (x, y)=0, the region is represented not
It is candidate target region, we are by the numerical value zero setting in the region.The only image containing candidate target is thus generated, referred to herein as
Target prediction image P.
Step 3, the above-mentioned residual image being calculated and target prediction image are blended, accurately fixed to carry out target
Position, and carry out binarization segmentation and obtain final target detection result.Comprise the following steps that：
(6) target prediction image P and residual image G are merged, i.e., to the non-zero region in target prediction image
Numerical value is converted to the numerical value of respective regions in residual image, thus generates determination of target prediction residual error image S.
(7) detected for single goal, it is contemplated that the target gray value in determination of target prediction residual error image S is larger, it is other empty
Alert target gray value is smaller, so need to only carry out average ratio to the non-zero region of some non-interconnected in image S compared with mean value computation is public
Formula is as follows：
Wherein, piThe gray value of ith pixel in non-zero region is represented, n represents the element number in the region, and mean is then
Represent the average of the non-zero region.The maximum non-zero region of average is exactly target area.Then the target area is mapped to original
In beginning image, while doing zero setting treatment to the nontarget area in original image.
(8) binaryzation is carried out to target area, it is possible to obtain final target detection result.As shown in figure 4, being the present invention
The testing result figure of embodiment of the method.
Above example is only explanation technological thought of the invention, it is impossible to limit protection scope of the present invention with this, every
According to technological thought proposed by the present invention, any change done on the basis of technical scheme each falls within the scope of the present invention
Within.
Claims (5)
1. the infrared target detection method of local rarefaction representation and contrast is based on, it is characterised in that comprised the following steps：
Step 1, some image blocks are divided into by original infrared image, and each image is built using local rarefaction representation and inner product approach
The non-similarity dictionary of block, and based on non-similarity dictionary, the residual image block of each image block is calculated, according to residual image block
Obtain the residual image of original infrared image；
Step 2, for original infrared image, using the candidate target region in the method prognostic chart picture of grey-scale contrast, will scheme
The gray value zero setting of non-candidate target area pixel, obtains target prediction image as in；
Step 3, the target prediction image that the residual image and step 2 that step 1 is obtained are obtained is blended, and obtains target prediction residual
Difference image, according to determination of target prediction residual error framing target area, and carries out binarization segmentation and obtains final mesh to target area
Mark testing result.
2. the infrared target detection method of local rarefaction representation and contrast is based on according to claim 1, it is characterised in that
Original infrared image is divided into some image blocks described in step 1, each image is built using local rarefaction representation and inner product approach
The non-similarity dictionary of block, detailed process is as follows：
Step 11, utilizesThe sliding window of size by original infrared image Y by from top to bottom, from left to right in the way of
Traversal, obtains m size identical image blockK=1,2 ..., m, BsIt is image block YkSize；
Step 12, for each image block, expands the training window image of L × L sizes, using sliding window centered on itWill training window image by from top to bottom, from left to right in the way of travel through, obtain h size identical image blockG=1,2 ..., h, by each image block fgColumn vector is carried out, matrix is constituted
Step 13, calculates image block YkRelative to the inner product of each column vector in matrix F：ρg=<Yk,fg>, wherein, ρgRepresent Yk
Relative to g-th inner product of column vector in matrix F,<·,·>Represent inner product operation；
Step 14, calculates image block YkWith the inner product β of its own：β=<Yk,Yk>；
Step 15, calculates β and ρgAbsolute difference And therefrom find out c maximum absolute difference,
C image block in c maximum absolute difference correspondence training window image, will be arranged in one after c image block column vector
Individual matrix, obtains image block YkNon-similarity dictionary
3. the infrared target detection method of local rarefaction representation and contrast is based on according to claim 2, it is characterised in that
Non-similarity dictionary is based on described in step 1, the residual image block of each image block is calculated, obtains original red according to residual image block
The residual image of outer image, detailed process is as follows：
Step 16, image block Y is calculated using OMP algorithmskSparse coding coefficient X under non-similarity dictionaryk；
Step 17, calculates image block YkReconstructive residual error vector εk：εk=Yk-DkXk；
Step 18, reconstructive residual error vector is reduced toThe matrix of size, BsIt is image block YkSize, obtain image
Block YkResidual image block；
Step 19, to each image block obtained through step 11, all carries out the operation of step 12 to step 18, and obtain all
Residual image block according to its position of the corresponding image block in original infrared image, generate new residual image.
4. the infrared target detection method of local rarefaction representation and contrast is based on according to claim 1, it is characterised in that
The detailed process of the step 2 is as follows：
Step 21, utilizesThe sliding window of size by original infrared image Y by from top to bottom, from left to right in the way of
Traversal, obtains T size same levels image block Yt∈RU, t=1,2 ..., T, U are image block YtSize；
Step 22, by each image block YtAs region to be detected, will be with image block YtCentered on,Size is searched
Rope window is intraoral except YtOther 8 image blocks judge whether region to be detected is time as background area according to below equation in addition
Select target area：
Wherein, (x, y) represents pixel coordinate in region to be detected, MtIt is the average of grey scale pixel value in region to be detected, MbFor
The average of grey scale pixel value, Std in background areabIt is the standard deviation of grey scale pixel value in background area, λ is parameter；
Step 23, when R (x, y)=1, represents that region to be detected is candidate target region, and the gray value in the region is retained, when
During R (x, y)=0, represent that region to be detected is not candidate target region, by the gray value zero setting in the region, obtain target prediction
Image.
5. the infrared target detection method of local rarefaction representation and contrast is based on according to claim 1, it is characterised in that
The detailed process of the step 3 is as follows：
Step 31, the gray value of respective regions in the gray value residual image of non-zero region in target prediction image is replaced,
Obtain determination of target prediction residual error image；
Step 32, the grey scale pixel value average of the non-zero region of each non-interconnected in comparison object prediction residual image, average is most
Big non-zero region is target area；
Step 33, target area is mapped in original infrared image, while being put to the nontarget area in original infrared image
Zero, binarization segmentation is carried out to target area, obtain final target detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611127333.6A CN106709512B (en) | 2016-12-09 | 2016-12-09 | Infrared target detection method based on local sparse representation and contrast |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611127333.6A CN106709512B (en) | 2016-12-09 | 2016-12-09 | Infrared target detection method based on local sparse representation and contrast |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106709512A true CN106709512A (en) | 2017-05-24 |
| CN106709512B CN106709512B (en) | 2020-03-17 |
Family
ID=58936674
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611127333.6A Active CN106709512B (en) | 2016-12-09 | 2016-12-09 | Infrared target detection method based on local sparse representation and contrast |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709512B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107452230A (en) * | 2017-07-28 | 2017-12-08 | 明见（厦门）技术有限公司 | A kind of obstacle detection method, device, terminal device and storage medium |
| CN108062523A (en) * | 2017-12-13 | 2018-05-22 | 苏州长风航空电子有限公司 | A kind of infrared remote small target detecting method |
| CN108898589A (en) * | 2018-06-19 | 2018-11-27 | 南通大学 | The quick-fried pearl intelligent detecting method of filter stick based on high speed machines vision |
| CN112749714A (en) * | 2019-10-29 | 2021-05-04 | 中国科学院长春光学精密机械与物理研究所 | Method for detecting polymorphic dark and weak small target in single-frame infrared image |
| CN114463619A (en) * | 2022-04-12 | 2022-05-10 | 西北工业大学 | Infrared dim target detection method based on integrated fusion features |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090016587A1 (en) * | 2007-07-09 | 2009-01-15 | Siemens Corporate Research, Inc. | System and method for two-dimensional visualization of temporal phenomena and three dimensional vessel reconstruction |
| US7894668B1 (en) * | 2006-09-28 | 2011-02-22 | Fonar Corporation | System and method for digital image intensity correction |
| CN103632153A (en) * | 2013-12-05 | 2014-03-12 | 宁波大学 | Region-based image saliency map extracting method |
| CN103996040A (en) * | 2014-05-13 | 2014-08-20 | 西北工业大学 | Bottom-up visual saliency generating method fusing local-global contrast ratio |
| CN104021537A (en) * | 2014-06-23 | 2014-09-03 | 西北工业大学 | Infrared and visible image fusion method based on sparse representation |
| EP2797033A1 (en) * | 2013-04-22 | 2014-10-29 | Ricoh Company, Ltd. | Method and apparatus for processing sparse disparity map, and method and apparatus for detecting object |
| US20160239969A1 (en) * | 2015-02-14 | 2016-08-18 | The Trustees Of The University Of Pennsylvania | Methods, systems, and computer readable media for automated detection of abnormalities in medical images |
| CN106296567A (en) * | 2015-05-25 | 2017-01-04 | 北京大学 | The conversion method of a kind of multi-level image style based on rarefaction representation and device |
-
2016
- 2016-12-09 CN CN201611127333.6A patent/CN106709512B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7894668B1 (en) * | 2006-09-28 | 2011-02-22 | Fonar Corporation | System and method for digital image intensity correction |
| US20090016587A1 (en) * | 2007-07-09 | 2009-01-15 | Siemens Corporate Research, Inc. | System and method for two-dimensional visualization of temporal phenomena and three dimensional vessel reconstruction |
| EP2797033A1 (en) * | 2013-04-22 | 2014-10-29 | Ricoh Company, Ltd. | Method and apparatus for processing sparse disparity map, and method and apparatus for detecting object |
| CN103632153A (en) * | 2013-12-05 | 2014-03-12 | 宁波大学 | Region-based image saliency map extracting method |
| CN103996040A (en) * | 2014-05-13 | 2014-08-20 | 西北工业大学 | Bottom-up visual saliency generating method fusing local-global contrast ratio |
| CN104021537A (en) * | 2014-06-23 | 2014-09-03 | 西北工业大学 | Infrared and visible image fusion method based on sparse representation |
| US20160239969A1 (en) * | 2015-02-14 | 2016-08-18 | The Trustees Of The University Of Pennsylvania | Methods, systems, and computer readable media for automated detection of abnormalities in medical images |
| CN106296567A (en) * | 2015-05-25 | 2017-01-04 | 北京大学 | The conversion method of a kind of multi-level image style based on rarefaction representation and device |
Non-Patent Citations (4)
| Title |
|---|
| XIN WANG ET AL: "A sparse representation-based method for infrared dim target detection under sea–sky background", 《INFRARED PHYSICS & TECHNOLOGY》 * |
| YI ZENG，YI XU: "Saliency Detection Using Quaternion Sparse Reconstruction", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOPS》 * |
| 张宝华 等: "基于显著性检测和稀疏表示的多聚焦图像融合算法", 《小型微型计算机系统》 * |
| 张旭东: "结合区域协方差分析的图像显著性检测", 《中国图象图形学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107452230A (en) * | 2017-07-28 | 2017-12-08 | 明见（厦门）技术有限公司 | A kind of obstacle detection method, device, terminal device and storage medium |
| CN108062523A (en) * | 2017-12-13 | 2018-05-22 | 苏州长风航空电子有限公司 | A kind of infrared remote small target detecting method |
| CN108062523B (en) * | 2017-12-13 | 2021-10-26 | 苏州长风航空电子有限公司 | Infrared far-small target detection method |
| CN108898589A (en) * | 2018-06-19 | 2018-11-27 | 南通大学 | The quick-fried pearl intelligent detecting method of filter stick based on high speed machines vision |
| CN108898589B (en) * | 2018-06-19 | 2022-06-07 | 南通大学 | Filter rod bead explosion intelligent detection method based on high-speed machine vision |
| CN112749714A (en) * | 2019-10-29 | 2021-05-04 | 中国科学院长春光学精密机械与物理研究所 | Method for detecting polymorphic dark and weak small target in single-frame infrared image |
| CN114463619A (en) * | 2022-04-12 | 2022-05-10 | 西北工业大学 | Infrared dim target detection method based on integrated fusion features |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106709512B (en) | 2020-03-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709512A (en) | Infrared target detection method based on local sparse representation and contrast | |
| CN110378308B (en) | Improved port SAR image near-shore ship detection method based on fast R-CNN | |
| CN107194408B (en) | Target tracking method of mixed block sparse cooperation model | |
| CN103186904B (en) | Picture contour extraction method and device | |
| CN105574534A (en) | Significant object detection method based on sparse subspace clustering and low-order expression | |
| CN105513041A (en) | Large-scale remote sensing image sea-land segmentation method and system | |
| CN108229551B (en) | Hyperspectral remote sensing image classification method based on compact dictionary sparse representation | |
| CN105549009B (en) | A kind of SAR image CFAR object detection methods based on super-pixel | |
| CN109919910A (en) | The SAR image change detection of level set is merged and improved based on disparity map | |
| CN108428220A (en) | Satellite sequence remote sensing image sea island reef region automatic geometric correction method | |
| CN101976436A (en) | Pixel-level multi-focus image fusion method based on correction of differential image | |
| CN105160355A (en) | Remote sensing image change detection method based on region correlation and visual words | |
| CN109977968B (en) | SAR change detection method based on deep learning classification comparison | |
| CN107392887B (en) | Heterogeneous remote sensing image change detection method based on homogeneous pixel point conversion | |
| Tang et al. | A multiple-point spatially weighted k-NN method for object-based classification | |
| CN107194896A (en) | A kind of background suppression method and system based on neighbour structure | |
| CN104599288A (en) | Skin color template based feature tracking method and device | |
| CN102360503B (en) | SAR (Specific Absorption Rate) image change detection method based on space approach degree and pixel similarity | |
| CN111738332A (en) | Underwater multi-source acoustic image substrate classification method and system based on feature level fusion | |
| Liang et al. | An extraction and classification algorithm for concrete cracks based on machine vision | |
| CN109002792B (en) | SAR image change detection method based on layered multi-model metric learning | |
| CN103093472B (en) | Based on the remote sensing image change detecting method of doubledictionary intersection rarefaction representation | |
| Kekre et al. | SAR image segmentation using vector quantization technique on entropy images | |
| CN105354845A (en) | Method for semi-supervised detection on changes in remote sensing images | |
| Dong et al. | Optical remote sensing water-land segmentation representation based on proposed SNS-CNN network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |