CN100552716C - Under the global abnormal signal environment based on the associating remarkable figure robust image registration method - Google Patents

Under the global abnormal signal environment based on the associating remarkable figure robust image registration method Download PDF

Info

Publication number
CN100552716C
CN100552716C CNB2007100393744A CN200710039374A CN100552716C CN 100552716 C CN100552716 C CN 100552716C CN B2007100393744 A CNB2007100393744 A CN B2007100393744A CN 200710039374 A CN200710039374 A CN 200710039374A CN 100552716 C CN100552716 C CN 100552716C
Authority
CN
China
Prior art keywords
registration
associating
remarkable
image
abnormal signal
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.)
Expired - Fee Related
Application number
CNB2007100393744A
Other languages
Chinese (zh)
Other versions
CN101038669A (en
Inventor
顾志俊
秦斌杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CNB2007100393744A priority Critical patent/CN100552716C/en
Publication of CN101038669A publication Critical patent/CN101038669A/en
Application granted granted Critical
Publication of CN100552716C publication Critical patent/CN100552716C/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06V10/24—Aligning, centring, orientation detection or correction of the image

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

Based on the robust image registration method of the remarkable figure of associating, step is: (1) at first utilizes traditional method for registering to carry out thick registration under a kind of global abnormal signal environment; (2) figure is significantly estimated in current registration results extraction, calculate the associating of corresponding overlapping region and significantly scheme.Unite remarkable figure and provided the modeling of abnormal signal point present position, strengthen the contribution weight of public marking area in similarity measure calculates, thus the influence of removing abnormal signal adaptively; (3) utilize the similarity measure of the remarkable figure optimization of associating, obtain the optimum geometrical registration parameter under the remarkable figure of current associating based on gray scale; (4) loop optimization finishes registration with the variation of the registration parameter of twice of front and back as end condition, obtains final registration results.The present invention carries out to the modeling of abnormal signal with to the registration iteration of image, is applicable to the multi-period image registration of multimode under the global abnormal signal, the realistic demands of applications of registration accuracy.

Description

Under the global abnormal signal environment based on the associating remarkable figure robust image registration method
Technical field
What the present invention relates to is a kind of method of technical field of image processing, specifically is based on the robust image registration method of uniting remarkable figure under a kind of global abnormal signal environment.
Background technology
Image registration is a basis of applications such as computer vision, pattern-recognition, graphical analysis and important work, it estimates the optimum geometric transformation parameter of objective function and then searching image coupling by optimizing image similarity, make multiple image can eliminate otherness on space time, be integrated in the unified coordinate frame, thereby the complementation of being convenient between the image information is understood.The image that the different periods of registration different mode gather is all extremely important in fields such as medical imaging application, remote sensing, computer vision, national defence imagings.Because the physical characteristics difference of different imaging sensors and the variation of different time scenery, some notable features that exist in piece image may only be the part existence in other piece image or not have correspondence fully.These features that do not have one-to-one relationship are exactly the global abnormal signal in image registration.It specifically can show as the excision of tumour in the operation, sightless procedural characteristics point, vegetation change or the covering of cloud layer etc. in the image before art.If these abnormal signals are not correctly handled, will cause registration error to increase.
With regard to present technical merit, the method for image registration is broadly divided into two classes: based on the method for voxel gray scale with based on the method for feature.First kind method promptly based on voxel gray scale method for registering, has an important hypothesis prerequisite: the gray scale field of two width of cloth images has the relation of statistics (or linear), and this hypothesis is having the establishment of will being difficult under the situation of abnormal signal.The existence of abnormal signal will have a strong impact on the homogeneity of gray scale, causes the gray scale corresponding relation of two width of cloth images to become uncertain, and even more serious is this influence often is not a space invariance.Second class methods, promptly based on the method for registering images of feature, although extracting feature, seeking the advantage that has in the link of feature correspondence on speed and the efficient, be difficult to satisfy simultaneously the influence of not only repelling the global abnormal signal, but also the effective double requirements of reconstruction features corresponding relation.Therefore, how to reduce too much pre-service, the image automatic, accurate, that robust ground registration has abnormal signal is a technical barrier of this area as far as possible.The conventional method of global abnormal signal is in conjunction with the advantage of aforementioned two class method for registering and incorporates some and detect unusual means in the current solution image registration, to remove or to reduce the influence of abnormal signal in the registration process.But existing method seldom relates to the public marking area of how to locate in two width of cloth images subject to registration (marking area is exactly the comparatively significant area-of-interest of regional area characteristics of image contrast or gray scale complexity), and utilizes between these public marking areas correspondence to its registration that gives top priority to what is the most important.
Through the literature search of prior art is found to exist following patent to approach the present invention most, its Chinese patent publication No. is CN1920882, open day is 2007.02.28, patent name is: based on the system and method for the medical image three-dimensional multimode registration of marking area feature, be described as: " sign notable feature zone in first image and second image, wherein each zone is relevant with space scale; Come the location feature zone by each regional central point; According to local gray level the unique point of piece image and the unique point of another width of cloth image are carried out registration." wherein special processing is arranged for getting rid of overflow value (unusually); but this method does not consider to cause the global abnormal signal of characteristics of image correspondence disappearance; thereby do not make special optimization at the global abnormal signal, having under a large amount of unusual situations, its registration results is not as can be known.
Summary of the invention
The objective of the invention is to overcome the deficiencies in the prior art, propose under a kind of global abnormal signal environment based on the robust image registration method of uniting remarkable figure, make it can be under the situation that does not reduce registration accuracy, weaken the influence of the global abnormal signal that causes image correspondence disappearance, the final accurate registration of realizing image adaptive.The present invention can be applicable to various modes, multidimensional, multi-period image registration.
The present invention is achieved by the following technical solutions, comprises that step is as follows:
(1) at first utilizes traditional method for registering, treat registering images and carry out thick registration, make that the overlapping area of image is enough big;
(2) significantly figure of associating is calculated in the result images overlapping region under the current registration parameter, unite remarkable figure and provided modeling the abnormal signal present position, can show intensity of anomaly, promptly this position is the whether all corresponding public marking area of two width of cloth images specific to each overlapping this picture signal of voxel location.If this location point is the public marking area of correspondence image not, just be set to abnormal signal, do not participate in ensuing registration;
(3) associating that utilizes (2) to obtain is significantly schemed, statistical weight associating grey level histogram, and optimization obtains the optimum geometric match parameter under the remarkable figure of current associating based on the similarity measure of gray scale;
(4) loop optimization between step (2), (3) finishes registration with the variation of the registration parameter of twice of front and back as end condition, obtains final registration results.
The present invention is further described as follows:
1, associating is significantly schemed
Uniting remarkable figure is one of crucial innovation of the present invention, its proposition is mainly based on a kind of universally recognized hypothesis: different mode has the mutual correspondence (the conspicuousness degree in image local zone measured the contrast of image local provincial characteristics or the complexity of half-tone information) of local marking area at the image of same object collection under the different periods, promptly because being equal to of institutional framework, the image of gathering in the different periods of different mode can show the similarity of local salient region or correspondence one by one.
1) figure is significantly estimated in extraction
The present invention at first extracts the marking area of two width of cloth images, and selects a kind of tolerance conspicuousness degree of estimating for use, obtain image subject to registration separately significantly estimate figure.
2) locate based on the modeling of the remarkable figure of associating
The present invention proposes to utilize similarity analysis tolerance significantly to estimate the position and the degree of public marking area among the figure, and the associating that obtains between the image subject to registration is significantly schemed, the location that realization is an abnormal signal point to non-public marking area.
The codomain of uniting remarkable figure is normalized to [0,1] (significantly is worth) hereinafter referred to as the normalization associating, its application in registration process is explained as follows: if the remarkable value of normalization associating approaches 1, represent that image subject to registration has similar notable feature at this point and distributes, assert that then this point is non-abnormal signal point, belong to public marking area, continue to participate in further registration; If normalization associating significantly value approaches 0, then be illustrated in this and put two width of cloth images and do not have corresponding notable feature to distribute, this point is an abnormal signal, does not belong to public marking area, should be excluded next step registration process of participation.So just realized modeling location to the global abnormal signal.
Wherein need to prove, can select a threshold value, point direct tax the in the remarkable figure of associating that image subject to registration remarkable measure value separately is lower than threshold value is 0, do not comprise the conspicuousness feature because the point of low conspicuousness means, do not have the carrying out that is beneficial to registration.
2, in conjunction with the calculating based on the gray scale similarity measure of uniting remarkable figure
The associating of setting forth in above-mentioned 1 is significantly schemed, and must just can be used for concrete registration in conjunction with corresponding similarity measure, and the present invention is conceived to the improvement based on the gray scale similarity measure.
When calculating based on the gray scale similarity measure, thereby the probability distribution, the joint probability distribution that often need to treat registering images are made the value of estimating to calculate entropy.A kind of generally accepted simpler and easy method of estimation efficiently is that the associating grey level histogram by statistical picture is similar to probability distribution.The remarkable information fusion of associating that the present invention comprises the remarkable figure of associating that obtains in 1 can strengthen the weight of public marking area in registration, thereby reach the effect of adaptive eliminating global abnormal signal in the statistics of associating grey level histogram.
Concrete fusion method is as follows: in the process of statistical picture associating grey level histogram, because it and registration transformation parameter correlation, need and wherein to add up with reference image R after a width of cloth such as the floating image F conversion, if the position after the conversion not on the regular coordinate grid of reference picture, then needs interpolation calculation to go out the reference picture gray scale of this position.The present invention use normalization associating significantly value determine that every pair of voxel on the image is to uniting the contribution weight of grey level histogram, promptly in the gray scale that has obtained floating image F and reference image R to (f, r) after, with normalization associating significantly value count histogram frequency h (f, r) in.If its weights are very little, these are judged to abnormal signal or the very little point of conspicuousness is just got rid of outside the registration process automatically.
The step that the present invention proposes requires registration process calculating significantly figure and optimize to circulate between the geometric transformation parameter and carry out of associating, is in essence to carry out combining for the modeling of abnormal signal and registration.Can estimate comparatively ideal associating by the registration results that both gets and significantly scheme, and can reflect the residing position of abnormal signal more accurately based on uniting remarkable figure more accurately, both sides are hocketed and will be made registration results obtain improvement progressively.
The present invention is by proposing a kind of method to the modeling of abnormal signal position---and associating is figure significantly, locate the public marking area in two width of cloth images subject to registration, and utilizing between these public marking areas correspondence to its registration that gives top priority to what is the most important, those are identified as and cause that the gray scale field concerns that the voxel of uncertain abnormal signal then is excluded outside registration process between image.The present invention has realized under the global abnormal signal environment in multimode, the multi-period image registration the special optimization for abnormal signal thus, also registration preferably under the situation that a large amount of global abnormal signals are arranged.In the design of method, absorbed based on (extract significantly estimate figure) of feature with based on the advantage of (adopt based on gray scale similarity measure as the optimization aim function) two traditional methods of gray scale voxel, make that the robustness of method and efficient can both practical requirement, can be applicable to medical image, the registration of multiple occasion such as the image of taking photo by plane, video image.Through many test shows, the present invention has improved the success ratio at global abnormal signal hypograph registration greatly, and the precision of registration can reach sub-pixel (being registration bias＜1 pixel).
Description of drawings
Fig. 1 is a process flow diagram of the present invention.
Fig. 2 is sample result figure; Wherein (a) figure is MRI-PD and MRI-T1 registration example; (b) figure is a CT-PET registration example; (c) figure is the different period registration examples of MRI.Every width of cloth instance graph distributes and is: upper left reference image R, and upper right floating image F, lower-left registration finish the final associating in back and significantly scheme (significantly value is big more for the white more expression normalization associating of color), and the bottom right is a registration results.
Embodiment
Below embodiments of the invention are elaborated: present embodiment has provided detailed embodiment and process being to implement under the prerequisite with the technical solution of the present invention, but protection scope of the present invention is not limited to following embodiment
The object of registration is actual two dimensional image (accompanying drawing 2), the process flow diagram of registration as shown in Figure 1:
After having obtained thick registration results, extract operator by conspicuousness and obtain reference picture and floating image " significantly estimating figure " separately, can get both " associating is figure significantly " through similarity analysis, calculate " based on the similarity measure of gray scale " in conjunction with " associating is figure significantly " and certain interpolation algorithm, as objective function, optimization obtains optimum geometrical registration parameter with it.If do not reach " end condition ", then recomputate " significantly estimating figure ", " associating is figure significantly " with this registration parameter, carry out the optimization of a new round, otherwise then stop, the output net result.Concrete enforcement with regard to some part elaborates below:
1, thick registration
Employing is based on the method for traditional normalized mutual information, and transformation model is chosen two-dimentional rigid body translation model, and optimized Algorithm is a simplex algorithm.
2, calculate significantly figure of associating
(1) figure is significantly estimated in extraction
The algorithm of traditional Flame Image Process is all regarded each voxel in the image as an independent random variables, and this obviously is ungratified in real image, and each voxel can influence its neighborhood voxel at least.Computation model based on the local energy function of multidimensional random field has been considered this neighborhood relationships, and this example is specialized this computation model that to extract operator as conspicuousness as follows:
E ( v ) = Σ m , n ( I ( v ) - I ( v x - m , v y - n ) 2 - - - ( 1 )
Specify that v is a bivector, represents two-dimentional voxel (hereinafter referred to as pixel), I (v) represents v point corresponding gray, v x, v yIt is the two dimensional component of v, m, n are the neighborhood range threshold that certain type of neighborhood system is determined, during common 3*3 neighborhood, m, n be 1,0,1}, in order to obtain the rotational invariance energy, what the present invention implemented is the circumference neighborhood, is that circle is done at center certain radius (length in pixels) with the v point promptly, and the half-tone information of circle inside is as the method for its neighborhood information.E (v) is the corresponding remarkable measure value of v point.
(2) calculate significantly figure of associating
Here choose the square similarity analysis as the method that produces the remarkable figure of associating.Each can obtain separately inertial matrix major axes orientation vector by the square analysis to the position of overlaid pixel in two width of cloth images subject to registration.If this position belongs to the public marking area of both sides, then this will have similarity to vector, in the depth of parallelism or the degree of correlation that show as on the mathematics between vector.
Specific implementation process is as follows: to overlaid pixel, is inertial matrix and eigenwert and the proper vector of center calculation its neighborhood territory pixel (adopt round domain system (1) in) with this point in two width of cloth images each, writes down eigenvalue of maximum characteristic of correspondence vector.Calculate significantly value of normalization associating with following formula:
W ( v ) = | eig ( v f ) T eig ( v r ) eig ( v f ) T eig ( v f ) eig ( v r ) T eig ( v r ) | - - - ( 2 )
Eig (v f) and eig (v r) represent the gray scale energy field major axes orientation vector of floating image and reference picture corresponding point neighborhood respectively.It (v) is 0 that the point that wherein will be lower than 0.005 times of energy mxm. in remarkable figure is directly given W.
3, estimate in conjunction with the normalized mutual information of the remarkable figure of associating
This example is chosen the similarity measure of normalized mutual information as registration, and it is based on similarity measure a kind of of gray scale.In the process of statistical picture associating grey level histogram, choose the easy bilinear interpolation of main flow as interpolating method, (f, r) gradation of image is right can to obtain floating image and reference picture.With the normalization associating of the lap position point that obtains in 2 significantly value join that (f is r) in the pairing histogram frequency.All the other parts of calculating normalized mutual information are similar with classic method, do not do detailed description at this.
4, prioritization scheme
Utilize the normalized mutual information of mentioning in above-mentioned 3 to estimate the destination object of optimizing as registration, pairing geometric transformation parameter was the optimum registration parameter under the remarkable figure of current associating when it reached peaked.With this registration parameter floating image F is carried out conversion and obtain F ', can carry out the calculating of the remarkable figure of associating of a new round F ' and reference image R.After having obtained the remarkable figure of associating, carry out the registration optimizing process of a new round again.So just realized that remarkable figure of calculating associating and registration optimization replace iteration and carries out.
Euler's distance of twice resulting optimum registration parameter chooses 0.00001 as end condition before and after adopting among the embodiment.Set a greatest iteration step number 200 in addition and prevented that the situation that program can't be jumped out from taking place.
5, registration results
(1) many group sample results explanation
Carried out above-mentioned registration for MRI-PD with MRI-T1, CT and the image example of PET, different periods of MRI (the image size is 256*256) respectively in the example.All obtained gratifying result.Chosen in the accompanying drawing 2 wherein several routine MRI-PD and MRI-T1 (figure a), CT-PET (figure b) and the image (scheming c) of different periods of MRI show the present invention for the multi-period image of multimode in following registration results that can reach of the situation that abnormal signal is arranged.Wherein MRI-PD-MRI-T1 has added artificial noise at floating image; CT-PET is owing to the difference of pattern, and both have very dissimilar characteristic, contain tumour; The image of different periods of MRI then is to gather the image before and after tumor resection with patient.The arrangement mode of accompanying drawing 2 every width of cloth images is: upper left reference image R, and upper right floating image F, (showing with gray scale) significantly schemed in the associating that the final iteration in lower-left obtains, the final registration results in bottom right, the right-angled intersection of straight line has shown the precision of registration.In iteration for several times after all automatically convergence withdraw from.As seen from the figure, the associating that the is drawn reflection that significantly figure is all good the position of abnormal signal.
(2) registration accuracy explanation
The good one group of MRI image of registration in the application example is as standard, floating image F is carried out the conversion of displacement and rotation, produce the image subject to registration of different displacements of 30 width of cloth and rotation, use this method registration with reference image R respectively, the registration error that finally draws (average registration error ± standard deviation) is: horizontal shift is 0.31 ± 0.59 pixel, perpendicular displacement is-0.38 ± 0.43 pixel, and the anglec of rotation is 0.11 ± 0.21 degree.The whole registration successes of 30 width of cloth images.From the result, registration accuracy has reached sub-pixel, can be applied to reality fully.
(3) registration curve comparative descriptions
Chosen one group from ITK (Insight ToolKit, www.itk.org) in the accurate example of registration, do the comparison of registration curve.Specific implementation is to make the measure curve figure of variation range in the x of [20,20] direction, y direction and the anglec of rotation respectively, relatively its similarities and differences.
The result shows that after the initial registration result has been arranged, it will be bigger that new normalized mutual information is estimated than the original gradient of estimating, and meaning newly to estimate has speed of convergence faster.Its reason is that the present invention has been dissolved into public marking area characteristic information in the calculating of estimating, and has accelerated the speed of convergence of similarity measure.

Claims (8)

  1. Under the global abnormal signal environment based on the robust image registration method of the remarkable figure of associating, may further comprise the steps:
    (1) at first utilizes traditional method for registering, treat registering images and carry out thick registration; Image subject to registration refers to floating image and reference picture here;
    (2) extract operator by conspicuousness and obtain reference picture and floating image " significantly estimating figure " separately, significantly figure of associating is calculated in the result images overlapping region under the current registration parameter; The codomain of uniting remarkable figure is normalized to [0,1], and normalization is united remarkable figure and provided modeling to the abnormal signal present position, can show specific to the floating image of each overlapping voxel location and the intensity of anomaly of reference picture;
    (3) utilize the similarity measure of the remarkable figure optimization of associating, obtain the optimum geometrical registration parameter under the remarkable figure of current associating based on gray scale;
    (4) loop optimization between step (2), (3) finishes registration with the variation of the registration parameter of twice of front and back as end condition, obtains final registration results.
  2. 2. under the global abnormal signal environment according to claim 1 based on the associating remarkable figure robust image registration method, it is characterized in that, significantly figure is united in the result images overlapping region under the current registration parameter is calculated described in the step (2), and the specific implementation way is as follows:
    A) extract image subject to registration separately significantly estimate figure;
    B) significantly estimate figure according to what obtain, calculate each, obtain uniting remarkable figure the remarkable value of the normalization associating of overlapping tissue points.
  3. 3. under the global abnormal signal environment according to claim 2 based on the associating remarkable figure robust image registration method, it is characterized in that, described a), extract image subject to registration separately significantly estimate figure, be meant: utilize conspicuousness extraction operator that the marking area of image subject to registration is extracted, and select a kind of tolerance conspicuousness degree of estimating for use, obtain image subject to registration separately significantly estimate figure.
  4. 4. under the global abnormal signal environment according to claim 2 based on the associating remarkable figure robust image registration method, it is characterized in that, described b), be meant: utilize similarity analysis tolerance significantly to estimate the position and the degree of public marking area among the figure, obtain each remarkable value of normalization associating, thereby the associating that makes up between floating image subject to registration and the reference picture is schemed significantly to overlapping tissue points.
  5. According under claim 1 or the 4 described global abnormal signal environments based on the associating remarkable figure robust image registration method, it is characterized in that, it is [0 that remarkable value scope is united in the normalization that remarkable figure is united in described normalization, 1], in registration process, if the remarkable value of normalization associating approaches 1, represent that image subject to registration has similar notable feature at this point and distributes, assert that then this point is non-abnormal signal point, belong to public marking area, continue to participate in further registration; If normalization associating significantly value approaches 0, then be illustrated in this and put two width of cloth images and do not have corresponding notable feature to distribute, this point is an abnormal signal, does not belong to public marking area, should be excluded the registration process that participates in next step, so just realize modeling location the global abnormal signal.
  6. 6. under the global abnormal signal environment according to claim 4 based on the associating remarkable figure robust image registration method, it is characterized in that, pass through setting threshold, point direct tax the in the remarkable figure of associating that image subject to registration remarkable measure value separately is lower than threshold value is 0, and the location point that remarkable measure value is low is excessively got rid of outside the calculating of the remarkable figure of associating.
  7. 7. under the global abnormal signal environment according to claim 1 based on the associating remarkable figure robust image registration method, it is characterized in that, in the described step (3), add up in the following manner based on the used associating grey level histogram of the similarity measure of gray scale:
    The normalization associating of uniting remarkable figure with normalization significantly value determines that every pair of voxel on the image is to uniting the contribution weight of grey level histogram, obtained the gray scale of floating image F and reference image R to (f in interpolation, r) after, with normalization associating significantly value count grey level histogram frequency h (f, r) in.
  8. 8. based on the robust image registration method of the remarkable figure of associating, it is characterized in that under the global abnormal signal environment according to claim 1, described step (4), specific as follows:
    Remarkable figure of calculating associating and optimum geometrical registration parameter alternately iteration are carried out, after having obtained the remarkable figure of associating each time, on this basis similarity measure is optimized, draws the optimum geometrical registration parameter under the remarkable figure of this associating, so just finished iteration one time; The registration parameter of last iteration gained will be as the initial parameter of next iteration, iteration stopping when satisfying end condition.
CNB2007100393744A 2007-04-12 2007-04-12 Under the global abnormal signal environment based on the associating remarkable figure robust image registration method Expired - Fee Related CN100552716C (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CNB2007100393744A CN100552716C (en) 2007-04-12 2007-04-12 Under the global abnormal signal environment based on the associating remarkable figure robust image registration method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CNB2007100393744A CN100552716C (en) 2007-04-12 2007-04-12 Under the global abnormal signal environment based on the associating remarkable figure robust image registration method

Publications (2)

Publication Number Publication Date
CN101038669A CN101038669A (en) 2007-09-19
CN100552716C true CN100552716C (en) 2009-10-21

Family

ID=38889553

Family Applications (1)

Application Number Title Priority Date Filing Date
CNB2007100393744A Expired - Fee Related CN100552716C (en) 2007-04-12 2007-04-12 Under the global abnormal signal environment based on the associating remarkable figure robust image registration method

Country Status (1)

Country Link
CN (1) CN100552716C (en)

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8055101B2 (en) * 2008-04-29 2011-11-08 Adobe Systems Incorporated Subpixel registration
CN101763633B (en) * 2009-07-15 2011-11-09 中国科学院自动化研究所 Visible light image registration method based on salient region
CN101706961B (en) * 2009-11-10 2012-12-12 北京航空航天大学 Image registration method and image registration device
CN101847264B (en) * 2010-05-28 2012-07-25 北京大学 Image interested object automatic retrieving method and system based on complementary significant degree image
CN102274042B (en) * 2010-06-08 2013-09-04 深圳迈瑞生物医疗电子股份有限公司 Image registration method, panoramic imaging method, ultrasonic imaging method and systems thereof
CN103003811A (en) * 2010-06-09 2013-03-27 惠普发展公司，有限责任合伙企业 Determining similarity scores of anomalies
CN102005057B (en) * 2010-11-17 2012-07-25 中国科学院声学研究所 Method for detecting region of interest of color image
CN102129693B (en) * 2011-03-15 2012-07-25 清华大学 Image vision significance calculation method based on color histogram and global contrast
US20130044927A1 (en) * 2011-08-15 2013-02-21 Ian Poole Image processing method and system
CN103279968B (en) * 2013-05-09 2015-11-18 四三九九网络股份有限公司 Detect the method and apparatus of GPU rendering animation pixel collision on the equipment of mobile terminal
CN103747240B (en) * 2013-12-25 2015-10-21 浙江大学 The vision significance filtering method of Fusion of Color and movable information
JP6641696B2 (en) * 2015-02-05 2020-02-05 富士通株式会社 Image display device, image display program, and image display method
US11538176B2 (en) * 2015-12-15 2022-12-27 Koninklijke Philips N.V. Image processing systems and methods
US11175398B2 (en) * 2016-12-21 2021-11-16 The Boeing Company Method and apparatus for multiple raw sensor image enhancement through georegistration
CN109584283A (en) * 2018-11-29 2019-04-05 合肥中科离子医学技术装备有限公司 A kind of Medical Image Registration Algorithm based on convolutional neural networks
CN112150404B (en) * 2019-06-27 2023-10-27 上海交通大学 Global-to-local non-rigid image registration method and device based on joint saliency map
CN114648564B (en) * 2022-05-23 2022-08-23 四川大学 Visible light and infrared image optimization registration method and system for unsteady state target

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
全局异常信号环境下基于体素灰度多模医学图像配准研究. 秦斌杰，庄天戈.航天医学与医学工程，第17卷第2期. 2004
全局异常信号环境下基于体素灰度多模医学图像配准研究. 秦斌杰，庄天戈.航天医学与医学工程，第17卷第2期. 2004 *
基于特征的遥感图像自动配准算法. 韦燕凤，赵忠明，闫冬梅，曾庆业.电子学报，第33卷第1期. 2005
基于特征的遥感图像自动配准算法. 韦燕凤，赵忠明，闫冬梅，曾庆业.电子学报，第33卷第1期. 2005 *
基于视觉注意的遥感图像分析方法. 张鹏，王润生.电子与信息学报，第27卷第12期. 2005
基于视觉注意的遥感图像分析方法. 张鹏，王润生.电子与信息学报，第27卷第12期. 2005 *

Also Published As

Publication number Publication date
CN101038669A (en) 2007-09-19

Similar Documents

Publication Publication Date Title
CN100552716C (en) Under the global abnormal signal environment based on the associating remarkable figure robust image registration method
Xia et al. Geometric primitives in LiDAR point clouds: A review
CN104536009B (en) Above ground structure identification that a kind of laser infrared is compound and air navigation aid
CN102938142B (en) Based on the indoor LiDAR missing data complementing method of Kinect
CN105787439A (en) Depth image human body joint positioning method based on convolution nerve network
CN101777181B (en) Ridgelet bi-frame system-based SAR image airfield runway extraction method
CN101882308A (en) Method for improving accuracy and stability of image mosaic
CN109389062B (en) Method for extracting lake water and land dividing line by using high-resolution satellite-borne SAR image
CN102982334B (en) The sparse disparities acquisition methods of based target edge feature and grey similarity
CN106488122A (en) A kind of dynamic auto focusing algorithm based on improved sobel method
CN105678733A (en) Infrared and visible-light different-source image matching method based on context of line segments
CN102446358A (en) Multi-mode medical image registration method based on edge features and CS (Cauchy-Schwarz) information
CN109472802A (en) A kind of surface grid model construction method constrained certainly based on edge feature
CN102867309A (en) SAR (Synthetic Aperture Radar) image change detection method based on mixing model
CN106225774A (en) A kind of unmanned agriculture tractor road measurement apparatus based on computer vision and method
CN103714547A (en) Image registration method combined with edge regions and cross-correlation
CN101246595A (en) Multi-view point data splitting method of optical three-dimensional scanning system
CN109978919A (en) A kind of vehicle positioning method and system based on monocular camera
CN108876861A (en) A kind of solid matching method of objects outside Earth rover
Guo et al. Exploring GIS knowledge to improve building extraction and change detection from VHR imagery in urban areas
Tatar et al. High-resolution satellite stereo matching by object-based semiglobal matching and iterative guided edge-preserving filter
Long et al. Image segmentation based on the minimum spanning tree with a novel weight
CN104361597A (en) Liver DTI image segmentation method based on active contour model
CN101661623A (en) Three-dimensional tracking method of deformable body based on linear programming
Li et al. Automatic Road Extraction from High-Resolution Remote Sensing Image Based on Bat Model and Mutual Information Matching.

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20091021

Termination date: 20160412