CN103337080A - Registration technology of infrared image and visible image based on Hausdorff distance in gradient direction - Google Patents
Registration technology of infrared image and visible image based on Hausdorff distance in gradient direction Download PDFInfo
- Publication number
- CN103337080A CN103337080A CN2013102946761A CN201310294676A CN103337080A CN 103337080 A CN103337080 A CN 103337080A CN 2013102946761 A CN2013102946761 A CN 2013102946761A CN 201310294676 A CN201310294676 A CN 201310294676A CN 103337080 A CN103337080 A CN 103337080A
- Authority
- CN
- China
- Prior art keywords
- gradient direction
- angle
- infrared image
- angle point
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a registration technology of an infrared image and a visible image based on a Hausdorff distance in a gradient direction, and belongs to the field of image matching. The technology is characterized by comprising the following steps of 1) extracting angular points from the infrared image and the visible image respectively, and obtaining an initial angular point set, 2) calculating the gradient direction of each angular point, and decomposing the initial angular point set into eight subsets, and 3) only looking for registration point pairs in the corresponding subsets when the Hausdorff distance is calculated for registration, thereby reducing interference of the points in the corresponding subsets in different intervals. Compared with the existing Hausdorff distance registration algorithm, the technology has better accuracy in the registration of the infrared image and the visible image.
Description
Technical field
The present invention relates to infrared image and visible light image registration method, belong to technical field of image matching.
Background technology
Image registration is exactly that two width or multiple image that obtain (weather, illumination, camera position and angle etc.) under different time, different sensors (imaging device) or different condition carry out the process of best match, that is, finding one makes the most close or most like conversion of two images.Image registration has been widely used in multiple fields, for example:The fields such as remotely-sensed data analysis, computer vision, image procossing.
Infrared image is different with the image-forming principle of visible images, and what infrared image was obtained is thermal image, and the gray value of image is corresponding with scene temperature profile;And it is the power of visible ray that visible images, which are obtained, its gray scale is that reflection by object to light and dash area are determined.Due to infrared image and the different qualities of visible images so that the two has complementarity.This complementary effect, makes to result in more comprehensively information after infrared image and visual image fusion, and image registration is the indispensable step of image co-registration.
Conventional heterologous method for registering images, can be divided into the algorithm of algorithm and feature based based on pixel, the algorithm based on pixel is suitable for having the image of certain correlation between the intensity profile of two images at present;The algorithm of feature based is applied to the obvious image of structure.The algorithm of feature based can in most cases obtain preferable registration effect in actual applications.
Conventional feature includes characteristic point and edge at present.Because edge pixel number can be relatively more, when edge is as feature, amount of calculation can be made larger;Angle point has geometric figure consistency, is a kind of effective characteristic point, therefore calculated in the present invention from angle point as characteristic point.Based on Hausdorff apart from matching algorithm, due to the simplicity of its calculating process, the registration of heterologous image is had been widely used for.But existing Hausdorff distance algorithms are directly directly to calculate Hausdorff distances with edge pixel set to carry out registration mostly, this causes 1)Operand is larger;2)Due to not classifying to characteristic point, therefore noise spot is more, influences the degree of accuracy of registration.
The content of the invention
Accuracy and reduction amount of calculation of the invention in order to improve infrared image and visible light image registration, it is proposed that a kind of Hausdorff based on gradient direction is apart from infrared image and visible light image registration algorithm.Extract the angle point of infrared image and visible images respectively first;Then the angle point of extraction is divided into 8 subclass according to gradient direction, and calculates the Hausdorff distances between corresponding subset conjunction;Finally calculate and obtain the Hausdorff distances based on gradient direction.The present invention not only eliminates the interference of noise spot compared with traditional Hausdorff distance methods, reduces registration error, and reduce amount of calculation.
Current Hausdorff algorithms are typically that selection edge is calculated, and the pixel at edge is more, and amount of calculation can be made larger.Calculated to reduce selection characteristic point in the amount of calculation present invention, due to the premium properties that Harris angle points have, therefore the present invention uses Harris angle points as characteristic point.The basic thought of Harris Corner Detection Algorithms is to determine angle point using variation of image grayscale rate, this method is by calculating a distance feature being associated with image auto-correlation function, i.e. the single order curvature of auto-correlation function come judge certain point whether as angle point, if two curvature values are all high, then think that the point is angle point.The pixel quantity of angle point is far smaller than the pixel count at edge, therefore greatly reduces amount of calculation from Corner Feature progress registration, shortens the registering time.
The present invention using Harris detective operators after angle point is extracted, and the angle point obtained to extraction is divided into 8 characteristic point subclass according to the angular interval where the gradient direction angle of angle point;Then the Hausdorff distances between corresponding subset conjunction are calculated;The average of 4 Hausdorff distances of minimum is finally calculated, the Hausdorff distances based on gradient direction are used as.The present invention divides angle point set according to gradient direction, calculates the Hausdorff distances between corresponding subset conjunction, eliminates the interference of angle point in other subclass, improves the degree of accuracy of registration.
The present invention is realized by following technical scheme:Hausdorff based on gradient direction is apart from registration Algorithm, it is characterised in that including following 4 big step:1)Extract Harris angle points and calculate the gradient direction angle of each angle point;2)It is interval according to the difference residing for angle point gradient direction, angle point set is divided into 8 different subclass;3)Calculate the Hausdorff distances between the subclass in two images corresponding to same angular interval;4)The average of 4 minimum Hausdorff distances is calculated, the Hausdorff distances based on gradient direction are used as.
The step 1)In, the Harris angle points of infrared image and visible images are extracted first, obtain corner location figure;Then the gradient direction angle of each pixel of image is calculated, gradient direction figure is obtained;Finally the gradient direction angle that point multiplication operation obtains each angle point is carried out with corner location figure and gradient direction figure.
The step 2)In, by 1)In the obtained scope of gradient direction angle be, 8 intervals are divided into, each interval is accounted for, 8 angular intervals are respectively:,,,,,,,.According to the division methods of 8 Direction intervals, the angle point of infrared image and visible ray figure is all divided into 8 subclass.
The step 3)In, the Hausdorff distances between the corresponding subclass of same angular interval in two images with punctual, is only calculated.
The step 4)In, step 3)Middle calculating obtains the corresponding 8 Hausdorff distance values of 8 subclass, and the present invention takes the average of 4 distance values of minimum as the Hausdorff distances based on gradient direction.
Brief description of the drawings
Fig. 1 is the flow chart of the registration Algorithm of the Hausdorff distances of the invention based on gradient direction;
Fig. 2 is the division methods schematic diagram of gradient direction angle range.
Embodiment
With reference to embodiment, the present invention is described in further detail:
1st, the imaging features of infrared image and visible images
The imaging system of infrared image is the Temperature Distribution using object each several part, and formation is temperature profile, and its gradation of image is changed with the radiation variation of different parts.Visible images are that reflection by object to light and dash area are determined.Infrared image is different with the image-forming principle of visible images, the two is had complementary effect.Infrared image can overcome the obstacle on partial visual and detect target, and night being capable of normal imaging, the necessity as detection device, but have more noise in infrared image.There is less noise in visible images, but can not normal imaging at night.More comprehensively information is resulted in after this complementary effect, infrared image and visual image fusion, image registration is the essential step of fusion.
2nd, the Hausdorff distances based on gradient direction realize the registration of infrared image and visible images:
To visible imagesTranslated, rotated and change of scale, obtain, whereinWithTranslational movement horizontally and vertically is represented respectively;Want to obtain preferable matching effect visible ray and infrared image at least 1/9 lap, that is to say, that what visible ray and infrared image at least 1/9 part were represented is Same Scene, so in the present inventionWithSpan be,, hereWithIt is the height and width of image respectively;Rotation parameter is represented, it is changed with 2 for step-length every time;Because visible ray resolution ratio is more than infrared resolution rate, therefore change of scale parameter, it is changed every time with 0.1 for step-length.For each transformation parameter, step 2-1 is performed), obtain a corresponding Hausdorff distance based on gradient direction of conversion。
2-1)Reference picture 1, the Hausdorff distances based on gradient direction comprise the following steps:
The first step, extracts angle point, obtains corner location figure.To infrared imageWith the image after visible images translation, rotation and change of scaleHarris angle points are extracted respectively, obtain the angle point set of two images.The principle of Harris operators detection angle point is whether the angle point response for judging each pixel is more than threshold value, is then judged to angle point if greater than threshold value, is not angle point otherwise.Angle point response depends on correlation matrix,It is:
Response diagramFor:
Wherein,Represent that two matrix corresponding elements are multiplied;,,,It is Gaussian function,WithIt is respectivelyWithThe difference in direction,Value be typically that 0.04 ~ 0.06, * represents convolution.Obtain after response diagram R, the value to each element in R is analyzed, the value of some element is more than threshold value T set in advance, be considered as the element correspondence position for angle point.Corner location figure representation is:
Wherein,Corner location figure is represented,RepresentFigure position isValue, ifValue be 1, represent positionIt is angle point, ifValue be 0, represent positionIt is not angle point;RepresentFigure position isValue;For angle point response threshold value.The present invention is selected when extracting angle point;When handling infrared imageTake 67, during processing visible imagesTake 500.
Second step, calculates the gradient direction angle of each pixel of image, obtains gradient direction figure.Here the gradient direction angle of each pixel is calculated with Sobel operators, specifically, using the both direction template of Sobel operators, calculated level and vertical gradient value, then calculate according to Grad and obtain gradient direction angle respectively.The both horizontally and vertically template of wherein Sobel operators is respectively:
Wherein,For pending image,,,WithIt is Sobel both horizontally and vertically template pair respectivelyResult after being handled.Representing matrixWithPointwise is divided by.
3rd step, obtains the gradient direction angle of each angle point, and is classified according to gradient direction angle angle steel joint.Corner location figure and gradient direction figure are carried out into corresponding element to be multiplied, the gradient direction angle of each angle point is obtained.Here the two carry out computing be:, whereinIt is angle point gradient direction figure.
Will8 intervals are divided into, the interval according to where the gradient direction angle of each angle point is different, and angle point set is decomposed into 8 subclass.Wherein interval division methods reference picture 2, subinterval is respectively:,,,,,,,。
4th step, calculates the Hausdorff distances between each subclass, can obtain 8 Hausdorff distance values.
a)Hausdorff distances
Hausdorff distances are the direct similarities calculated between two point sets, and calculating process is easy.The definition of Hausdorff distances is minimax distance, it is assumed that is extracted from two images and obtains two point sets, the two set of characteristic points are designated as respectivelyWith, traditional Hausdorff distance definitions are:
Wherein,,WithIt is the angle point in two images respectively,WithRepresent respectivelyIn setmIndividual point andnThe corresponding coordinate value horizontally and vertically of individual angle point.Find out that it is more sensitive to noise spot, as long as there is a point far from the minimum range of image subject to registration from Hausdorff definition, then can make that the distance of calculating is very big, this makes registration result error easily occur.
b)Hausdorff distance algorithms based on gradient direction
After decomposition by diagonal point set, 8 point sets in infrared image are designated as, it is seen that 8 point sets in light image are designated as.In order to reduce the interference of noise spot, between calculating corresponding point set apart from when, add restrictive condition, i.e., when in larger distance between two points, for example distance is more than 6 pixel distances, it is believed that the two is unlikely to be registering point pair.Distance in this patentAnd distanceComputing formula uses improved Hausdorff distances, is expressed as:
Wherein,WithIt is respectivelyWithIn arbitrfary point;;For distance threshold, 6 are taken in the present invention.WithRepresent respectively in correspondence formulaThe number being not zero.
Calculate the distance between corresponding subset conjunction, the point pair of registration is only looked in corresponding subset conjunction, computing formula is:
Calculate the Hausdorff distance values between 4 minimum subclassAverage, this average is used as to the Hausdorff distances based on gradient direction.In order to eliminate the distance caused registration error bigger than normal between Direction interval correspondence point set, the average value of less 4 distance values is only taken, the Hausdorff distances based on gradient direction are expressed as:
WhereinIt isOrdered series of numbers after being arranged by ascending order,Represent correspondent transformWhen the Hausdorff distances based on gradient direction.
2-2)The calculating of registration parameter
Pass through 2-1)Calculate and obtain a series of Hausdorff distances based on gradient direction;The distance value of gained is compared, a minimum distanceCorresponding translation, rotation and change of scale parameter,,,It is exactly optimal registration parameter:I.e.
Claims (6)
1. a kind of infrared image and the method for registering of visible images of the Hausdorff distances based on gradient direction, it is characterized in that comprising the following steps:
1)To visible imagesTranslated, rotated and change of scale, obtain, whereinWithTranslational movement horizontally and vertically is represented respectively;Want to obtain preferable matching effect visible ray and infrared image at least 1/9 lap, that is to say, that what visible ray and infrared image at least 1/9 part were represented is Same Scene, so in the present inventionWithSpan be,, hereWithIt is the height and width of image respectively;Rotation parameter is represented, it is changed with 2 for step-length every time;Because visible ray resolution ratio is more than infrared resolution rate, therefore change of scale parameter, it is changed every time with 0.1 for step-length;
3)The gradient direction angle of each pixel of image is calculated, gradient direction figure is obtained;
4)The gradient direction angle of each angle point is obtained, according to the gradient direction angle of each angle point, one in 8 subintervals divided in advance is referred to, after the completion of all angle points are sorted out, 8 angle point subclass can be obtained;
5)The Hausdorff distances between the conjunction of same direction corresponding subset are calculated, 8 distance values are obtained;
6)The average of the Hausdorff distance values between 4 subclass of minimum is sought, the Hausdorff distances based on gradient direction are used as;
2. the infrared image and the method for registering of visible images of a kind of Hausdorff distances based on gradient direction according to claim 1, it is characterized in that the step 2)Extract infrared imageAnd imageAngle point, utilize Harris operator extractions infrared image and the angle point of visible images, calculate the angle point response of each pixel,During more than threshold value set in advance, the pixel is judged as angle point;Otherwise, pixel is not angle point, and the computing formula of response diagram is:
Wherein,Represent that two matrix corresponding elements are multiplied;,,,It is Gaussian function,WithIt is respectivelyWithThe difference in direction, Matrix point multiplication operation is represented,Value be typically that 0.04 ~ 0.06, * represents convolution;Obtain after response diagram R, the value to each element in R is analyzed, the value of some element is more than threshold value T set in advance, be considered as the element correspondence position for angle point, corner location figure representation is:
Wherein,Corner location figure is represented,RepresentFigure position isValue, ifValue be 1, represent positionIt is angle point, ifValue be 0, represent positionIt is not angle point;RepresentFigure position isValue;For angle point response threshold value, the present invention is selected when extracting angle point;When handling infrared imageTake 67, during processing visible imagesTake 500.
3. the infrared image and the method for registering of visible images of a kind of Hausdorff distances based on gradient direction according to claim 1, it is characterized in that the step 3)The gradient direction angle of each pixel of image is calculated, gradient direction figure is obtained, the gradient direction angle of each pixel is calculated using Sobel operators, the both horizontally and vertically template of Sobel operators is respectively:
The computing formula of deflection is:
4. the infrared image and the method for registering of visible images of a kind of Hausdorff distances based on gradient direction according to claim 1, it is characterized in that the step 4)The gradient direction angle of each angle point is obtained, according to the gradient direction angle of each angle point, one in 8 subintervals divided in advance is referred to, after the completion of all angle points are sorted out, 8 angle point subclass can be obtained;Will8 subintervals are divided into, the division methods in subinterval are:,,,,,,,;Obtained gradient direction figure will be calculatedComputing is carried out with corner location figure, the gradient direction angle of each angle point is obtained, computing formula is:
WhereinIt is angle point gradient direction figure.
5. the infrared image and the method for registering of visible images of a kind of Hausdorff distances based on gradient direction according to claim 1, it is characterized in that the step 5)The corresponding Hausdorff distances of same direction between subclass are calculated, 8 distances are obtained;Specially after angle steel joint is classified, infrared image and visible images obtain 8 subclass respectively, and wherein 8 in infrared image point set is designated as, it is seen that 8 point sets in light image are designated as, the definition of distance between corresponding subset is closed is:
Wherein,WithIt is respectivelyWithEach point element in set;;For distance threshold, 6 are taken in the present invention,WithRepresent respectively in correspondence formulaThe number being not zero.
6. the infrared image and the method for registering of visible images of a kind of Hausdorff distances based on gradient direction according to claim 1, it is characterized in that the step 6)The average of the Hausdorff distance values between 4 subclass of minimum is sought, the Hausdorff distances based on gradient direction are used as, it is expressed as
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013102946761A CN103337080A (en) | 2013-07-15 | 2013-07-15 | Registration technology of infrared image and visible image based on Hausdorff distance in gradient direction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013102946761A CN103337080A (en) | 2013-07-15 | 2013-07-15 | Registration technology of infrared image and visible image based on Hausdorff distance in gradient direction |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103337080A true CN103337080A (en) | 2013-10-02 |
Family
ID=49245228
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2013102946761A Pending CN103337080A (en) | 2013-07-15 | 2013-07-15 | Registration technology of infrared image and visible image based on Hausdorff distance in gradient direction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103337080A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104966281A (en) * | 2015-04-14 | 2015-10-07 | 中测新图(北京)遥感技术有限责任公司 | IMU/GNSS guiding matching method of multi-view images |
CN105371957A (en) * | 2015-10-23 | 2016-03-02 | 国家电网公司 | Transformer station equipment infrared temperature registration positioning and method |
CN108765358A (en) * | 2018-05-22 | 2018-11-06 | 烟台艾睿光电科技有限公司 | The double light fusion methods and plug-in type thermal imager system of visible light and infrared light |
US10169681B2 (en) | 2014-12-15 | 2019-01-01 | Koninklijke Philips N.V. | Quality control of image registration |
CN110751620A (en) * | 2019-08-28 | 2020-02-04 | 宁波海上鲜信息技术有限公司 | Method for estimating volume and weight, electronic device, and computer-readable storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1567977A (en) * | 2003-06-27 | 2005-01-19 | 光宝科技股份有限公司 | Automatic correction method for oblique image |
CN102298779A (en) * | 2011-08-16 | 2011-12-28 | 淮安盈科伟力科技有限公司 | Image registering method for panoramic assisted parking system |
CN102567979A (en) * | 2012-01-20 | 2012-07-11 | 南京航空航天大学 | Vehicle-mounted infrared night vision system and multi-source images fusing method thereof |
CN103020956A (en) * | 2012-11-20 | 2013-04-03 | 华中科技大学 | Image matching method for judging Hausdorff distance based on decision |
-
2013
- 2013-07-15 CN CN2013102946761A patent/CN103337080A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1567977A (en) * | 2003-06-27 | 2005-01-19 | 光宝科技股份有限公司 | Automatic correction method for oblique image |
CN102298779A (en) * | 2011-08-16 | 2011-12-28 | 淮安盈科伟力科技有限公司 | Image registering method for panoramic assisted parking system |
CN102567979A (en) * | 2012-01-20 | 2012-07-11 | 南京航空航天大学 | Vehicle-mounted infrared night vision system and multi-source images fusing method thereof |
CN103020956A (en) * | 2012-11-20 | 2013-04-03 | 华中科技大学 | Image matching method for judging Hausdorff distance based on decision |
Non-Patent Citations (2)
Title |
---|
WU JIAN-MING等: "Study on an improved Hausdorff distance for multi-sensor image matching", 《COMMUN NONLINEAR SCI NUMER SIMULAT》, vol. 17, no. 2, 28 February 2012 (2012-02-28), pages 514 - 515 * |
李柏林: "基于特征点图像拼接的配准算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》, no. 08, 15 August 2009 (2009-08-15), pages 138 - 1059 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10169681B2 (en) | 2014-12-15 | 2019-01-01 | Koninklijke Philips N.V. | Quality control of image registration |
CN104966281A (en) * | 2015-04-14 | 2015-10-07 | 中测新图(北京)遥感技术有限责任公司 | IMU/GNSS guiding matching method of multi-view images |
CN104966281B (en) * | 2015-04-14 | 2018-02-02 | 中测新图(北京)遥感技术有限责任公司 | The IMU/GNSS guiding matching process of multi-view images |
CN105371957A (en) * | 2015-10-23 | 2016-03-02 | 国家电网公司 | Transformer station equipment infrared temperature registration positioning and method |
CN108765358A (en) * | 2018-05-22 | 2018-11-06 | 烟台艾睿光电科技有限公司 | The double light fusion methods and plug-in type thermal imager system of visible light and infrared light |
CN110751620A (en) * | 2019-08-28 | 2020-02-04 | 宁波海上鲜信息技术有限公司 | Method for estimating volume and weight, electronic device, and computer-readable storage medium |
CN110751620B (en) * | 2019-08-28 | 2021-03-16 | 宁波海上鲜信息技术有限公司 | Method for estimating volume and weight, electronic device, and computer-readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11244197B2 (en) | Fast and robust multimodal remote sensing image matching method and system | |
Wang et al. | Automated estimation of reinforced precast concrete rebar positions using colored laser scan data | |
CN111222395B (en) | Target detection method and device and electronic equipment | |
CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
CN108470356B (en) | Target object rapid ranging method based on binocular vision | |
CN103389310B (en) | Online sub-pixel optical component damage detection method based on radiation calibration | |
CN107516322B (en) | Image object size and rotation estimation calculation method based on log polar space | |
CN107993258A (en) | A kind of method for registering images and device | |
CN104268853A (en) | Infrared image and visible image registering method | |
CN103337080A (en) | Registration technology of infrared image and visible image based on Hausdorff distance in gradient direction | |
KR20150112656A (en) | Method to calibrate camera and apparatus therefor | |
Yan et al. | Towards automated detection and quantification of concrete cracks using integrated images and lidar data from unmanned aerial vehicles | |
CN102789578A (en) | Infrared remote sensing image change detection method based on multi-source target characteristic support | |
CN101976436A (en) | Pixel-level multi-focus image fusion method based on correction of differential image | |
CN105934757A (en) | Method and apparatus for detecting incorrect associations between keypoints of first image and keypoints of second image | |
Flesia et al. | Sub-pixel straight lines detection for measuring through machine vision | |
CN111563896A (en) | Image processing method for catenary anomaly detection | |
CN103150725B (en) | Based on SUSAN edge detection method and the system of non-local mean | |
KR20180098945A (en) | Method and apparatus for measuring speed of vehicle by using fixed single camera | |
CN104966283A (en) | Imaging layered registering method | |
CN107818583A (en) | Cross searching detection method and device | |
CN105678731A (en) | Identification method of buckling of band steel | |
US9727780B2 (en) | Pedestrian detecting system | |
CN106778822B (en) | Image straight line detection method based on funnel transformation | |
CN110969601B (en) | Structure rotation response non-contact identification method based on visual characteristic tracking algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131002 |