CN104318549A - Axial interpolation based registration description sub-direction calculation method - Google Patents
Axial interpolation based registration description sub-direction calculation method Download PDFInfo
- Publication number
- CN104318549A CN104318549A CN201410532061.2A CN201410532061A CN104318549A CN 104318549 A CN104318549 A CN 104318549A CN 201410532061 A CN201410532061 A CN 201410532061A CN 104318549 A CN104318549 A CN 104318549A
- Authority
- CN
- China
- Prior art keywords
- interpolation
- point
- axis
- image
- calculation
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an axial interpolation based registration description sub-direction calculation method and relates to the image data processing field. The axial interpolation based registration description sub-direction calculation method aims at solving the problems that the effect on the image light and view change processing is poor and the accuracy of the calculation on the image rotating degree is low when the image processing is performed by the existing method. Sub-direction factors can be described by the axial interpolation based registration description sub-direction calculation method rapidly, wherein the response to the rotating directions is the same. The real-time registration is performed on an image after the image is obtained through a sensor and image data after the registration are provided for a subsequent image processing algorithm. An image brightness centroid invariant moment of direction vector calculation method is adopted and accordingly the accuracy of the calculation on the comprehensive rotating angle is balanced, the accuracy is slightly improved in comparison with an SURF (Speeded Up Robust Features) method, and the calculation speed is more than four times of SURF direction vectors. The coefficient matrix calculation mode is adopted and accordingly only the multiplication of a coefficient matrix and the image is required during the registration and the parallel processing is easy. The floating-point calculation accuracy is achieved and meanwhile the calculation speed is greatly optimized.
Description
Technical field
The present invention relates to image real time transfer field, be specifically related to moment invariants and calculate, for image registration and image fusion technology.
Background technology
Image registration techniques is image mosaic, image co-registration, target identification, the technical foundation of numerous image processing techniquess such as 3D reconstruction, Postprocessing technique, camera location, computer vision and key link, mainly can be divided into based on the method for gray scale and the large class of the method two of feature based.The latter only calculates feature after extracting feature, the former its calculated amount is less relatively, insensitive to noise, illumination, visual angle and dimensional variation, efficiency of algorithm and registration accuracy high, there is good robustness, thus become the main direction of studying in present image registration field.The algorithm of early stage feature based comprises the methods such as Morvec, Harris.
In May, 2006, the people such as Bay [1] propose famous SURF (speeded up robust features) algorithm, and its comprehensive matching effect is suitable with SIFT algorithm, and has increased substantially registration speed.The computing velocity of SURF algorithm can be faster than SIFT 3 times, and it can maintain the invariance in the change such as, illumination flexible to the rotation of image, yardstick, visual angle, but when processing image irradiation and visual angle change not as SIFT algorithm.Devise the direction vector computing method based on brightness of image barycenter not bending moment herein, the method higher to the computational accuracy rotated, computing velocity is more than 4 times of SURF direction vector.
Summary of the invention
The present invention solves to exist when adopting existing method process image to the illumination of image and visual angle change poor processing effect and to problems such as the swing computational accuracy of image are low, provides a kind of high precision based on the registration descriptor direction calculating method of axial interpolation.
Step one, select radius to be the circular image regions of r, the center of circle of described circular image regions is as the center of unique point, and described unique point centre coordinate is O (x
0, y
0), be that the circumference of r is divided into one section of circular arc at interval of π/4 from 0 ° by described radius, then the length of X-axis corresponding to circular arc is cx, and the length of corresponding Y-axis is cy, cx>cy, and above-mentioned interval is called X-axis interpolation section; Cx<cy, above-mentioned interval is called Y-axis interpolation section, in X-axis interpolation section, adopts X-direction interpolation calculation, at Y-axis interpolation section, adopts Y direction interpolation calculation, obtains whole interpolation points circumferentially;
Step 2, calculate the interpolation coefficient of circumferentially adjacent interpolation point, in X-axis interpolation section, interpolation point I (x
i, y
i) by two consecutive point
with
interpolation forms,
with
be respectively described consecutive point
with
corresponding I (x
i, y
i) interpolation coefficient; In Y-axis interpolation section, I (x
i, y
i) by consecutive point
with
interpolation forms,
with
be respectively the corresponding I (x of these two consecutive point
i, y
i) interpolation coefficient;
Step 3, consecutive point that the interpolation coefficient of the adjacent interpolation point obtained in step 2 is added to calculate, be that the matrix of coefficients of the circular image regions of r is set as that the length of side is the square array of 2*r+1 to described radius, this matrix and the pixel one_to_one corresponding in the square area centered by unique point
First, all elements in matrix of coefficients is set to 0; Then 1 is set to for the coefficient that the pixel in circular image regions is corresponding; Corresponding to all interpolation points finally step 2 obtained, the interpolation coefficient of axial consecutive point is added on the corresponding pixel points position of matrix of coefficients, obtains the interpolation coefficient matrix of image;
Step 4, definition image moment formula are
k
xyfor (x, y) puts final interpolation coefficient, I (x, y) is the brightness that (x, y) puts, and p, q be x respectively, and the exponent number that y-axis is corresponding, works as p=1, during q=0,
work as p=0, during q=1,
work as p=0, during q=0,
then its intensity centroid is:
the vector in representative image direction is set up from unique point O to intensity centroid C
then vector
represent the direction of this unique point, adopt the folder tangent of an angle of this vector and X-axis to represent:
final acquisition descriptor direction.
Beneficial effect of the present invention:
One, The present invention gives a kind of direction vector computing method based on intensity centroid not bending moment, its computing velocity is 4 times of SURF algorithm, and computational accuracy also can meet or exceed SURF algorithm, and effect as shown in Figure 2.
Two, the account form of the matrix of coefficients of the present invention's employing, thinking is provided for optimizing intensity centroid Invariant Moment Method further, namely intensity centroid can be calculated and image interpolation etc. calculates the form being all optimized to matrix of coefficients and calculates before registration, only need complete the multiplication of matrix of coefficients and image during registration, and be easy to parallel processing.This mode, while having floating type computational accuracy, greatly optimizes computing velocity.
Accompanying drawing explanation
Fig. 1 is the registration hardware structural drawing of the registration descriptor direction calculating method based on axial interpolation of the present invention;
Fig. 2 is of the present invention is the axial method of interpolation schematic diagram of 9 based on radius in the registration descriptor direction calculating method of axial interpolation;
Fig. 3 is the registration experiment effect figure adopting the registration descriptor direction calculating method based on axial interpolation of the present invention.
Embodiment
Embodiment one, composition graphs 1 to Fig. 3 illustrate present embodiment, propose in present embodiment a kind of energy fast, the method for high precision computation descriptor direction vector, after sensor obtains image, in real time registration is carried out to image, for successive image Processing Algorithm provides the view data after registration.
Based on the registration descriptor direction calculating method of axial interpolation, the method is realized by following steps:
The calculating in step one, unique point direction adopts usually centered by unique point, and r is the circular image panel region of radius, and r can choose according to actual needs, and each pixel is regarded as a point without size by this method.If image sheet center pixel (i.e. unique point) coordinate is O (x
0, y
0), be then that the whole circumference of r is divided into one section of circular arc at interval of π/4 from 0 ° by radius, then the length of X-axis corresponding to circular arc is cx, the length of corresponding Y-axis is cy, then at (π/4, pi/2), (pi/2,3 π/4), (5 π/4,3 pi/2s) and (3 pi/2s, 7 π/4) interval interior cx>cy, above-mentioned interval is called X-axis interpolation section; At (0, π/4), (3 π/4, π), (π, 5 π/4) and (7 π/4,2 π) interval interior cx<cy, above-mentioned interval is called Y-axis interpolation section.In X-axis interpolation section, adopt X-direction interpolation calculation, first use formula (1) to obtain all interpolation point horizontal ordinates, then use formula (2) to try to achieve the ordinate of respective point; In like manner, at Y-axis interpolation section, adopt Y direction interpolation calculation, first use formula (3) to obtain all interpolation point ordinates, then use formula (4) to try to achieve the horizontal ordinate of respective point. obtain whole interpolation points circumferentially in this manner.
represent and round downwards.(1)
represent and round downwards.(3)
Step 2, calculate the interpolation coefficient of circumferentially adjacent interpolation point, in X-axis interpolation section, interpolation point I (x
i, y
i) by two consecutive point
with
interpolation forms,
with
be respectively described consecutive point
with
corresponding
interpolation coefficient; In Y-axis interpolation section, I (x
i, y
i) by consecutive point
with
interpolation forms,
with
be respectively the corresponding I (x of these two consecutive point
i, y
i) interpolation coefficient;
Described brightness of circumferentially putting corresponding to X-axis interpolation section, adopts formula (5) to calculate, the brightness of circumferentially putting corresponding to Y-axis interpolation section, adopts formula (6) to calculate, i.e. I (x
i, y
i) by two consecutive point
with
interpolation forms,
with
be respectively the corresponding I (x of these two consecutive point
i, y
i) interpolation coefficient.
The mode that the design of step 3, this method precalculates matrix of coefficients accelerates Interpolation Process, i.e. the brightness of not actual computation interpolation point, but the consecutive point that are added to by the interpolation coefficient of its consecutive point calculate, and doing so avoids interpolation operation.The computing method of matrix of coefficients are as follows: pair radius is the image sheet of r, square array, this matrix and the pixel one_to_one corresponding in the square area centered by unique point of its matrix of coefficients to be the length of side be 2*r+1.First, all elements in matrix of coefficients is set to 0, then for circle territory in pixel (namely meet ((x, y) | (x-x
0)
2+ (y-y
0)
2≤ r
2and x, y ∈ R) a little) corresponding coefficient is set to 1.Corresponding to all interpolation points finally step 2 obtained, the interpolation coefficient of axial consecutive point is added on the corresponding pixel points position of matrix of coefficients, namely obtains the interpolation coefficient matrix of this image sheet.Definition image moment formula is
k
xyfor (x, y) puts final interpolation coefficient, I (x, y) is the brightness that (x, y) puts, and p, q be x respectively, and the exponent number that y-axis is corresponding, p+q represents the exponent number of image moment.Work as p=1, during q=0,
work as p=0, during q=1,
work as p=0, during q=0,
then its intensity centroid is:
the vector in a representative graph photo direction is set up from unique point O to intensity centroid C
then vector
the direction of this unique point can be represented, the folder tangent of an angle of this vector and X-axis can be adopted to represent:
wherein k
xyy, k
xyx item all can calculate before registration.
Method described in present embodiment is a part for the method for registering images of feature based, and the method for registering images of feature based mainly comprises the structure of multi-scale image, the selection of unique point, the calculating in unique point direction and the generation of descriptor.The calculation stages in this method mainly distinguished point based direction.
One, the registration hardware environment that present embodiment adopts adopts CPU, 4GB internal memory of Intel4960X, 3.6GHz and the operating system of XP SP3.Capture card adopts the CLSAS video processing board-card of Canadian IO Industries company, the maximum picking rate of this capture card can reach 450MBps, two data stream (camera 1 and camera 2) can be gathered simultaneously, capture card is connected to the SAS hyperdisk array of 4TB, main frame Graphics Processing is uploaded to, as shown in Figure 1 by pci bus partial data.
Two, first build integrogram and pyramid, the first floor adopts the Hession matrix filter of 9*9 size to carry out filtering to image, and build five groups of Octaves, often group is formed by four layers, and each layer wave filter size is as shown in table 1.Table 1 is pyramid construction parameter list, this ensure that except third and fourth layer except the ground floor of first group and the 5th group, every tomographic image just participates in a feature point detection as middle layer and calculates, and then second and third layer often organized is carried out to the non-maxima suppression of 3*3*3 neighborhood.Threshold method is utilized to obtain unique point set according to the screening of unique point second order gradient.
Table 1
| ? | First | Second | Third | Forth |
| Oct1 | 9 | 15 | 21 | 27 |
| Oct2 | 15 | 27 | 39 | 51 |
| Oct3 | 27 | 51 | 75 | 99 |
| Oct4 | 51 | 99 | 147 | 195 |
| Oct5 | 99 | 195 | 291 | 387 |
Three, after obtaining unique point, apply method of the present invention and calculate unique point direction.For the round territory of different radii, the degree of uniformity of its circumferentially pixel distribution is not identical yet.In the round territory of radius 2 to 10, radius be 6 and 9 circumference pixel distribution comparatively even, this method adopt radius be 9 calculating unique point direction, round territory, it is (0 years old, pi/2) axial interpolation point as shown in Figure 2, radius be 9 this axial interpolation method matrix of coefficients as shown in table 2.Table 2 to be radiuses be 9 axial method of interpolation matrix of coefficients.
Table 2
| OK | -9 row---+9 row |
| 9 | 0,0,0,0,0,0.0623,0.4853,0.775,0.9443,1,0.9443,0.775,0.4853,0.0623,0,0,0,0,0 |
| 8 | 0,0,0,0,0.4833,1.9377,1.5147,1.225,1.0557,1,1.0557,1.225,1.5147,1.9377,0.4833,0,0,0,0 |
| 7 | 0,0,0,0.7082,1.5167,1,1,1,1,1,1,1,1,1,1.5167,0.7082,0,0,0 |
| 6 | 0,0,0.7082,1.5836,1,1,1,1,1,1,1,1,1,1,1,1.5836,0.7082,0,0 |
| 5 | 0,0.4833,1.5167,1,1,1,1,1,1,1,1,1,1,1,1,1,1.5167,0.4833,0 |
| 4 | 0.0623,1.9377,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.9377,0.0623 |
| 3 | 0.4853,1.5147,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.5147,0.4853 |
| 2 | 0.7750,1.2250,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.2250,0.7750 |
| 1 | 0.9443,1.0557,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.0557,0.9443 |
| 0 | 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1 |
| 1 | 0.9443,1.0557,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.0557,0.9443 |
| 2 | 0.7750,1.2250,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.2250,0.7750 |
| 3 | 0.4853,1.5147,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.5147,0.4853 |
| 4 | 0.0623,1.9377,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1.9377,0.0623 |
| 5 | 0,0.4833,1.5167,1,1,1,1,1,1,1,1,1,1,1,1,1,1.5167,0.4833,0 |
| 6 | 0,0,0.7082,1.5836,1,1,1,1,1,1,1,1,1,1,1,1.5836,0.7082,0,0 |
| 7 | 0,0,0,0.7082,1.5167,1,1,1,1,1,1,1,1,1,1.5167,0.7082,0,0,0 |
| 8 | 0,0,0,0,0.4833,1.9377,1.5147,1.225,1.0557,1,1.0557,1.225,1.5147,1.9377,0.4833,0,0,0,0 |
| 9 | 0,0,0,0,0,0.0623,0.4853,0.775,0.9443,1,0.9443,0.775,0.4853,0.0623,0,0,0,0,0 |
Four, after obtaining direction vector, adopt the SURF method of H.Bay to extract 64 descriptors tieed up based on the little wave response of Haar and carry out image registration, registration result is shown in Fig. 3.Experiment shows, the direction vector computing time adopting this axial method of interpolation is approximately 1/4 of SURF method.In registration repeatability, the performance of axial method of interpolation in some angle is more than SURF algorithm.
Claims (2)
1., based on the registration descriptor direction calculating method of axial interpolation, it is characterized in that, the method is realized by following steps:
Step one, select radius to be the circular image regions of r, the center of circle of described circular image regions is as the center of unique point, and described unique point centre coordinate is O (x
0, y
0), be that the circumference of r is divided into one section of circular arc at interval of π/4 from 0 ° by described radius, then the length of X-axis corresponding to circular arc is cx, and the length of corresponding Y-axis is cy, cx>cy, and above-mentioned interval is called X-axis interpolation section; Cx<cy, above-mentioned interval is called Y-axis interpolation section, in X-axis interpolation section, adopts X-direction interpolation calculation, at Y-axis interpolation section, adopts Y direction interpolation calculation, obtains whole interpolation points circumferentially;
Step 2, calculate the interpolation coefficient of circumferentially adjacent interpolation point, in X-axis interpolation section, interpolation point I (x
i, y
i) by two consecutive point
with
interpolation forms,
with
be respectively described consecutive point
with
corresponding
interpolation coefficient; In Y-axis interpolation section, I (xi, yi) is by consecutive point
with
interpolation forms,
with
be respectively the corresponding I (x of these two consecutive point
i, y
i) interpolation coefficient;
Step 3, consecutive point that the interpolation coefficient of the adjacent interpolation point obtained in step 2 is added to calculate, be that the matrix of coefficients of the circular image regions of r is set as that the length of side is the square array of 2*r+1 to described radius, this matrix and the pixel one_to_one corresponding in the square area centered by unique point
First, all elements in matrix of coefficients is set to 0; Then 1 is set to for the coefficient that the pixel in circular image regions is corresponding; Corresponding to all interpolation points finally step 2 obtained, the interpolation coefficient of axial consecutive point is added on the corresponding pixel points position of matrix of coefficients, obtains the interpolation coefficient matrix of image;
Step 4, definition image moment formula are
k
xyfor (x, y) puts final interpolation coefficient, I (x, y) is the brightness that (x, y) puts, and p, q be x respectively, and the exponent number that y-axis is corresponding, works as p=1, during q=0,
Work as p=0, during q=1,
Work as p=0, during q=0,
then its intensity centroid is:
the vector in representative image direction is set up from unique point O to intensity centroid C
then vector
represent the direction of this unique point, adopt the folder tangent of an angle of this vector and X-axis to represent:
final acquisition descriptor direction.
2. the registration descriptor direction calculating method based on axial interpolation according to claim 1, it is characterized in that, be that the whole circumference of r is divided into one section of circular arc at interval of π/4 from 0 ° by radius, then X-axis interpolation section is (π/4, pi/2), (pi/2,3 π/4), (5 π/4,3 pi/2s) and (3 pi/2s, 7 π/4), be (0, π/4), (3 π/4, π), (π at Y-axis interpolation section, 5 π/4) and (7 π/4,2 π);
In X-axis interpolation section, adopt X-direction interpolation calculation, first obtain all interpolation point horizontal ordinates with formula one, then try to achieve the ordinate of respective point with formula two;
At Y-axis interpolation section, adopt Y direction interpolation calculation, first obtain all interpolation point ordinates with formula three, then try to achieve the horizontal ordinate of respective point with formula four, until obtain whole interpolation points circumferentially;
Formula one, x
i=k ± x
0,
represent and round downwards;
Formula two,
Formula three, y
i=k ± y
0,
Formula four,
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410532061.2A CN104318549B (en) | 2014-10-10 | 2014-10-10 | A kind of method for describing sub- direction for registration of the determination based on axial interpolation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410532061.2A CN104318549B (en) | 2014-10-10 | 2014-10-10 | A kind of method for describing sub- direction for registration of the determination based on axial interpolation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104318549A true CN104318549A (en) | 2015-01-28 |
| CN104318549B CN104318549B (en) | 2017-07-11 |
Family
ID=52373775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410532061.2A Expired - Fee Related CN104318549B (en) | 2014-10-10 | 2014-10-10 | A kind of method for describing sub- direction for registration of the determination based on axial interpolation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104318549B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090010539A1 (en) * | 2007-07-03 | 2009-01-08 | Stmicroelectronics S.R.L. | Method and relative device of color interpolation of an image acquired by a digital color sensor |
| WO2011006781A1 (en) * | 2009-07-15 | 2011-01-20 | Trident Microsystems, Inc. | 3d image processing |
| CN103997609A (en) * | 2014-06-12 | 2014-08-20 | 四川川大智胜软件股份有限公司 | Multi-video real-time panoramic fusion splicing method based on CUDA |
-
2014
- 2014-10-10 CN CN201410532061.2A patent/CN104318549B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090010539A1 (en) * | 2007-07-03 | 2009-01-08 | Stmicroelectronics S.R.L. | Method and relative device of color interpolation of an image acquired by a digital color sensor |
| WO2011006781A1 (en) * | 2009-07-15 | 2011-01-20 | Trident Microsystems, Inc. | 3d image processing |
| CN103997609A (en) * | 2014-06-12 | 2014-08-20 | 四川川大智胜软件股份有限公司 | Multi-video real-time panoramic fusion splicing method based on CUDA |
Non-Patent Citations (4)
| Title |
|---|
| ETHAN RUBLEE 等: "ORB: an efficient alternative to SIFT or SURF", 《2011 INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| NARESH MARTURI 等: "Fast Image Drift Compensation in Scanning Electron Microscope Using Image Registration", 《2013 IEEE INTERNATIONAL CONFERENCE ON AUTOMATION SCIENCE AND ENGINEERING（CASE）》 * |
| 李光鑫 等: "基于亮度-对比度传递技术的彩色图像融合算法", 《中国光学学会2010年光学大会论文集》 * |
| 符祥 等: "图像插值技术综述", 《计算机工程与设计》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104318549B (en) | 2017-07-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104298990B (en) | A kind of Fast Graphics matching based on skeleton drawing is with knowing method for distinguishing | |
| CN103839265B (en) | SAR image registration method based on SIFT and normalized mutual information | |
| CN103456022B (en) | A kind of high-resolution remote sensing image feature matching method | |
| CN107103323A (en) | A kind of target identification method based on image outline feature | |
| Chen et al. | Robust affine-invariant line matching for high resolution remote sensing images | |
| CN103727930A (en) | Edge-matching-based relative pose calibration method of laser range finder and camera | |
| CN105809173B (en) | A kind of image RSTN invariable attribute feature extraction and recognition methods based on bionical object visual transform | |
| CN104008379A (en) | Object recognition method based on surf | |
| CN107180436A (en) | A kind of improved KAZE image matching algorithms | |
| Fan et al. | Registration of multiresolution remote sensing images based on L2-siamese model | |
| CN104834931A (en) | Improved SIFT algorithm based on wavelet transformation | |
| CN102509293B (en) | Method for detecting consistency of different-source images | |
| CN110222661B (en) | Feature extraction method for moving target identification and tracking | |
| CN108182705A (en) | A kind of three-dimensional coordinate localization method based on machine vision | |
| CN111199558A (en) | Image matching method based on deep learning | |
| Zhang et al. | GPU-accelerated large-size VHR images registration via coarse-to-fine matching | |
| Qi et al. | Exploring illumination robust descriptors for human epithelial type 2 cell classification | |
| Pei et al. | Fast normalized cross-correlation image matching based on multiscale edge information | |
| CN111898589B (en) | Unmanned aerial vehicle image rapid registration method based on GPU+feature recognition | |
| Gao et al. | Fast corner detection using approximate form of second-order Gaussian directional derivative | |
| CN105551048A (en) | Space surface patch-based three-dimensional corner detection method | |
| Kang et al. | Image registration based on harris corner and mutual information | |
| Dong et al. | Superpixel-based local features for image matching | |
| Wang et al. | Unified detection of skewed rotation, reflection and translation symmetries from affine invariant contour features | |
| CN104318549A (en) | Axial interpolation based registration description sub-direction calculation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| 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: 20170711 Termination date: 20181010 |