CN102663720A - Image splicing method based on minimum mean square error criterion - Google Patents
Image splicing method based on minimum mean square error criterion Download PDFInfo
- Publication number
- CN102663720A CN102663720A CN2012100929413A CN201210092941A CN102663720A CN 102663720 A CN102663720 A CN 102663720A CN 2012100929413 A CN2012100929413 A CN 2012100929413A CN 201210092941 A CN201210092941 A CN 201210092941A CN 102663720 A CN102663720 A CN 102663720A
- Authority
- CN
- China
- Prior art keywords
- zone
- registration
- pixel
- square error
- area
- 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
Images
Abstract
The invention provides an image splicing method based on a minimum mean square error criterion, relating to a high precision image splicing method, belonging to the field of image processing technology. A purpose of the invention is applying the minimum mean square error criterion to realize the registration of best overlap regions with high accuracy. The method comprises the following steps of: carrying out rough registration; sliding a registration area; calculating epsilon [L*D], sliding an image splicing area around a rough registration splicing range, and calculating the mean square error value epsilon [L*D] of an overlapping area successively; carrying out epsilon [min] updating, updating a large value with a minimum mean square error, and recording a corresponding overlapping area; repeating the operation in a sliding range, wherein when L is not equal to m or D is not equal to n, epsilon [L*D] is larger than epsilon [opt]; when a mean square error of an examine overlapping area shows an obvious valley, expressing an area reaching a minimum pixel difference, and finally using the overlapping area corresponding to the minimum mean square error epsilon [min]= epsilon [opt] as an optimal overlapping area X[M-m:M-1,N-n:n-1] and Y[0:m-1,0:n-1] of registration; and carrying out area ratio Gauss transition. The method is suitable for the image splicing of two-dimensional planar graphs with a same pixel density degree, and is applied to splicing of continuously shot images of a same image acquisition device.
Description
Technical field
The present invention relates to a kind of high precision image joining method, belong to technical field of image processing.
Background technology
Image mosaic is applied to the various fields that need to obtain high resolving power or wide visual angle image as an important branch of image processing techniques.For the current images acquisition technique, can occur owing to the different faint differences of brightness of image that cause of acquisition angles.Therefore in the splicing of two high-definition pictures, can't overlap the zone on the border of two pictures and find the desirable pixel zone that overlaps, therefore stay tangible splicing vestige or crack, influence visual effect.Image mosaic comprises two processes of fusion of the registration and the image of image.The method of the image registration of generally adopting at present is the method for extract minutiae.
Summary of the invention
The purpose of this invention is to provide a kind of image split-joint method, realize the best zone that overlaps of high registration accuracy to use minimum mean square error criterion based on minimum mean square error criterion.The present invention only is directed against and is spliced into the condition proposition between the identical picture of 2 d plane picture and pixel density degree, is applied to the completion splicing to same image capture device continuously shot images.
The present invention solves the problems of the technologies described above the technical scheme of taking to be:
A kind of image split-joint method based on minimum mean square error criterion of the present invention: image X and Y that to set two resolution be M * N splice, and it is best, and to overlap zone be the zone of a m * n; Setting these two picture formats is black and white BMP file, and the pixel span is X
I, j, Y
I, j∈ 0 ..., 255}, (i=0,1 ... M-1, j=0,1 ... N-1); With the image upper left corner is X
0,0, the lower right corner is X
M-1, N-1, then best coincidence zone can be expressed as respectively in X and Y: X
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1
Said method realizes according to following steps:
Step 1, thick registration: the splicing regions of image X and Y is accomplished thick registration, confirm that tentatively image mosaic zone sliding scale is:
Step 2, slip registration region: near slip image mosaic zone thick registration splicing scope;
Step 3, calculating ε
L * D:
Investigate to overlap zone and be (as shown in fig. 1) L * D when big or small, the expression formula of square error is:
Even near slip image mosaic zone thick registration splicing scope is X
M-L, N-DAnd Y
L, DSatisfy:
And
And calculating overlaps regional square mean error amount ε one by one
L * D
Step 4, ε
MinUpgrade: upgrade big value with little value square error, and the corresponding coincidence zone of record; In sliding scale, repeat this operation, when L ≠ m or D ≠ n, ε is arranged
L * D>ε
OptWhen tangible valley appearred in the square error of investigating the coincidence zone, expression arrived the zone of minimum pixel difference, finally with minimum value square error ε
Min=ε
OptCorresponding coincidence zone overlaps regional X for the best of registration
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1
Then when registration region was best coincidence zone, the square error minimum was:
x
M-m+i, N-n+jBe pixel X
M-m+i, N-n+jPixel value, y
I, jBe pixel Y
I, jPixel value;
Step 5, area are than Gauss transition: after confirming the optimal registration zone; Be each the pixel structure Gaussian distribution model corresponding in the registration region with pixel value; Calculate the upper left bottom right area ratio of each pixel in registration overlaps the zone, and confirm the pixel value of pixel than the mode of approaching according to area.
The invention has the beneficial effects as follows:
The present invention mainly stresses to use minimum mean square error criterion and realizes the best zone that overlaps of high registration accuracy.Image mosaic comprises two processes of fusion of the registration and the image of image.Different with the method registration of the extract minutiae that generally adopts at present, the present invention is the location that the image mosaic zone is accomplished in registration decision rule to investigate least mean-square error (MMSE) criterion that overlaps on the region area, avoids the generation in crack; The inventive method is used minimum mean square error criterion and has been realized the best zone that overlaps of high registration accuracy; And the method that adopts transition to merge to the pixel of splicing regions.And then, greatly improved the joining quality of image from visually eliminating the splicing vestige that causes owing to luminance difference XOR difference in exposure.
Description of drawings
Fig. 1 is the splicing synoptic diagram of the inventive method, and Fig. 2 realizes the procedure chart of area than Gauss transition, and Fig. 3 is the inventive method FB(flow block).
Embodiment
Embodiment one: shown in Fig. 1~3, the described a kind of image split-joint method based on minimum mean square error criterion of this embodiment: image X and Y that to set two resolution be M * N splice, and it is best, and to overlap zone be the zone of a m * n; Setting these two picture formats is black and white BMP file, and the pixel span is X
I, j, Y
I, j∈ 0 ..., 255}, (i=0,1 ... M-1, j=0,1 ... N-1); With the image upper left corner is X
0,0, the lower right corner is X
M-1, N-1, then best coincidence zone can be expressed as respectively in X and Y: X
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1
Said method realizes according to following steps:
Step 1, thick registration: the splicing regions of image X and Y is accomplished thick registration, confirm that tentatively image mosaic zone sliding scale is:
Thick registration can be realized through the mode of feature point extraction; The mode of feature point extraction is the prior art category;
Step 2, slip registration region: near slip image mosaic zone thick registration splicing scope;
Step 3, calculating ε
L * D:
Investigate to overlap zone and be (as shown in fig. 1) L * D when big or small, the expression formula of square error is:
Even near slip image mosaic zone thick registration splicing scope is X
M-L, N-DAnd Y
L, DSatisfy:
And
And calculating overlaps regional square mean error amount ε one by one
L * D
Step 4, ε
MinUpgrade: upgrade big value with little value square error, and the corresponding coincidence zone of record; In sliding scale, repeat this operation, when L ≠ m or D ≠ n, ε is arranged
L * D>ε
OptWhen tangible valley appearred in the square error of investigating the coincidence zone, expression arrived the zone of minimum pixel difference, finally with minimum value square error ε
Min=ε
OptCorresponding coincidence zone overlaps regional X for the best of registration
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1
Then when registration region was best coincidence zone, the square error minimum was:
x
M-m+i, N-n+jBe pixel X
M-m+i, N-n+jPixel value, y
I, jBe pixel Y
I, jPixel value;
Step 5, area are than Gauss transition: after confirming the optimal registration zone; Be each the pixel structure Gaussian distribution model corresponding in the registration region with pixel value; Calculate the upper left bottom right area ratio of each pixel in registration overlaps the zone, and confirm the pixel value of pixel than the mode of approaching according to area.
The registration process of step 1 to four image, step 5 are the fusion processs of image.Through investigating the square error situation of change in the sliding area, search the best zone that overlaps, realize high-precision image mosaic.
Embodiment two: as shown in Figure 3, in this embodiment, the described area of step 5 than the detailed process of Gauss transition is:
Step 5 (one), definition overlap regional Z
M * nOn pixel be Z
L, d, the pixel value of corresponding pixel points is respectively x in image X and Y
M-m+l, N-n+dAnd y
L, dThrough pixel Z
L, dAs slope is the separatrix of m/n, and separatrix top left region area is S
X, separatrix lower right area area is S
YCross the separatrix area this moment than being S
X/ S
Y
Step 5 (two), the transition of employing Gauss model make the luminance transition in image mosaic zone more level and smooth, in Gaussian distribution figure, find respective pixel value z
L, d, make z
L, dLeft side area Z
XWith right side area Z
YSatisfy:
Z
X/Z
Y→S
X/S
Y (4)
" → " expression approaches in the formula;
Can confirm to overlap regional Z after step 5 (three), above-mentioned two steps of completion
M * nAt Z
L, dLocational pixel value; The variance of Gaussian distribution can be decided according to the luminance difference that is spliced picture, and the average of Gaussian distribution is μ
z=(x
M-m+l, N-n+d+ y
L, d)/2;
Step 5 (four), repetition above-mentioned steps can confirm that thus registration overlaps the pixel value of all pixels in the zone.Other step is identical with embodiment one.
Principle of work:
Square error is used to weigh receiving terminal to sending the estimated performance of terminal signaling in the communications field.If the transmission terminal signaling is t, receiving terminal is expressed as the estimated value of the sending terminal signaling estimated value square error ε for
receiving end:
Wherein E{} representes the expectation of element or vector.The receiving terminal estimated value is more little with the square error of the signal that sends the terminal, and then demodulating of the easier zero defect of receiving end sent the terminal information sequence.Be that square error is the important indicator that influences the receiving terminal bit error rate.
With the minimum mean square error criterion is the judgment condition of registration, and two width of cloth picture region that are about to registration region are regarded as sending terminal signaling and receiving terminal signal.Move the coincidence zone of two pictures; Investigate the square mean error amount that overlaps two corresponding on region area pixels; When minimum value appears in square mean error amount, represent the difference of two width of cloth pictures on the registration coincidence zone of this moment minimum, therefore judgement is the registration coincidence zone of two pictures.
M * N image X that with two resolution shown in Figure 1 is and Y are spliced into example, and it is best, and to overlap zone be the zone of a m * n.If these two picture formats are black and white BMP file, the pixel span is X
I, j, Y
I, j∈ 0 ..., 255}, (i=0,1 ... M-1, j=0,1 ... N-1).With the image upper left corner is X
0,0, the lower right corner is X
M-1, N-1, then best coincidence zone can be expressed as respectively in X and Y: X
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1Then when registration region was best coincidence zone, the square error minimum was:
x
M-m+i, N-n+jBe pixel X
M-m+i, N-n+jPixel value, y
I, jBe pixel Y
I, jPixel value.When investigate overlapping zone is L * D as shown in fig. 1 when big or small, and the expression formula of square error is:
Then when L ≠ m or D ≠ n, ε is arranged
L * D>ε
Opt, therefore when tangible valley appearred in the square error of investigating the coincidence zone, expression arrived the zone of minimum pixel difference, so registration region is best coincidence zone.
Behind the best coincidence of the registration zone, the pixel that overlaps the zone must the suitable excessive method of pixel of employing owing to luminance difference.Traditional Gauss's class transition algorithm can reasonablely address this problem.Because splicing regions is the two dimension splicing usually, adopt based on area than Gauss's transition method of dividing.More excessive promptly according to the position of pixel in registration region based on area than the Gauss who divides; Size by the registration region rectangle is divided its upper left and lower right area; And in gaussian distribution curve, finding suitable pixel, the left and right area of changing the time is than approaching the upper left area ratio with bottom right pixel point of this pixel in the registration region.With coincidence zone Z shown in Figure 2
M * nOn pixel Z
L, dBe example, this moment, the pixel value of corresponding pixel points in X and Y was respectively x
M-m+l, N-n+dAnd y
L, dCross and be Z
L, dSlope is the separatrix of m/n, and separatrix top left region area is S
X, separatrix lower right area area is S
YCross the separatrix area this moment than being S
X/ S
Y
Excessively more level and smooth for the brightness that makes the image mosaic zone, adopt Gauss model excessive.Then as shown in Figure 2ly can in brachymemma Gaussian distribution figure, find respective pixel value z
L, d, make z
L, dLeft side area Z
XWith right side area Z
YSatisfy:
Z
X/Z
Y→S
X/S
Y (4)
And then confirm at Z
L, dLocational pixel value.The variance of brachymemma Gaussian distribution can be decided according to the luminance difference that is spliced picture, and the average of Gaussian distribution is μ
z=(x
M-m+l, N-n+d+ y
L, d)/2.Can confirm that thus registration overlaps the pixel value of all pixels in the zone.
Claims (2)
1. image split-joint method based on minimum mean square error criterion, image X and Y that to set two resolution be M * N splice, and it is best, and to overlap zone be the zone of a m * n; Setting these two picture formats is black and white BMP file, and the pixel span is X
I, j, Y
I, j∈ 0 ..., 255}, (i=0,1 ... M-1, j=0,1 ... N-1); With the image upper left corner is X
0,0, the lower right corner is X
M-1, N-1, then best coincidence zone can be expressed as respectively in X and Y: X
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1
It is characterized in that: said method realizes according to following steps:
Step 1, thick registration: the splicing regions of image X and Y is accomplished thick registration, confirm that tentatively image mosaic zone sliding scale is:
Step 2, slip registration region: near slip image mosaic zone thick registration splicing scope;
Step 3, calculating ε
L * D:
Investigate to overlap zone and be L * D when big or small, the expression formula of square error is:
Even near slip image mosaic zone thick registration splicing scope is X
M-L, N-DAnd Y
L, DSatisfy:
And
And calculating overlaps regional square mean error amount ε one by one
L * D
Step 4, ε
MinUpgrade: upgrade big value with little value square error, and the corresponding coincidence zone of record; In sliding scale, repeat this operation, when L ≠ m or D ≠ n, ε is arranged
L * D>ε
OptWhen tangible valley appearred in the square error of investigating the coincidence zone, expression arrived the zone of minimum pixel difference, finally with minimum value square error ε
Min=ε
OptCorresponding coincidence zone overlaps regional X for the best of registration
M-m:M-1, N-n:N-1And Y
0:m-1,0:n-1
Then when registration region was best coincidence zone, the square error minimum was:
x
M-m+i, N-n+jBe pixel X
M-m+i, N-n+jPixel value, y
I, jBe pixel Y
I, jPixel value;
Step 5, area are than Gauss transition: after confirming the optimal registration zone; Be each the pixel structure Gaussian distribution model corresponding in the registration region with pixel value; Calculate the upper left bottom right area ratio of each pixel in registration overlaps the zone, and confirm the pixel value of pixel than the mode of approaching according to area.
2. a kind of image split-joint method based on minimum mean square error criterion according to claim 1 is characterized in that:
The described area of step 5 than the detailed process of Gauss transition is:
Step 5 (one), definition overlap regional Z
M * nOn pixel be Z
L, d, the pixel value of corresponding pixel points is respectively x in image X and Y
M-m+l, N-n+dAnd y
L, dThrough pixel Z
L, dAs slope is the separatrix of m/n, and separatrix top left region area is S
X, separatrix lower right area area is S
YCross the separatrix area this moment than being S
X/ S
Y
Step 5 (two), the transition of employing Gauss model make the luminance transition in image mosaic zone more level and smooth, in Gaussian distribution figure, find respective pixel value z
L, d, make z
L, dLeft side area Z
XWith right side area Z
YSatisfy:
Z
X/Z
Y→S
X/S
Y (4)
" → " expression approaches in the formula;
Can confirm to overlap regional Z after step 5 (three), above-mentioned two steps of completion
M * nAt Z
L, dLocational pixel value; The variance of Gaussian distribution can be decided according to the luminance difference that is spliced picture, and the average of Gaussian distribution is μ
z=(x
M-m+l, N-n+d+ y
L, d)/2;
Step 5 (four), repetition above-mentioned steps can confirm that thus registration overlaps the pixel value of all pixels in the zone.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210092941.3A CN102663720B (en) | 2012-03-31 | 2012-03-31 | Image splicing method based on minimum mean square error criterion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210092941.3A CN102663720B (en) | 2012-03-31 | 2012-03-31 | Image splicing method based on minimum mean square error criterion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102663720A true CN102663720A (en) | 2012-09-12 |
CN102663720B CN102663720B (en) | 2014-06-04 |
Family
ID=46773199
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210092941.3A Active CN102663720B (en) | 2012-03-31 | 2012-03-31 | Image splicing method based on minimum mean square error criterion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102663720B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104184950A (en) * | 2014-09-10 | 2014-12-03 | 北京奇艺世纪科技有限公司 | Video image stitching method and device |
CN105554449A (en) * | 2015-12-11 | 2016-05-04 | 浙江宇视科技有限公司 | Method and device for quickly splicing camera images |
CN105844683A (en) * | 2016-03-23 | 2016-08-10 | 深圳市富途网络科技有限公司 | Pixel difference frame-by-frame animation realization method based on Canvas and WebWorker |
CN105976319A (en) * | 2016-05-06 | 2016-09-28 | 安徽伟合电子科技有限公司 | Boundary reproduction method applied to image splicing |
CN108510439A (en) * | 2017-02-28 | 2018-09-07 | 上海小桁网络科技有限公司 | Joining method, device and the terminal of point cloud data |
WO2020037570A1 (en) * | 2018-08-22 | 2020-02-27 | 深圳市真迈生物科技有限公司 | Method and device for image registration, and computer program product |
CN111598177A (en) * | 2020-05-19 | 2020-08-28 | 中国科学院空天信息创新研究院 | Self-adaptive maximum sliding window matching method facing low-overlapping image matching |
US11170506B2 (en) | 2018-08-22 | 2021-11-09 | Genemind Biosciences Company Limited | Method for constructing sequencing template based on image, and base recognition method and device |
US11847766B2 (en) | 2018-08-22 | 2023-12-19 | Genemind Biosciences Company Limited | Method and device for detecting bright spots on image, and computer program product |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6018349A (en) * | 1997-08-01 | 2000-01-25 | Microsoft Corporation | Patch-based alignment method and apparatus for construction of image mosaics |
CN101052100A (en) * | 2007-03-29 | 2007-10-10 | 上海交通大学 | Multiple exposure image intensifying method |
CN101221661A (en) * | 2008-01-29 | 2008-07-16 | 深圳市迅雷网络技术有限公司 | Image registration method and device |
-
2012
- 2012-03-31 CN CN201210092941.3A patent/CN102663720B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6018349A (en) * | 1997-08-01 | 2000-01-25 | Microsoft Corporation | Patch-based alignment method and apparatus for construction of image mosaics |
CN101052100A (en) * | 2007-03-29 | 2007-10-10 | 上海交通大学 | Multiple exposure image intensifying method |
CN101221661A (en) * | 2008-01-29 | 2008-07-16 | 深圳市迅雷网络技术有限公司 | Image registration method and device |
Non-Patent Citations (2)
Title |
---|
程焱等: "《基于SIFT特征遥感影像自动配准与拼接》", 《遥感技术与应用》, vol. 23, no. 6, 31 December 2008 (2008-12-31), pages 721 - 728 * |
谭康: "《图像拼接技术研究与实现》", 《南京理工大学硕士论文》, 31 January 2007 (2007-01-31), pages 1 - 60 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104184950A (en) * | 2014-09-10 | 2014-12-03 | 北京奇艺世纪科技有限公司 | Video image stitching method and device |
CN104184950B (en) * | 2014-09-10 | 2017-12-08 | 北京奇艺世纪科技有限公司 | A kind of video image joining method and device |
CN105554449A (en) * | 2015-12-11 | 2016-05-04 | 浙江宇视科技有限公司 | Method and device for quickly splicing camera images |
CN105554449B (en) * | 2015-12-11 | 2018-04-27 | 浙江宇视科技有限公司 | A kind of method and device for being used to quickly splice camera review |
CN105844683A (en) * | 2016-03-23 | 2016-08-10 | 深圳市富途网络科技有限公司 | Pixel difference frame-by-frame animation realization method based on Canvas and WebWorker |
CN105844683B (en) * | 2016-03-23 | 2019-02-05 | 深圳市富途网络科技有限公司 | The implementation method of pixel difference frame-by-frame cartoons based on Canvas and WebWorker |
CN105976319A (en) * | 2016-05-06 | 2016-09-28 | 安徽伟合电子科技有限公司 | Boundary reproduction method applied to image splicing |
CN108510439A (en) * | 2017-02-28 | 2018-09-07 | 上海小桁网络科技有限公司 | Joining method, device and the terminal of point cloud data |
WO2020037570A1 (en) * | 2018-08-22 | 2020-02-27 | 深圳市真迈生物科技有限公司 | Method and device for image registration, and computer program product |
US11170506B2 (en) | 2018-08-22 | 2021-11-09 | Genemind Biosciences Company Limited | Method for constructing sequencing template based on image, and base recognition method and device |
US11847766B2 (en) | 2018-08-22 | 2023-12-19 | Genemind Biosciences Company Limited | Method and device for detecting bright spots on image, and computer program product |
CN111598177A (en) * | 2020-05-19 | 2020-08-28 | 中国科学院空天信息创新研究院 | Self-adaptive maximum sliding window matching method facing low-overlapping image matching |
Also Published As
Publication number | Publication date |
---|---|
CN102663720B (en) | 2014-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102663720A (en) | Image splicing method based on minimum mean square error criterion | |
US11830222B2 (en) | Bi-level optimization-based infrared and visible light fusion method | |
CN103345755A (en) | Chessboard angular point sub-pixel extraction method based on Harris operator | |
CN108596975A (en) | A kind of Stereo Matching Algorithm for weak texture region | |
CN110889398A (en) | Multi-modal image visibility detection method based on similarity network | |
CN104851089A (en) | Static scene foreground segmentation method and device based on three-dimensional light field | |
CN103293168B (en) | Fruit surface defect detection method based on visual saliency | |
CN105488485A (en) | Lane line automatic extraction method based on vehicle trajectory | |
CN103226831A (en) | Image matching method utilizing block Boolean operation | |
CN103106663B (en) | Realize the method for SIM card defects detection based on image procossing in computer system | |
Qu et al. | CrackT-net: a method of convolutional neural network and transformer for crack segmentation | |
CN111881914B (en) | License plate character segmentation method and system based on self-learning threshold | |
CN113326846A (en) | Rapid bridge apparent disease detection method based on machine vision | |
CN105427299A (en) | Camera focal length solving method based on distortion correction | |
CN106355613B (en) | The method for automatically extracting cross centre of figure based on least square fitting iteration | |
Shit et al. | An encoder‐decoder based CNN architecture using end to end dehaze and detection network for proper image visualization and detection | |
CN102567992B (en) | Image matching method of occluded area | |
CN103473566A (en) | Multi-scale-model-based vehicle detection method | |
CN116523885A (en) | PCB defect detection method based on multi-scale fusion and deep learning | |
CN116630917A (en) | Lane line detection method | |
Chu et al. | Fine‐grained crack segmentation for high‐resolution images via a multiscale cascaded network | |
CN113920317B (en) | Semantic segmentation method based on visible light image and low-resolution depth image | |
CN114926456A (en) | Rail foreign matter detection method based on semi-automatic labeling and improved deep learning | |
Mojidra et al. | Vision-based fatigue crack detection using global motion compensation and video feature tracking | |
CN103970885B (en) | Method for achieving 3D video search engine based on video content analyzing |
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 |