CN103530612B - Fast target detection method based on a small amount of sample - Google Patents
Fast target detection method based on a small amount of sample Download PDFInfo
- Publication number
- CN103530612B CN103530612B CN201310479987.5A CN201310479987A CN103530612B CN 103530612 B CN103530612 B CN 103530612B CN 201310479987 A CN201310479987 A CN 201310479987A CN 103530612 B CN103530612 B CN 103530612B
- Authority
- CN
- China
- Prior art keywords
- hog
- sample
- histogram
- picture
- small amount
- 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
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of fast target detection method based on a small amount of sample, it mainly constructs the voting space about target by the histogram of gradients HOG feature using a small amount of sample, and calculate the HOG feature detecting image under the HOG feature of query image and different scale, the mode using sliding window calculates query image and the histogram of gradients distance of detection image block, thus positions the target in detection picture；Accurately extract finally by the target in the mean shift algorithm detection image to navigating in previous step, thus the detection block overlaped is blended, its procedure is simple, can detect based on less sample simultaneously, and the result degree of accuracy drawn is higher.
Description
Technical field
The invention belongs to computer vision and fuzzy recognition technology field, a kind of apply in picture library and video based on
The design of the fast target detection method of a small amount of sample.
Background technology
Along with the application technology of the high speed development of electronic information technology, image and video has been deep into each of people's live and work
Individual corner, retrieval of such as based on image information certification, safety monitoring and image information data etc., in such application
In scene, it usually needs a certain appearance target in an image or a video is detected and identifies.Due in image or video
Often there is noise, illumination deformation in target, so the detection to a certain target is still that a difficult problem in an image or a video.
Generally, people can select a kind of to train the mode of grader to carry out target detection based on great amount of samples, and such scheme often needs
Will for a certain target collect thousands of sample, and need long time to train grader, mesh to be detected
Mark change, needs again again to collect sample and is trained.In many application scenarios, collecting thousands of training sample is not
In the cards, such as: in the image information verification process of airport security, a passenger often only has two or three photo and basis
People carries out authentication, the most just cannot obtain the thousands of pictures about passenger and verify, but needs
The technology verified is carried out from several photos；And for example, in shopping at network, in client's hands, there is oneself commodity figure interested several
Sheet can not be again trade name, then needs to search out on website these commodity by the way of picture searching, the most just
Corresponding goods picture can only be just searched by these samples；It addition, in safety defense monitoring system, owing to the target occurred is each
Formula various kinds, if by collecting thousands of sample training grader, by being a job the greatest, therefore, it can
The scheme selecting logical too small amount of sample is carried out.During additionally, carry out target detection in an image or a video, in addition it is also necessary to consider speed
Degree problem, i.e. can quickly locate target in image and video, if a searching algorithm is too slow, influences whether relevant art
Popularization and application.
In prior art, " SAR based on the Primal Sketch algorithm figure of Liu Fang of Xian Electronics Science and Technology University et al. invention
As object detection method ", number of patent application is: 2011101028551, and it is right that this cannot realize mainly for object detection method
Different types of made Target carries out the shortcoming detected, and it realizes process and is: 1) former SAR image is used Primal Sketch
Algorithm obtains the line segment aggregate of image result information；2) define and calculate the regularity of all line segments in line segment aggregate and regular ratio
Rate；3) the seed line segment aggregate appreciated for region is determined；4) on the basis of seed line segment, carry out region growing, wrapped
Candidate target region set containing made Target and natural target；5) obtain according to the Feature Selection of line segment in candidate target region
Whole made Target.First this invention calculates SAR image, then uses Primal Sketch algorithm to represent in SAR image
Line segment aggregate, generate seed line segment region by calculating line segment regularity and regular ratio.But the method exists as follows
Shortcoming: 1) feature of the method is line segment based on image, such feature representation is easily subject to the impact of illumination, contrast,
And there is no local feature information；2) speed of service is slower, it is difficult to meet network picture and video frequency searching.
Summary of the invention
The technical problem to be solved is to carry for shortcoming present in target recognition detection method in above-mentioned prior art
Go out a kind of fast target detection method based on a small amount of sample.
The present invention solves its technical problem and the technical scheme is that fast target detection method based on a small amount of sample, specifically wraps
Include:
S1, collect about the sample of target to be detected, and form sample set R；
S2, by the sample scaling collected in step S1 to unified size, calculate the gradient of each samples pictures in sample set R
Rectangular histogram HOG feature, uses the mode of combination of two to calculate two samples pictures respectively the samples pictures in sample set R every
The HOG Histogram distance of individual pixel；
S3, for size, the mean μ of the HOG Histogram distance in each location of pixels tries to achieve described step S2
With variance ∑；
S4, randomly choose from sample set R a samples pictures as inquiry picture Q, calculate its HOG feature；For defeated
The test picture T entered carries out scaling on multiple yardsticks, and calculates its HOG feature respectively on each yardstick；
S5, with inquiry picture Q HOG feature the test picture T under multiple yardsticks is calculated similar in the way of sliding window
Degree, when similarity is more than being then considered when setting threshold value to detect target；
The detection target that S6, merging overlap have calculated is as final testing result.
Further, described HOG Histogram distance computing formula is:Wherein, h1And h2Point
It is not two histogram vectors,WithRepresent histogram vectors h respectively1And h2I-th element.
Further, described step S2 specifically includes:
S21, to each samples pictures use medium filtering mode denoising；
The HOG feature of each samples pictures after S22, the denoising of calculation procedure S21, each picture of each samples pictures after calculating
The all corresponding histogram vectors in element position；
S23, samples pictures to combination of two calculate the HOG Histogram distance in each pixel respectively.
Further, in described step S2, two samples pictures are all three-dimensional matrices after calculating HOG feature, its
HOG feature is designated as m × n × p, and wherein m and n represents the length and width in the plane of HOG feature respectively, and p represents
The length of each pixel position histogram vectors.
Further, described step S4 specifically includes:
S41, to inquiry picture Q use medium filtering mode denoising, calculate its HOG feature；
S42, to input test picture T use medium filtering mode denoising, multiple yardsticks carry out scaling, at each chi
Its HOG feature is calculated respectively on degree.
Further, described step S5 includes:
S51, the mode of employing sliding window calculate phase in inquiry picture Q and the test picture T under an equal amount of a certain yardstick
Image block T with sizeiHOG Histogram distance；
If S52 is inquiry picture Q and image block TiThe HOG Histogram distance described pixel corresponding with step S3 of respective pixel
The absolute value of the difference of the mean μ of the HOG Histogram distance of position is less than the variance of the HOG Histogram distance of respective pixel position
∑, then have a similarity to accumulate TiOn this image block, cumulative TiThe similarity of upper all positions just becomes image block TiWith
The similarity that inquiry picture Q is overall, as image block TiOverall similarity is more than being then considered when setting threshold value to detect target.
Further, in described step S5, inquiry picture Q and image block TiBetween HOG Histogram distance be:Wherein, j ＜ m, k ＜ n, djkFor inquiry picture Q and image block TiBetween jth × k
HOG Histogram distance between rectangular histogram；
Inquiry picture Q and image block TiBetween calculating formula of similarity be:Wherein
μjkRepresent jth × k the element in μ, ∑jkRepresent jth × k the element in ∑, functionDefinition be:
Wherein N (djk,μjk,∑jk) it is gauss of distribution function, should
The average of function is μjk, variance is ∑jk。
Further, described step S6 use mean-shift algorithm merge the detection target that overlap has calculated.
Beneficial effects of the present invention: present invention fast target based on a small amount of sample detection method is by using the ladder of a small amount of sample
Degree rectangular histogram HOG feature constructs the voting space about target, and calculates HOG feature and the different scale of query image
The HOG feature of lower detection image, use sliding window mode calculate query image and detection image block histogram of gradients away from
From, thus the target in detection picture is positioned；Finally by mean-shift algorithm to previous step navigates to
Target in detection image is accurately extracted, thus the detection block overlaped is blended, and its procedure is simple, with
Time can detect based on less sample, the result degree of accuracy drawn is higher.
Accompanying drawing explanation
Fig. 1 is the FB(flow block) of the fast target detection method based on a small amount of sample of the embodiment of the present invention；
Fig. 2 is the testing process block diagram as a example by face；
Fig. 3 is the schematic diagram of calculating using the mode of sliding window to carry out similarity；
Fig. 4 is the testing result schematic diagram as a example by face；
Fig. 5 is the testing result schematic diagram as a example by automobile；
Fig. 6 is the testing result schematic diagram as a example by heart and flower；
Fig. 7 is the testing result schematic diagram of the artificial row with walking.
Detailed description of the invention
The invention will be further elaborated with specific embodiment below in conjunction with the accompanying drawings.
Being illustrated in figure 1 the FB(flow block) of the fast target detection method based on a small amount of sample of the embodiment of the present invention, it specifically wraps
Include,
S1, collect a small amount of sample about target to be detected, and form sample set R；Wherein sample set R should come voluntarily
Data set open and generally acknowledged in the industry or the picture from network search engines, and allow certain deformation；
S2, by the sample scaling collected in step S1 to unified size, calculate the gradient of each samples pictures in sample set R
Rectangular histogram HOG feature, uses the mode of combination of two to calculate two samples pictures respectively the samples pictures in sample set R every
The HOG Histogram distance of individual pixel；
S3, for size, the mean μ of the HOG Histogram distance in each location of pixels tries to achieve described step S2
With variance ∑；
S4, randomly choose from sample set R a samples pictures as inquiry picture Q, calculate its HOG feature；For defeated
The test picture T entered carries out scaling on multiple yardsticks, and calculates its HOG feature respectively on each yardstick；
S5, with inquiry picture Q HOG feature the test picture T under multiple yardsticks is calculated similar in the way of sliding window
Degree, when similarity is more than being then considered when setting threshold value to detect target；
The detection target that S6, merging overlap have calculated is as final testing result；Wherein it is possible to use mean-shift
Algorithm merges overlap to detection target.
Wherein, described HOG Histogram distance computing formula is:Wherein, h1And h2Respectively
It is two histogram vectors,WithRepresent respectively histogram vectors h1And h2Take i-th element respectively.
Described step S2 specifically includes:
S21, to each samples pictures use medium filtering mode denoising；
S22, each samples pictures is calculated HOG feature, each location of pixels of each samples pictures corresponding one after calculating
Individual histogram vectors；
S23, samples pictures to combination of two calculate the HOG Histogram distance in each pixel respectively.
Being all a three-dimensional matrice after two samples pictures calculate HOG feature, its HOG feature is designated as m × n × p, its
Middle m and n represents the length and width in the plane of HOG feature respectively, and p represents each pixel position histogram vectors
Length.In like manner, the inquiry picture Q related in following step and image block TiBetween similarity calculating in, it is also desirable to use
To above-mentioned three-dimensional matrice.
Described step S4 specifically includes:
S41, to inquiry picture Q use medium filtering mode denoising, calculate its HOG feature；
S42, to input test picture T use medium filtering mode denoising, multiple yardsticks carry out scaling, at each chi
Its HOG feature is calculated respectively on degree.
Described step S5 includes:
S51, the mode of employing sliding window calculate phase in inquiry picture Q and the test picture T under an equal amount of a certain yardstick
Image block T with sizeiHOG Histogram distance；
If S52 is inquiry picture Q and image block TiThe HOG Histogram distance described pixel corresponding with step S3 of respective pixel
The absolute value of the difference of the mean μ of the HOG Histogram distance of position is less than the variance of the HOG Histogram distance of respective pixel position
∑, then have a similarity to accumulate TiOn this image block, cumulative TiThe similarity of upper all positions just becomes image block TiWith
The similarity that inquiry picture Q is overall, as image block TiOverall similarity is more than being then considered when setting threshold value to detect target.
Wherein, described step S5 calculates query graph sheet Q and image block TiSimilarity time, need to use in described step S3
Mean μ and variance ∑, wherein mean μ and variance ∑ be respectively size be the matrix of m × n,
Inquiry picture Q and image block TiBetween HOG Histogram distance be:Wherein,
J ＜ m, k ＜ n, djkFor inquiry picture Q and image block TiBetween HOG Histogram distance between jth × k rectangular histogram；
Inquiry picture Q and image block TiBetween calculating formula of similarity be:Wherein
μjkRepresent jth × k the element in μ, ∑jkRepresent jth × k the element in ∑, functionDefinition be:
Wherein N (djk,μjk,∑jk) it is gauss of distribution function, should
The average of function is μjk, variance is ∑jk。
Described step S6 particularly as follows: the detection target location that described step S5 examination is gone out coordinate (x, y) and each Similarity value
It is updated in mean-shift algorithm, by the iteration of limited step, the multiple target frames in same target is fused into one.
In order to skilled artisans appreciate that and implement present invention fast target based on a small amount of sample detection method, will knot
Close specific embodiment the present invention program is described in detail:
We are as a example by Face datection, are illustrated in figure 2 the FB(flow block) of detection process, and the face samples pictures wherein comprised is only
Having 4, the size of samples pictures is 80 × 80 pixels, calculates each after each samples pictures is carried out medium filtering denoising
The HOG feature of sample, each size obtained after having calculated is 10 × 10 × 32；Size to 4 samples pictures
It is the HOG feature combination of two calculating HOG Histogram distance respectively of 10 × 10 × 32, for two samples, calculation
Use the rectangular histograms to 10 × 10 32 dimensions and calculate Histogram distance respectively, calculated and obtained a size afterwards and be
The matrix of 10 × 10, each element representation of matrix is Histogram distance.4 face samples, combination of two, it is possible to obtainThe individual matrix of such 10 × 10.
Obtained the matrix of 6 10 × 10 by 4 face samples, be designated as D respectively1,D2,…,D6, then Histogram distance is equal
Value matrix μ is:
μi,jRepresent the i-th row jth column element of Mean Matrix μ,Represent the i-th row jth column element of kth matrix.Variance
Matrix Σ is calculated as:
When carrying out target detection, from existing four samples, randomly choose a face picture slide as in test picture
Inquiry picture, and calculate its HOG feature, be designated as Q.
Picture to be detected to input, initially with medium filtering, then carries out multiple dimensioned scaling, then calculates on each yardstick
HOG feature.On each yardstick of the picture to be detected of input, the mode of sliding window is used to carry out the calculating of similarity, as
Shown in Fig. 3, the size of window is identical with the size of inquiry picture Q.Each window all calculates a rectangular histogram with inquiry picture Q
Similarity, if rectangular histogram similarity is more than the threshold value of a certain setting, then it is assumed that comprise face in window.To some video in window
Ti, TiCorresponding HOG feature is Hi, HOG feature corresponding for Q is H, HiIt is all the matrix of 10 × 10 × 32 with H,
Then TiAnd rectangular histogram D between QiIt is the matrix of 10 × 10, Q and TiBetween calculating formula of similarity be:
Wherein djkIt is DiJth × k element.
Owing to detection process is sliding window pixel-by-pixel in test picture, so around the same face of test picture the most all
There will be the similarity situation more than threshold value set in advance, therefore after sliding window completes, we pass through mean-shift
These multiple position candidate in same target of algorithm fusion, the input of mean-shift algorithm is that the image of these position candidate is sat
(x, y) and the similarity of this position, this algorithm can be by overlapped candidate frame merging after having run for mark.Wherein mean-shift
Algorithm is the common knowledge of those skilled in the art, is not described in detail in the present patent application scheme.
In order to further illustrate the practicality of the present patent application scheme, this programme additionally lists on the basis of above-mentioned example
The testing result that several different detection targets are completed by method of the present invention, the most as shown in FIG. 4,5,6, 7, its tool
The detection process of body is similar with the detection process of examples detailed above, is not repeated at this.
Those of ordinary skill in the art is it will be appreciated that embodiment described here is to aid in the former of the reader understanding present invention
Reason, it should be understood that protection scope of the present invention is not limited to such special statement and embodiment.The ordinary skill of this area
Personnel can according to these technology disclosed by the invention enlightenment make various other various concrete deformation without departing from essence of the present invention and
Combination, these deformation and combination are the most within the scope of the present invention.
Claims (8)
1. fast target detection method based on a small amount of sample, it is characterised in that specifically include:
S1, collect about the sample of target to be detected, and form sample set R；
S2, by the sample scaling collected in step S1 to unified size, calculate the gradient of each samples pictures in sample set R
Rectangular histogram HOG feature, uses the mode of combination of two to calculate two samples pictures respectively the samples pictures in sample set R every
The HOG Histogram distance of individual pixel；
S3, for size, the mean μ of the HOG Histogram distance in each location of pixels tries to achieve described step S2
With variance ∑；
S4, randomly choose from sample set R a samples pictures as inquiry picture Q, calculate its HOG feature；For defeated
The test picture T entered carries out scaling on multiple yardsticks, and calculates its HOG feature respectively on each yardstick；
S5, with inquiry picture Q HOG feature the test picture T under multiple yardsticks is calculated similar in the way of sliding window
Degree, when similarity is more than being then considered when setting threshold value to detect target；
The detection target that S6, merging overlap have calculated is as final testing result.
2. fast target detection method based on a small amount of sample as claimed in claim 1, it is characterised in that described HOG is straight
Side's map distance computing formula is:Wherein, h1And h2It is respectively two histogram vectors,With
Represent histogram vectors h respectively1And h2I-th element.
3. fast target detection method based on a small amount of sample as claimed in claim 1, it is characterised in that described step S2
Specifically include:
S21, to each samples pictures use medium filtering mode denoising；
The HOG feature of each samples pictures after S22, the denoising of calculation procedure S21, each picture of each samples pictures after calculating
The all corresponding histogram vectors in element position；
S23, samples pictures to combination of two calculate the HOG Histogram distance in each pixel respectively.
4. fast target detection method based on a small amount of sample as claimed in claim 3, it is characterised in that described step S2
In, two samples pictures are all three-dimensional matrices after calculating HOG feature, and its HOG feature is designated as m × n × p, wherein
M and n represents the length and width in the plane of HOG feature respectively, and p represents the length of each pixel position histogram vectors
Degree.
5. fast target detection method based on a small amount of sample as claimed in claim 1, it is characterised in that described step S4
Specifically include:
S41, to inquiry picture Q use medium filtering mode denoising, calculate its HOG feature；
S42, to input test picture T use medium filtering mode denoising, multiple yardsticks carry out scaling, at each chi
Its HOG feature is calculated respectively on degree.
6. fast target detection method based on a small amount of sample as claimed in claim 1, it is characterised in that described step S5
Including:
S51, the mode of employing sliding window calculate phase in inquiry picture Q and the test picture T under an equal amount of a certain yardstick
Image block T with sizeiHOG Histogram distance；
If S52 is inquiry picture Q and image block TiThe HOG Histogram distance described pixel corresponding with step S3 of respective pixel
The absolute value of the difference of the mean μ of the HOG Histogram distance of position is less than the variance of the HOG Histogram distance of respective pixel position
∑, then have a similarity to accumulate TiOn this image block, cumulative TiThe similarity of upper all positions just becomes image block TiWith
The similarity that inquiry picture Q is overall, as image block TiOverall similarity is more than being then considered when setting threshold value to detect target.
7. fast target detection method based on a small amount of sample as claimed in claim 6, it is characterised in that described step S5
In, inquiry picture Q and image block TiBetween HOG Histogram distance be:Wherein,
J ＜ m, k ＜ n, djkFor inquiry picture Q and image block TiBetween HOG Histogram distance between jth × k rectangular histogram；
Inquiry picture Q and image block TiBetween calculating formula of similarity be:Wherein
μjkRepresent jth × k the element in μ, ΣjkRepresent jth × k the element in Σ, functionDefinition be:
Wherein N (djk,μjk,Σjk) it is gauss of distribution function, should
The average of function is μjk, variance is Σjk。
8. the fast target detection method based on a small amount of sample as described in claim 1 to 7 any one claim, its feature
It is, described step S6 uses mean-shift algorithm merge the detection target that overlap has calculated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310479987.5A CN103530612B (en) | 2013-10-14 | 2013-10-14 | Fast target detection method based on a small amount of sample |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310479987.5A CN103530612B (en) | 2013-10-14 | 2013-10-14 | Fast target detection method based on a small amount of sample |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103530612A CN103530612A (en) | 2014-01-22 |
| CN103530612B true CN103530612B (en) | 2016-08-10 |
Family
ID=49932611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310479987.5A Expired - Fee Related CN103530612B (en) | 2013-10-14 | 2013-10-14 | Fast target detection method based on a small amount of sample |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103530612B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104866616B (en) * | 2015-06-07 | 2019-01-22 | 中科院成都信息技术股份有限公司 | Monitor video Target Searching Method |
| CN108932457B (en) * | 2017-05-24 | 2021-09-28 | 腾讯科技（深圳）有限公司 | Image recognition method, device and equipment |
| CN108200406A (en) * | 2018-02-06 | 2018-06-22 | 王辉 | Safety monitoring device |
| CN108763261B (en) * | 2018-04-03 | 2021-08-27 | 南昌奇眸科技有限公司 | Graph retrieval method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268497A (en) * | 2013-06-18 | 2013-08-28 | 厦门大学 | Gesture detecting method for human face and application of gesture detecting method in human face identification |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9922241B2 (en) * | 2011-12-01 | 2018-03-20 | Nokia Technologies Oy | Gesture recognition method, an apparatus and a computer program for the same |
-
2013
- 2013-10-14 CN CN201310479987.5A patent/CN103530612B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268497A (en) * | 2013-06-18 | 2013-08-28 | 厦门大学 | Gesture detecting method for human face and application of gesture detecting method in human face identification |
Non-Patent Citations (2)
| Title |
|---|
| "基于机器视觉的行人快速检测方法研究";薛冠超;《中国优秀硕士学位论文全文数据库（信息科技辑）》;20121015(第10期);第3-4章 * |
| "基于梯度方向直方图的快速人体检测算法";周健;《中国优秀硕士学位论文全文数据库（信息科技辑）》;20111215(第S2期);第4章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103530612A (en) | 2014-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Overview and methods of correlation filter algorithms in object tracking | |
| Li et al. | Road network extraction via deep learning and line integral convolution | |
| CN108875600A (en) | A kind of information of vehicles detection and tracking method, apparatus and computer storage medium based on YOLO | |
| Zhang et al. | Semi-automatic road tracking by template matching and distance transformation in urban areas | |
| Liu et al. | A contrario comparison of local descriptors for change detection in very high spatial resolution satellite images of urban areas | |
| CN103810717B (en) | A kind of human body behavioral value method and device | |
| Abdollahi et al. | Extraction of road features from UAV images using a novel level set segmentation approach | |
| CN102495998B (en) | Static object detection method based on visual selective attention computation module | |
| Hinz et al. | Car detection in aerial thermal images by local and global evidence accumulation | |
| CN105389562A (en) | Secondary optimization method for monitoring video pedestrian re-identification result based on space-time constraint | |
| CN103530612B (en) | Fast target detection method based on a small amount of sample | |
| Yue et al. | Texture extraction for object-oriented classification of high spatial resolution remotely sensed images using a semivariogram | |
| Almagbile | Estimation of crowd density from UAVs images based on corner detection procedures and clustering analysis | |
| CN112541403B (en) | Indoor personnel falling detection method by utilizing infrared camera | |
| Fang et al. | A variational method for multisource remote-sensing image fusion | |
| Liu et al. | Robust multi-scale image matching for deriving ice surface velocity field from sequential satellite images | |
| Martinez-Sanchez et al. | Skyline variations allow estimating distance to trees on landscape photos using semantic segmentation | |
| Abujayyab et al. | Integrating object-based and pixel-based segmentation for building footprint extraction from satellite images | |
| Sun et al. | Knowledge-based automated road network extraction system using multispectral images | |
| CN104200472A (en) | Non-local wavelet information based remote sensing image change detection method | |
| Elouedi et al. | Generalized multi-directional discrete Radon transform | |
| Sun et al. | Synergistic use of Landsat TM and SPOT5 imagery for object-based forest classification | |
| Dey et al. | A robust FLIR target detection employing an auto-convergent pulse coupled neural network | |
| van de Wouw et al. | Real-time change detection for countering improvised explosive devices | |
| Du et al. | The fast lane detection of road using RANSAC 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 | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160810 Termination date: 20161014 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |