CN106815824B - A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency - Google Patents
A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency Download PDFInfo
- Publication number
- CN106815824B CN106815824B CN201611124130.1A CN201611124130A CN106815824B CN 106815824 B CN106815824 B CN 106815824B CN 201611124130 A CN201611124130 A CN 201611124130A CN 106815824 B CN106815824 B CN 106815824B
- Authority
- CN
- China
- Prior art keywords
- image
- image pair
- pair
- characteristic point
- histogram
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000005457 optimization Methods 0.000 title claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims abstract description 22
- 238000012795 verification Methods 0.000 claims abstract description 13
- 238000001914 filtration Methods 0.000 claims abstract description 12
- 230000008859 change Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 238000004321 preservation Methods 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 2
- 230000006872 improvement Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 201000011243 gastrointestinal stromal tumor Diseases 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration using histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image neighbour's optimization methods for improving extensive three-dimensional reconstruction efficiency, obtain neighborhood matching image pair first, to image to progress Feature Points Matching;Geometry verification is carried out by basis matrix to reject and be unsatisfactory for the error hiding of epipolar-line constraint and obtain interior points, and is calculated homography matrix and obtained homograph rate;Then the variation between statistical match characteristic point on direction and scale, obtains corresponding histogram;To measure the similitude of image and redundant image therein is marked by the triple constraints of the variation histogram in interior points, homograph rate, scale and direction;The image pair comprising redundant image is rejected, and narrow baseline image pair is rejected by interior points, homograph rate;The image after filtering finally is saved to match information, redundant image pair and the relatively narrow image pair of baseline has been eliminated, has further improved the precision and efficiency of subsequent three-dimensional reconstruction.
Description
Technical field
The invention belongs to computer vision fields, more particularly, to a kind of figure for improving extensive three-dimensional reconstruction efficiency
As neighbour's optimization method.
Background technique
The three-dimensional scenic of extensive pictures is the popular research field of a comparison in recent years.Three-dimensional reconstruction is current
Algorithm generally is exercise recovery structure (Structure from Motion, SFM) algorithm of increment type, it is main comprising with
Lower four parts: 1) picture feature point extracts, 2) characteristic matching between image, 3) to images match to carrying out geometry verification,
4) according to matching estimation camera posture and sparse three-dimensional point cloud.For large-scale dataset, critical issue is efficiency.It presses
According to process above, the bottleneck of efficiency of algorithm mainly appears on second step and third step at present, wherein the original mode of second step
It is to be matched two-by-two, but for large-scale dataset, most of picture is that no scene is overlapped, is incoherent,
If these pictures, which carry out matching, will waste a large amount of time.Therefore the improvement for second step, main approach is exactly to pass through
Certain high efficiency mode approximatively finds the image pair for having scene to be overlapped, to reduce subsequent match time.Change in this respect
Into space it is very big, in fact many scholars are exactly to do a lot of work in this respect.Frahm GIST a kind of to image zooming-out is special
Sign, and clustered according to the similitude of this feature, presentation graphics therein are found, to reduce image pair.Agarwal
A words tree is obtained by characteristics of image training, the close of each image is found with a kind of Image Retrieval Mechanism by this tree
Neighbour, matching only carry out between neighbour.Chao has done a kind of improvement on the basis of Agarwal works, with the method for on-line study
To be ranked up to improve the accuracy of neighbour to search result.Wu is matched to obtain figure with the characteristic point after down-sampled
Similarity as between, to find image pair.Each characteristic point in picture is mapped to a trained spy by Havlena
Some word of dictionary is levied, the characteristic point of the corresponding same word is to match, and all pictures are all matched together.
For third step, the basis matrix F between image is mainly estimated in geometry verification, then carries out pole to matching
Line checksum filter error hiding.Many people propose efficient RANSAC modified version to estimate basis matrix to improve efficiency, separately
Outer Raguram proposes a kind of mode of on-line study to improve the efficiency of geometry verification.
Since the bottleneck of second step will lead to the efficiency entire lowering of follow-up process, many work at present is concentrated on to the
The improvement of two steps.Wherein more classical method is exactly to train to obtain a words tree by the SIFT feature of picture
(vocabulary tree) finds k neighbour of every picture according to this words tree, this k neighbour is the picture
Matching image.Although this method can preferably reject some invalid images pair, the neighbour wherein found is also contained
Some problems: the 1) image of redundancy, it is assumed that a kind of extreme situation (is equivalent to same same width picture reproduction many times
A position, many pictures of the same angle shot), then neighbour will be constituted between these pictures, but these pictures it
Between match information be nonsensical.In addition the neighbour that other pictures are found may be the duplicate of same picture,
The matching of these neighbours contains many redundancies, wastes many times.2) the very narrow neighbour of baseline, baseline are too narrow
Decline of the image to will lead to reconstruction precision.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, extensive three-dimensional reconstruction effect is improved the present invention provides a kind of
Image neighbour's optimization method of rate, this method is on the basis of progress image neighbor search obtains initial pictures pair, by feature
Point matching is verified by geometry, the scale between homograph rate and characteristic point and the triple constraints of direction change are come close to image
Neighbour is further optimized, and the precision and efficiency of subsequent three-dimensional reconstruction are substantially increased.Thus it solves in the prior art to warp
Cross existing redundancy after image neighbor search and the relatively narrow technical problem of baseline.
To achieve the above object, according to one aspect of the present invention, a kind of extensive three-dimensional reconstruction efficiency of raising is provided
Image neighbour's optimization method, comprising:
(1) inceptive filtering of feature point extraction and image pair: characteristic point is extracted to each image, then uses image retrieval
Method find the neighbour of each image, obtain the image pair of scene overlapping;
(2) Feature Points Matching and related data prepare: for each image pair, carrying out characteristic point to the image of image pair
Matching carries out geometry verification by the basis matrix between image pair image to reject error hiding, obtains meeting basis matrix
Feature Points Matching number, as in points, and calculate optimal homography matrix according to matched characteristic point and obtain homograph rate,
Then variation of the characteristic point of statistical match on direction and scale obtains the variation histogram of the variation histogram and scale in direction
Figure;
(3) mark redundant image: by the interior points of each image pair, homograph rate, direction variation histogram
And similarity is met default rule to measure the similarity between each image pair image by the triple constraints of variation histogram of scale
The image of all image pairs then forms multiple images set, chooses an image in each image collection as representative figure
Picture, marking other images is redundant image;
(4) further filtering optimizes image pair: rejecting the image pair comprising redundant image, and passes through interior points, Dan Ying
Interconversion rate rejects narrow baseline image pair;
(5) preservation of information after optimizing: between the image pair and image pair image obtained after filtering is saved
With information.
Preferably, step (2) specifically includes following sub-step:
(2.1) for each image pair, the characteristic point between image pair image is carried out using Feature Points Matching algorithm
Matching;
(2.2) geometry verification is carried out by the basis matrix F between estimation image pair image, rejects error hiding, obtains
It is the interior m that counts to the Feature Points Matching number for meeting F;
(2.3) according to meeting the matching characteristic point of F to calculate optimal homography matrix H and obtain corresponding homograph rate h,
Wherein, h is used to measure the baseline width between image pair image;
(2.4) direction for asking difference operation to obtain characteristic point is carried out on the direction o of characteristic point to the matching characteristic point for meeting F
Changes delta o on o, on the scale s of characteristic point carry out division arithmetic obtain characteristic point scale s on changes delta s, then from
Dispersion obtains the histogram of Δ o and the histogram of Δ s.
Preferably, step (3) specifically includes following sub-step:
(3.1) for each image pair, if the interior points m of image pair is greater than threshold alpha and homograph rate h is greater than threshold value
β thens follow the steps (3.2);
(3.2) if peak value p of the Δ s histogram of image pair near 1 is greater than the top of threshold value η and Δ o histogram
Value q is greater than threshold θ, then by the image of the image pair alternately redundancy;
(3.3) all alternative redundancys are formed into independent multiple images set, for each image collection, only retaining should
For a most image of matching number as representative image, remaining image tagged is redundant image in image collection.
Preferably, step (4) specifically includes following sub-step:
(4.1) image comprising redundant image is rejected if image pair includes redundant image for each image pair
It is right, it is no to then follow the steps (4.2);
(4.2) if the interior points m of image pair is greater than threshold gamma and homograph rate h is less than threshold value δ, retain the image
It is right, otherwise reject the image pair;
(4.3) judge whether that traversal completes all images pair and thens follow the steps (4.1) if not completing, otherwise terminate
Process.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, mainly have skill below
Art advantage:
(1) a kind of new method of measurement image similarity is proposed to be filtered to the excessively high redundant image of similitude,
Image is solved to the Redundancy by still remaining after traditional image neighbor search.
(2) it combines geometry verification and homograph rate to reject narrow baseline image pair, solves image to by traditional figure
As the relatively narrow problem of the baseline still remained after neighbor search.Almost without time complexity is increased, significantly
Improve the precision and efficiency of subsequent three-dimensional reconstruction.
Detailed description of the invention
Fig. 1 is a kind of image neighbour's optimization method for improving extensive three-dimensional reconstruction efficiency disclosed by the embodiments of the present invention
Flow diagram;
Fig. 2 is a kind of flow diagram for marking redundant image disclosed by the embodiments of the present invention;
Fig. 3 is the scale of two nonredundancy images of one kind and the specific embodiment of direction change histogram;
Fig. 4 is the scale of two redundant images of one kind and the specific embodiment of direction change histogram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
This method is related to feature point extraction, and large-scale image is to filtering, quick hash Feature Points Matching algorithm, geometry school
It tests, the calculating of homograph rate, the building in direction and change of scale histogram between matching characteristic point, a kind of new measurement image
The method of similitude, a kind of further redundant image of rejecting is to the technologies such as method with narrow baseline image pair, the image pair of optimization
Saving with corresponding match information can be directly used for subsequent three-dimensional reconstruction process, optimized reconstruction precision and effect at the form of file
Rate.
Fig. 1 is a kind of image neighbour's optimization method for improving extensive three-dimensional reconstruction efficiency disclosed by the embodiments of the present invention
Flow diagram, in method shown in Fig. 1 the following steps are included:
S1: the inceptive filtering of feature point extraction and image pair;
Wherein, step S1 specifically includes following operation: extracting characteristic point to each image, then uses the side of image retrieval
Method finds the neighbour of each image, obtains the image pair of scene overlapping.
Wherein, step S1 specifically includes following sub-step:
(S1.1) Scale invariant features transform (Scale-invariant feature transform, SIFT) feature is used
The characteristic point of extraction algorithm extraction each image;
(1.2) words tree (vocabulary tree) is obtained with SIFT feature training, searches for obtain by words tree
K neighbour of each image obtains the images match pair of scene overlapping.
S2: Feature Points Matching and related data prepare;
Wherein, step S2 specifically includes following operation: for each image pair, carrying out characteristic point to the image of image pair
Matching carries out geometry verification by the basis matrix between image pair image to reject error hiding, obtains meeting basis matrix
Feature Points Matching number, as in points, and calculate optimal homography matrix according to matched characteristic point and obtain homograph rate,
Then variation of the characteristic point of statistical match on direction and scale obtains the variation histogram of the variation histogram and scale in direction
Figure.
Wherein, by the interior points of the available each image pair of step S2, homograph rate, direction variation histogram
With the variation histogram of scale.
Wherein, step S2 specifically includes following sub-step:
(S2.1) for each image pair, the characteristic point between image pair image is carried out using Feature Points Matching algorithm
Matching;
Wherein it is possible to using quick hash Feature Points Matching algorithm to the characteristic point progress between image pair image
Match.
(S2.2) geometry verification is carried out by the basis matrix F between estimation image pair image, rejects error hiding, obtains
It is the interior m that counts to the Feature Points Matching number for meeting F;
Wherein it is possible to combine at 8 points using consistent (RANdom SAmple Consensus, the RANSAC) algorithm of random sampling
Method estimates the basis matrix F between image pair image.
(S2.3) foundation meets the matching characteristic point of F to calculate optimal homography matrix H and obtain corresponding homograph rate
H, wherein h is used to measure the baseline width between image pair image;
Wherein it is possible to calculate optimal homography matrix H by RANSAC strategy
(S2.4) side for asking difference operation to obtain characteristic point is carried out on the direction o of characteristic point to the matching characteristic point for meeting F
Changes delta o on o carries out the changes delta s on the scale s that division arithmetic obtains characteristic point, then on the scale s of characteristic point
Discretization obtains the histogram of Δ o and the histogram of Δ s.
Wherein it is possible to the histogram of Δ o and the histogram of Δ s be obtained in the following manner, for each image pair, to full
The matching characteristic point of sufficient F carries out asking on o (direction for representing characteristic point) difference operation and enterprising in s (scale for representing characteristic point)
Row division arithmetic obtains corresponding changes delta o, Δ s in two dimensions.Δ s is normalized to [0,4], Δ o normalizes to [- 2
π, 2 π], it is then divided into n bucket respectively.For each matching characteristic point, the changes delta o in two dimensions, Δ s are calculated, so
As soon as amplitude adds 1 inside nearest bucket afterwards, counting all matched characteristic points can be obtained about two kinds of changes delta o,
The histogram of Δ s.As shown in Figure 3, Figure 4, two images are the image being compared above, below the left side figure be statistics o
The Δ o histogram that difference changes;The right is the Δ s histogram for counting the ratio of s and changing.Δ o abscissa range is
[- 2 π, 2 π], the abscissa range of Δ s are [0,4].
It should be noted that how above-mentioned be only enumerated a kind of obtains the specific of the histogram of Δ o and the histogram of Δ s
Embodiment, above-mentioned numberical range should not be construed as limiting the uniqueness of the embodiment of the present invention.
Wherein, the dimensional information and directional information of characteristic point are contained in direction o and scale s:SIFT feature.
Δ o, Δ s: the variation on direction and scale.
S3: label redundant image;
Wherein, step S3 specifically includes following operation: passing through the interior points of each image pair, homograph rate, direction
Variation histogram and scale variation histogram it is triple constraint to measure the similarity between each image pair image, will be similar
Degree meets the image composition multiple images set of all image pairs of preset rules, chooses an image in each image collection
As representative image, marking other images is redundant image.
Wherein, pass through the redundant image in the available all images of step S3.
Wherein, step S3 specifically includes following sub-step, and it is superfluous to be illustrated in figure 2 a kind of label disclosed by the embodiments of the present invention
The flow diagram of remaining image includes following operation in method shown in Fig. 2:
(S3.1) for each image pair, if the interior points m of image pair is greater than threshold alpha and homograph rate h is greater than threshold
Value β thens follow the steps (3.2);
Wherein, α and β can be rule of thumb configured.
Wherein, if the interior points m of image pair is greater than threshold alpha and indicates that picture material is more similar.
The similitude of picture material: how much judge that matching characteristic point number is more, interior according to matching characteristic purpose of counting
Rong Yue is similar.
Wherein, homograph rate h, which is greater than threshold value beta, indicates that baseline is relatively narrow.
Baseline width: the homograph rate that is acquired according to homography matrix judges that homograph rate is bigger, and baseline is narrower.
(S3.2) if peak value p of the Δ s histogram of image pair near 1 is greater than the top of threshold value η and Δ o histogram
Value q is greater than threshold θ, then by the image of the image pair alternately redundancy;
Wherein, peak value p of the Δ s histogram of image pair near 1 is greater than the scale between threshold value η expression matching characteristic point
Close, the peak-peak q of Δ o histogram, which is greater than threshold θ, indicates that the rotation angle change between matching characteristic point is consistent.If the two
All meet, then it represents that the image is alternately superfluous by the two images of the image pair to for the very high image pair of similitude
Remainder.
Scale is close to indicate that the dimensional variation between matching characteristic point is little, can be judged by Δ s histogram.
Rotation angle change unanimously indicates that the rotation angle change between matching characteristic point is unanimous on the whole, can be straight by Δ o
Square figure is judged.
Wherein, η and θ can be rule of thumb configured.
As shown in figure 3, can be seen by following histogram, although Δ o histogram has high peaks in a certain position,
It is there is no obvious peak value in Δ s histogram, and there is no integrated distribution near 1.It can be apparent by original image
There is apparent different scale between image, therefore, it is determined that being non-similarity compared with hi-vision;It is tested by another set, such as Fig. 4 institute
Show.It may be seen that Δ o histogram there are high peaks near 0, and there is obvious peak value in Δ s histogram, and concentrates and divide
Cloth is near 1, therefore, it is determined that this is similitude compared with hi-vision to image.
Fig. 3, Fig. 4 indicate a kind of experimental verification to image neighbour optimization method proposed by the present invention, should not be construed as pair
Uniqueness of the invention limits.
(S3.3) all alternative redundancys are formed into independent multiple images set, for each image collection, only retained
For a most image of matching number as representative image, remaining image tagged is redundant image in the image collection.
Wherein, representative image indicates the most information comprising other images in gathering where it, it is ensured that rebuilds knot
The information that fruit is lost is minimum.
S4: further filtering optimization image pair;
Wherein, step S4 specifically includes following operation: reject include redundant image image pair, and by interior points,
Homograph rate rejects narrow baseline image pair.
Wherein, step (4) specifically includes following sub-step:
(S4.1) figure comprising redundant image is rejected if image pair includes redundant image for each image pair
It is no to then follow the steps (S4.2) as right;
(S4.2) if the interior points m of image pair is greater than threshold gamma and homograph rate h is less than threshold value δ, retain the figure
As right, the image pair is otherwise rejected;
Wherein, γ and δ can be rule of thumb configured.
(S4.3) judge whether that traversal completes all images pair and thens follow the steps (4.1) if not completing, otherwise tie
Line journey.
S5: the preservation of information after optimization.
Wherein, step S5 specifically includes following operation: save the image pair obtained after filtering and image pair image it
Between match information.
The match information between image pair and image pair image obtained after final filtration is saved in associated documents
In, these files can be used for subsequent reconstruction, and does not need to carry out geometry verification again during reconstruction and calculate single strain
It changes.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (2)
1. a kind of image neighbour's optimization method for improving extensive three-dimensional reconstruction efficiency characterized by comprising
(1) inceptive filtering of feature point extraction and image pair: characteristic point is extracted to each image, then uses the side of image retrieval
Method finds the neighbour of each image, obtains the image pair of scene overlapping;
(2) Feature Points Matching and related data prepare: for each image pair, carrying out characteristic point to the image of image pair
Match, geometry verification is carried out to reject error hiding by the basis matrix between image pair image, obtains meeting basis matrix
Feature Points Matching number, as interior points, and calculate optimal homography matrix according to matched characteristic point and obtain homograph rate, it connects
Variation of the characteristic point on direction and scale of statistical match, obtain the variation histogram of the variation histogram and scale in direction
Figure;
Step (2) specifically includes following sub-step:
(2.1) for each image pair, the characteristic point between image pair image is matched using Feature Points Matching algorithm;
(2.2) geometry verification is carried out by the basis matrix F between estimation image pair image, rejects error hiding, is expired
The Feature Points Matching number of sufficient F is the interior m that counts;
(2.3) according to meeting the matching characteristic point of F to calculate optimal homography matrix H and obtain corresponding homograph rate h,
In, h is used to measure the baseline width between image pair image, and homograph rate is bigger, and baseline is narrower;
(2.4) the matching characteristic point for meeting F is carried out on the direction o of characteristic point on the direction o for asking difference operation to obtain characteristic point
Changes delta o, the changes delta s on the scale s that division arithmetic obtains characteristic point is carried out on the scale s of characteristic point, it is then discrete
Change the histogram of the histogram and Δ s that obtain Δ o;
(3) mark redundant image: by the interior points of each image pair, homograph rate, direction variation histogram and ruler
Similarity is met preset rules to measure the similarity between each image pair image by the triple constraints of the variation histogram of degree
The image of all image pairs forms multiple images set, chooses an image in each image collection as representative image, mark
Remember that other images are redundant image;
Step (3) specifically includes following sub-step:
(3.1) for each image pair, if the interior points m of image pair is greater than threshold alpha and homograph rate h is greater than threshold value beta,
It executes step (3.2);
(3.2) if peak value p of the Δ s histogram of image pair near 1 is greater than threshold value η and the peak-peak q of Δ o histogram is big
In threshold θ, then by the image of the image pair alternately redundancy;
(3.3) all alternative redundancys are formed into independent multiple images set and the image is only retained for each image collection
For a most image of matching number as representative image, remaining image tagged is redundant image in set;
(4) further filtering optimizes image pair: rejecting the image pair comprising redundant image, and passes through interior points, homograph
Rate rejects narrow baseline image pair;
(5) preservation of information after optimizing: the matching letter between the image pair and image pair image obtained after filtering is saved
Breath.
2. the image neighbour's optimization method according to claim 1 for improving extensive three-dimensional reconstruction efficiency, which is characterized in that
Step (4) specifically includes following sub-step:
(4.1) image pair comprising redundant image is rejected if image pair includes redundant image for each image pair,
It is no to then follow the steps (4.2);
(4.2) if the interior points m of image pair is greater than threshold gamma and homograph rate h is less than threshold value δ, retain the image pair,
Otherwise the image pair is rejected;
(4.3) judge whether that traversal completes all images pair and thens follow the steps (4.1) if not completing, otherwise terminate process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611124130.1A CN106815824B (en) | 2016-12-08 | 2016-12-08 | A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611124130.1A CN106815824B (en) | 2016-12-08 | 2016-12-08 | A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106815824A CN106815824A (en) | 2017-06-09 |
CN106815824B true CN106815824B (en) | 2019-07-05 |
Family
ID=59107113
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611124130.1A Active CN106815824B (en) | 2016-12-08 | 2016-12-08 | A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106815824B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
SG11202001487XA (en) * | 2017-09-29 | 2020-03-30 | Nec Solution Innovators Ltd | Image processing apparatus, image processing method, and computer-readable recording medium |
CN108154531B (en) * | 2018-01-03 | 2021-10-08 | 深圳北航新兴产业技术研究院 | Method and device for calculating area of body surface damage region |
US10346998B1 (en) * | 2019-02-25 | 2019-07-09 | Nurulize, Inc. | Method of merging point clouds that identifies and retains preferred points |
CN110176060B (en) * | 2019-04-28 | 2020-09-18 | 华中科技大学 | Dense three-dimensional reconstruction method and system based on multi-scale geometric consistency guidance |
CN111402412B (en) * | 2020-04-16 | 2023-06-09 | Oppo广东移动通信有限公司 | Data acquisition method and device, equipment and storage medium |
CN112729254A (en) * | 2020-12-23 | 2021-04-30 | 西南交通大学 | Aerial triangulation calculation method suitable for multi-gradient aerial belt oblique photogrammetry |
CN112822478B (en) * | 2020-12-31 | 2022-10-18 | 杭州电子科技大学 | High-quality photo sequence acquisition method for three-dimensional reconstruction |
CN113936085B (en) * | 2021-12-17 | 2022-05-20 | 荣耀终端有限公司 | Three-dimensional reconstruction method and device |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101650178A (en) * | 2009-09-09 | 2010-02-17 | 中国人民解放军国防科学技术大学 | Method for image matching guided by control feature point and optimal partial homography in three-dimensional reconstruction of sequence images |
CN101877140A (en) * | 2009-12-18 | 2010-11-03 | 北京邮电大学 | Panorama-based panoramic virtual tour method |
CN101908231A (en) * | 2010-07-27 | 2010-12-08 | 清华大学 | Reconstruction method and system for processing three-dimensional point cloud containing main plane scene |
CN103617616A (en) * | 2013-12-03 | 2014-03-05 | 康江科技(北京)有限责任公司 | Affine invariant image matching method |
CN104392453A (en) * | 2014-12-01 | 2015-03-04 | 南京邮电大学 | Method of matching and optimizing ransac features based on polar-line insertion image |
-
2016
- 2016-12-08 CN CN201611124130.1A patent/CN106815824B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101650178A (en) * | 2009-09-09 | 2010-02-17 | 中国人民解放军国防科学技术大学 | Method for image matching guided by control feature point and optimal partial homography in three-dimensional reconstruction of sequence images |
CN101877140A (en) * | 2009-12-18 | 2010-11-03 | 北京邮电大学 | Panorama-based panoramic virtual tour method |
CN101908231A (en) * | 2010-07-27 | 2010-12-08 | 清华大学 | Reconstruction method and system for processing three-dimensional point cloud containing main plane scene |
CN103617616A (en) * | 2013-12-03 | 2014-03-05 | 康江科技(北京)有限责任公司 | Affine invariant image matching method |
CN104392453A (en) * | 2014-12-01 | 2015-03-04 | 南京邮电大学 | Method of matching and optimizing ransac features based on polar-line insertion image |
Non-Patent Citations (5)
Title |
---|
Point Sets Matching by Feature-Aware Mixture Point Matching Algorithm;Kun Sun 等;《EMMCVPR 2015》;20151231;392-405 |
Progressive match expansion via coherent subspace constraint;Kun Sun 等;《Information Sciences》;20160706;848-861 |
基于匹配点相似度引导采样的图像多平面检测;曹芳 等;《计算机应用研究》;20150331;第32卷(第3期);929-933 |
基于图像的大场景三维重建并行算法研究;蒋昭炎;《中国优秀硕士学位论文全文数据库 信息科技辑(月刊)》;20160315;第2016年卷(第3期);I138-7011 |
基于特征点精确配准的图像拼接技术的研究;仲明;《中国优秀硕士学位论文全文数据库 信息科技辑(月刊)》;20151015;第2015年卷(第10期);I138-319 |
Also Published As
Publication number | Publication date |
---|---|
CN106815824A (en) | 2017-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106815824B (en) | A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency | |
Ding et al. | Multi-scale fully convolutional network for gland segmentation using three-class classification | |
Liu et al. | Deep-seated features histogram: a novel image retrieval method | |
Li et al. | A weighted sparse coding framework for saliency detection | |
Xie et al. | Beyond classification: structured regression for robust cell detection using convolutional neural network | |
Reddy Mopuri et al. | Object level deep feature pooling for compact image representation | |
Xie et al. | Discriminative object tracking via sparse representation and online dictionary learning | |
CN108334848A (en) | A kind of small face identification method based on generation confrontation network | |
Benson et al. | Deep hourglass for brain tumor segmentation | |
WO2016066042A1 (en) | Segmentation method for commodity picture and device thereof | |
CN107341812A (en) | A kind of sequence Lung neoplasm image partition method based on super-pixel and Density Clustering | |
WO2021082168A1 (en) | Method for matching specific target object in scene image | |
Dey et al. | Script independent approach for multi-oriented text detection in scene image | |
CN104361096A (en) | Image retrieval method based on characteristic enrichment area set | |
Liu et al. | Scene text detection with fully convolutional neural networks | |
Szeliski et al. | Feature detection and matching | |
Ahmad et al. | Describing colors, textures and shapes for content based image retrieval-a survey | |
Tan et al. | Saliency detection by deep network with boundary refinement and global context | |
Xu et al. | A robust hierarchical detection method for scene text based on convolutional neural networks | |
Shanmugavadivu et al. | FOSIR: fuzzy-object-shape for image retrieval applications | |
CN107357834A (en) | Image retrieval method based on visual saliency fusion | |
Wang et al. | Semantic annotation for complex video street views based on 2D–3D multi-feature fusion and aggregated boosting decision forests | |
Nguyen | TableSegNet: a fully convolutional network for table detection and segmentation in document images | |
Bougourzi et al. | Emb-trattunet: a novel edge loss function and transformer-CNN architecture for multi-classes pneumonia infection segmentation in low annotation regimes | |
Xu et al. | Robust seed localization and growing with deep convolutional features for scene text detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |