CN106384363B - A kind of quick self-adapted weight solid matching method - Google Patents
A kind of quick self-adapted weight solid matching method Download PDFInfo
- Publication number
- CN106384363B CN106384363B CN201610819094.4A CN201610819094A CN106384363B CN 106384363 B CN106384363 B CN 106384363B CN 201610819094 A CN201610819094 A CN 201610819094A CN 106384363 B CN106384363 B CN 106384363B
- Authority
- CN
- China
- Prior art keywords
- point
- parallax
- pixel
- obtains
- disparity
- 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
Classifications
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20004—Adaptive image processing
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of quick self-adapted weight solid matching method, comprising: carries out polar curve correction process to left and right two images;Choosing left and right image respectively is to calculate matching cost using the combination of the luminance information of image, horizontal gradient information and vertical gradient information with reference to figure;The color weight and space length weight between pixel are calculated using cosine function, makees first polymerization using weighting aggregating algorithm and obtains disparity space;The disparity search range of each pixel is split, t sub- search ranges are obtained;In conjunction with WTA criterion and disparity space, local optimum of each pixel in the sub- search range of t parallax is calculated;The adaptive weighting between pixel is calculated, makees adaptive weighting polymerization at the local optimum parallax value selected, obtains new disparity space;Global optimum's parallax value is sought to new disparity space using WTA criterion;Obtain the more accurate disparity map of a width.The present invention can reduce calculation amount.
Description
Technical field
The invention belongs to the Stereo matching parts in computer stereo vision field, and it is vertical to be related to a kind of quick self-adapted weight
Body matching algorithm can be used for the three-dimensional reconstruction of image, provide guidance for medical image, robot navigation etc..
Background technique
Stereo matching is according to the two dimensional image of several different perspectivess in Same Scene, to obtain each object in the scene
Depth information, it is the important component and one of current research hotspot of computer vision field.Currently, three-dimensional
With being widely applied to multiple fields, such as monitor, vision guided navigation, human-computer interaction, virtual reality etc..
Stereo Matching Algorithm can be divided into local algorithm and Global Algorithm.Wherein, Global Algorithm substantially belongs to optimization and calculates
Method, it is to convert stereo matching problem to the optimization problem for finding global energy function, although can obtain relatively
Low error hiding rate, but algorithm complexity is high, is unfavorable for using in practical projects.
Local algorithm is mainly calculated using the local message around point to be matched, the information content being related to due to it compared with
Few, match time is shorter, therefore has received widespread attention.For example, some scientific research personnel propose to utilize fixed polymerizing windows knot
The method for closing adaptive weighting obtains disparity map (Adaptive support-weight approach for
correspondence search.IEEE Transactions on Pattern Analysis and Machine
Intelligence, 2006,28 (4): 650-656), although the method calculates simplicity, due in low texture and high texture area
Large scale window should be respectively adopted in domain and small size window is polymerize, and institute in this way lacks different scenes adaptive
Property.There are also scientific research personnel to propose adaptive weighting Stereo Matching Algorithm, which is applied to matching cost for bilateral filtering
Polymerization stage, although preferable matching precision can be obtained, due to needing to calculate each point to be matched at all parallax depths
Adaptive weighting, computationally intensive, long operational time is not suitable for applying in Practical Project.Currently, obtaining higher matching essence
It is able to maintain higher efficiency of algorithm while spending, is numerous scientific research personnel problem encountered in Stereo matching.In recent years, it counts
The development of calculation machine theory of stereo vision is acquisition high-precision and efficient Stereo Matching Algorithm provides theoretical foundation.
Summary of the invention
In order to solve the problems, such as that traditional adaptive weighting Stereo Matching Algorithm is computationally intensive, the present invention proposes one kind quickly certainly
Weight algorithm is adapted to, technical solution is as follows:
A kind of quick self-adapted weight Stereo Matching Algorithm, this method are improved weighing computation method and are calculated using polymerization twice
Method reduces calculation amount, including the following steps:
(1) polar curve correction process is carried out to left and right two images;
(2) choosing left and right image respectively is to utilize the luminance information of image, horizontal gradient information and vertical ladder with reference to figure
The combination for spending information calculates matching cost;
(3) horizontal line for calculating each pixel divides range, in this, as polymeric support window;Utilize cosine function meter
Calculate pixel P (x, y) and PkColor weight w between (x-k, y)c(p,pk) and space length weight ws(p,pk), wherein (x, y) and (x-k, y) respectively represents point p and pkCoordinate,
β and ζ is related coefficient, and I (x, y) represents the color of point P (x, y), and r ∈ { R, G, B } represents the three primary colors of color image, | p-pk|
Represent point p and pkBetween space length;In conjunction with the matching cost that step (2) obtains, using weighting aggregating algorithm in parallax depth
Make first polymerization in 0~D of degree and obtains disparity space;
(4) 0~D of disparity search range of each pixel is split, obtains t sub- search ranges;In conjunction with WTA standard
Then and the obtained disparity space of step (3), local optimum of each pixel in the sub- search range of t parallax is calculated;
(5) pixel p and p are calculatedkBetween adaptive weighting w (p, pk), choose the rectangular window conduct that size is N*N
Adaptive weighting polymeric support window, wherein N is the radius of rectangular window, in conjunction with the matching cost that step (2) obtains, only in step
Suddenly make adaptive weighting polymerization at the local optimum parallax value that (4) select, obtain new disparity space;
(6) global optimum's parallax value is sought using the new disparity space that WTA criterion obtains step (5), to obtain
Left and right initial parallax figure;
(7) left and right consistency desired result is carried out to left and right initial parallax figure, rejects and mismatch point, it is more accurate obtains a width
Disparity map.
The present invention is directed to the deficiency of traditional adaptive weighting Stereo Matching Algorithm, improves in original method, first
Matching cost is calculated using the combination of luminance information, horizontal gradient information and vertical gradient information, first polymerization is using rewriting
Cosine function calculates color weight and space length weight of each point relative to point to be polymerized in support window, is then weighed
Weight matching cost polymerization obtains the new matching cost of each pixel to form disparity space;To the disparity search of each pixel
Range is split to obtain disparity search subregion, in conjunction with first polymerization and obtains parallax in each subregion using WTA criterion
The optimal value in space, to obtain the local optimum parallax set of each pixel;Each pixel is only calculated when polymerizeing again
Adaptive weighting at respective optimal parallax depth obtains new disparity space to obtain original disparity map;Then pass through
Consistency detection is carried out to left and right disparity map, effective pixel point set is extracted, improves the accuracy of disparity map.The present invention with
The advantages of traditional adaptive weighting algorithm is compared be not only significantly improve the efficiency of original algorithm, but also can be improved
With precision.
In short, the present invention improves traditional adaptive weighting Stereo Matching Algorithm, a kind of quick self-adapted power is proposed
Weight Stereo Matching Algorithm, first polymerization calculate color weight and space length weight between two o'clock using cosine function, then
The local optimum parallax value for selecting point to be matched only calculates point to be matched at local optimum parallax depth when polymerizeing again
Adaptive weighting.By carrying out consistency detection to left and right disparity map, effective pixel point set is extracted, disparity map is improved
Accuracy.The present invention can not only significantly improve the efficiency of traditional adaptive weighting algorithm, and also improve matching precision,
Have many advantages, such as that accuracy rate is high, high-efficient, has a wide range of applications.
Detailed description of the invention
Fig. 1 is quick self-adapted weight Stereo Matching Algorithm flow chart of the invention.
Fig. 2 is that the first polymerization line of the present invention divides schematic diagram.
Fig. 3 is the present invention and traditional adaptive weighting Stereo Matching Algorithm to same width standard testing image matching effect pair
Than, it is (c) the final matching effect figure of traditional algorithm after parallax optimizes that (a), (b) figure, which are the test image corrected,
It (d) is the final matching effect figure of the present invention after parallax optimizes.
Specific embodiment
Specific steps and principle of the present invention are as follows:
(1) disparity estimation, the acquisition of initial parallax image
Successively choosing left and right figure is with reference to figure, with the luminance information of image, horizontal gradient information and vertical gradient information
In conjunction with calculating matching cost
C (x, y, d)=α * Cc(x,y,d)+(1-α)*(Chg(x,y,d)+Cvg(x,y,d))
Wherein, tri- channels k representative image R, G, B, ▽xIL、▽yILRespectively represent the horizontal, vertical of left figure RGB channel
Direction gradient figure, ▽xIR、▽yIRLevel, the vertical direction gradient map of right figure RGB channel are respectively represented, Tc, Thg and Tvg are to cut
Disconnected threshold value, Tc=60, Thg=15, Tvg=15.α is balance factor, value 0.1.Cc(x, y, d) represents reference image vegetarian refreshments
Brightness matching cost, Chg(x, y, d) and Cvg(x, y, d) respectively represents the horizontal and vertical gradient matching cost of reference image vegetarian refreshments,
C (x, y, d) represents the initial matching cost of reference image vegetarian refreshments.
(2) first polymerization
In order to improve matching precision, it is polymerize using certain weight matching cost method.
By taking point P (x, y) as an example, its horizontal line segmentation range is first found out, Fig. 2 is that point P (x, y) line divides schematic diagram.
Its midpoint PkWith point Pk'For the point similar most left and most right with point P (x, y).With solution point PkFor, first seek point P
(x, y) and point P1Similarity D between (x-1, y)c(P,P1):
It is wherein balance factor, value 0.8.Point P is judged with following two criterion1It is whether similar with point P:
1) point P1Whether D is metc(P,P1) < ψ, wherein ψ is threshold value, value 10;
2) point P1Whether D is mets(P,P1) < ζ, wherein Ds(P,P1)=| P-P1| represent point P1Space between point P away from
From.ζ=15 are point P1Maximum search range.
The most left P of above-mentioned condition will be met1Point is used as Pk, similarly, using the rightest point for the condition that meets as Pk'.With following public affairs
Formula calculates point PkSpace length weight w between point Ps(p,pk) and color weight wc(p,pk):
Wherein β is related coefficient, value 2.It is calculated after point P (x, y) polymerize for the first time at depth d with following formula
Matching cost C'(x, y, d):
(3) it polymerize again
After first polymerization, subregion, and parallax in each subregion are divided at equal intervals to 0~D of disparity search range
Number is 13.If parallax number is less than 13 in the last one subregion, as an independent subregion.With point P
For (x, y), the local optimum parallax collection selected is combined into D'={ d1',d2'...dt', wherein t represents subregion number,
di' (i=1~t) represent the local optimum parallax value that selects, andWherein Di(i=1~t)
Represent i-ththParallax in sub-regions.
Adaptive weighting of the point to be matched at local optimum parallax depth is only calculated, then point P (x, y) is in parallax depth d
Locate new matching cost are as follows:
Wherein N (p) represents the rectangle polymerizing windows chosen and radius represents point in window, C as 12, qaggr(p, d) generation
New matching cost of the table point p at parallax d is also known as disparity space, and w (p, q) is the weight between point p and point q, calculation formula
For
Wherein σcAnd σsFor related coefficient, value is respectively 0.4 and 9.
The calculating of parallax value is tactful using WTA (WinnerTakesAll), final parallax value d at point P (x, y)PFor
After obtaining initial parallax figure, parallax outlier is detected using left and right consistency detecting method, is unsatisfactory for above formula
Point is parallax outlier.
|Dl(x,y)-Dr(x-Dl(x, y), y) | < 2
D in formulalIt is the initial parallax figure obtained with reference to figure, D that (x, y), which is with left figure,rIt is with reference to figure that (x, y), which be with right figure,
The initial parallax figure arrived.By carrying out consistency desired result to horizontal parallax figure, the wrong disparity value in initial matching can be picked
It removes.
Detailed process of the invention is as follows:
(1) polar curve correction process is carried out to left and right two images.
(2) choosing left and right image respectively is to utilize the luminance information of image, horizontal gradient information and vertical ladder with reference to figure
The combination for spending information calculates matching cost.
(3) horizontal line for calculating each pixel divides range, in this, as polymeric support window.Utilize cosine function meter
Calculate pixel P (x, y) and PkColor weight w between (x-k, y)c(p,pk) and space length weight ws(p,pk), wherein β and ζ is related coefficient, IrRepresent the corresponding R of pixel,
G, B three primary colors.In conjunction with the matching cost that step (2) obtains, made in 0~D of parallax depth using weighting aggregating algorithm poly- for the first time
Conjunction obtains disparity space.
(4) 0~D of disparity search range of each pixel is split, obtains t sub- search ranges.In conjunction with WTA standard
Then and the obtained disparity space of step (3), local optimum of each pixel in the sub- search range of t parallax is calculated.
(5) pixel p and p are calculatedkBetween adaptive weighting w (p, pk), choose the rectangular window conduct that size is N*N
Adaptive weighting polymeric support window, in conjunction with the matching cost that step (2) obtains, the local optimum only selected in step (4)
Make adaptive weighting polymerization at parallax value, obtains new disparity space.
(6) global optimum's parallax value is sought using the disparity space that WTA criterion obtains step (5), to be controlled
Initial parallax figure.
(7) left and right consistency desired result is carried out to left and right initial parallax figure, rejects and mismatch point, it is more accurate obtains a width
Disparity map.
Claims (1)
1. a kind of quick self-adapted weight solid matching method, this method improves weighing computation method and using aggregating algorithm twice
To reduce calculation amount, including the following steps:
(1) polar curve correction process is carried out to left and right two images;
(2) choosing left and right image respectively is to be believed with reference to figure using the luminance information of image, horizontal gradient information and vertical gradient
The combination of breath calculates matching cost;
(3) horizontal line for calculating each pixel divides range, in this, as polymeric support window;By taking point P (x, y) as an example, first
Find out its horizontal line segmentation range;Set up an office PkWith point Pk' be and point that point P (x, y) is similar most left and most right, with solution point Pk
For, first ask point P (x, y) and point P1Similarity D between (x-1, y)c(P,P1):
Wherein λ is balance factor, value 0.8;Point P is judged with following two criterion1It is whether similar with point P:
1) point P1Whether D is metc(P,P1) < ψ, wherein ψ is threshold value, value 10;
2) point P1Whether D is mets(P,P1) < ζ, wherein Ds(P,P1)=| P-P1| represent point P1Space length between point P;ζ
=15 be point P1Maximum search range;
The most left P of above-mentioned condition will be met1Point is used as Pk, similarly, the most right P of above-mentioned condition will be met1Point is used as Pk';
Pixel P (x, y) and P are calculated using cosine functionkColor weight w between (x-k, y)c(p,pk) and space length power
Weight ws(p,pk), wherein
(x, y) and (x-k, y) respectively represents point p and pkSeat
Mark, β are related coefficient, and I (x, y) represents the color of point P (x, y), and r ∈ { R, G, B } represents the three primary colors of color image, | p-pk|
Represent point p and pkBetween space length;In conjunction with the matching cost that step (2) obtains, using weighting aggregating algorithm in parallax depth
The first polymerization of work obtains disparity space in 0~D of degree, calculates the matching cost C' after point P (x, y) polymerize for the first time at depth d
The formula of (x, y, d) is as follows, and wherein k is point P (x, y) similarity formula Dc(P,P1) point that finds:
(4) 0~D of disparity search range of each pixel is split, obtains t sub- search ranges;In conjunction with WTA criterion with
And the disparity space that step (3) obtains, calculate local optimum of each pixel in the sub- search range of t parallax;
(5) pixel p and p are calculatedkBetween adaptive weighting w (p, pk), choosing size is the rectangular window of N*N as adaptive
Weight polymeric support window is answered, wherein N is the radius of rectangular window, in conjunction with the matching cost that step (2) obtains, only in step
(4) make adaptive weighting polymerization at the local optimum parallax value selected, obtain new disparity space;
(6) global optimum's parallax value is sought using the new disparity space that WTA criterion obtains step (5), to be controlled
Initial parallax figure;
(7) left and right consistency desired result is carried out to left and right initial parallax figure, rejects and mismatch point, obtains the more accurate parallax of a width
Figure.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610819094.4A CN106384363B (en) | 2016-09-13 | 2016-09-13 | A kind of quick self-adapted weight solid matching method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610819094.4A CN106384363B (en) | 2016-09-13 | 2016-09-13 | A kind of quick self-adapted weight solid matching method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106384363A CN106384363A (en) | 2017-02-08 |
CN106384363B true CN106384363B (en) | 2019-09-06 |
Family
ID=57936348
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610819094.4A Active CN106384363B (en) | 2016-09-13 | 2016-09-13 | A kind of quick self-adapted weight solid matching method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106384363B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107993233B (en) * | 2016-10-26 | 2022-02-22 | 中国科学院深圳先进技术研究院 | Pit area positioning method and device |
CN107481246B (en) * | 2017-07-14 | 2019-08-16 | 天津大学 | Parallax method of refining based on matching cost update and image segmentation |
CN108876841B (en) * | 2017-07-25 | 2023-04-28 | 成都通甲优博科技有限责任公司 | Interpolation method and system in parallax refinement of parallax map |
CN107452028B (en) * | 2017-07-28 | 2020-05-26 | 浙江华睿科技有限公司 | Method and device for determining position information of target image |
CN109672876A (en) * | 2017-10-17 | 2019-04-23 | 福州瑞芯微电子股份有限公司 | Depth map processing unit and depth map processing unit |
CN108596975B (en) * | 2018-04-25 | 2022-03-29 | 华南理工大学 | Stereo matching algorithm for weak texture region |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102096919A (en) * | 2010-12-31 | 2011-06-15 | 北京航空航天大学 | Real-time three-dimensional matching method based on two-way weighted polymerization |
CN102572485A (en) * | 2012-02-02 | 2012-07-11 | 北京大学 | Self-adaptive weighted stereo matching algorithm, stereo display and collecting device and system |
CN103236057A (en) * | 2013-04-24 | 2013-08-07 | 上海大学 | Improved self-adaptive weight parallax matching method |
CN104966290A (en) * | 2015-06-12 | 2015-10-07 | 天津大学 | Self-adaptive weight three-dimensional matching method based on SIFT descriptor |
-
2016
- 2016-09-13 CN CN201610819094.4A patent/CN106384363B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102096919A (en) * | 2010-12-31 | 2011-06-15 | 北京航空航天大学 | Real-time three-dimensional matching method based on two-way weighted polymerization |
CN102572485A (en) * | 2012-02-02 | 2012-07-11 | 北京大学 | Self-adaptive weighted stereo matching algorithm, stereo display and collecting device and system |
CN103236057A (en) * | 2013-04-24 | 2013-08-07 | 上海大学 | Improved self-adaptive weight parallax matching method |
CN104966290A (en) * | 2015-06-12 | 2015-10-07 | 天津大学 | Self-adaptive weight three-dimensional matching method based on SIFT descriptor |
Non-Patent Citations (4)
Title |
---|
A Line-Based Adaptive-Weight Matching Algorithm Using Loopy Belief Propagation;Hui Li 等;《Mathematical Problems in Engineering》;20151231;第1-13页 |
Accurate stereo matching algorithm based on cost aggregation with adaptive support weight;C.-H. Lin 等;《The Imaging Science Journal》;20151231;第63卷(第8期);第423-432页 |
Real-Time Accurate Stereo Matching using Modified Two-Pass Aggregation and Winner-Take-All Guided Dynamic Programming;Xuefeng Chang 等;《2011 International Conference on 3D Imaging, Modeling, Processing, Visualization and Transmission》;20111231;第73-79页 |
基于可移动窗口和扩散距离的立体匹配算法;卞程飞 等;《***仿真学报》;20140930;第26卷(第9期);第2130-2135页 |
Also Published As
Publication number | Publication date |
---|---|
CN106384363A (en) | 2017-02-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106384363B (en) | A kind of quick self-adapted weight solid matching method | |
CN106600632B (en) | A kind of three-dimensional image matching method improving matching cost polymerization | |
CN103226821B (en) | Stereo matching method based on disparity map pixel classification correction optimization | |
CN103971366B (en) | A kind of solid matching method being polymerize based on double weights | |
CN105528785B (en) | A kind of binocular vision image solid matching method | |
CN104867135B (en) | A kind of High Precision Stereo matching process guided based on guide image | |
CN108615244B (en) | A kind of image depth estimation method and system based on CNN and depth filter | |
CN107301664A (en) | Improvement sectional perspective matching process based on similarity measure function | |
CN110148181A (en) | A kind of general binocular solid matching process | |
CN107944386B (en) | Visual scene recognition methods based on convolutional neural networks | |
CN104867133B (en) | A kind of quick substep solid matching method | |
Chen et al. | Transforming a 3-d lidar point cloud into a 2-d dense depth map through a parameter self-adaptive framework | |
CN106875443B (en) | The whole pixel search method and device of 3-dimensional digital speckle based on grayscale restraint | |
CN109255811A (en) | A kind of solid matching method based on the optimization of confidence level figure parallax | |
CN106355608B (en) | The solid matching method with S-census transformation is calculated based on Changeable weight cost | |
CN103325120A (en) | Rapid self-adaption binocular vision stereo matching method capable of supporting weight | |
CN102665086A (en) | Method for obtaining parallax by using region-based local stereo matching | |
CN104200453B (en) | Parallax image correcting method based on image segmentation and credibility | |
CN104463870A (en) | Image salient region detection method | |
CN108322724A (en) | Image solid matching method and binocular vision equipment | |
CN106952304B (en) | A kind of depth image calculation method using video sequence interframe correlation | |
CN104574366A (en) | Extraction method of visual saliency area based on monocular depth map | |
CN107316326A (en) | Applied to disparity map computational methods of the binocular stereo vision based on side and device | |
CN106815594A (en) | Solid matching method and device | |
CN107560592A (en) | A kind of precision ranging method for optronic tracker linkage target |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |