CN102567964A - Filtering method for stereoscopic vision parallax image - Google Patents
Filtering method for stereoscopic vision parallax image Download PDFInfo
- Publication number
- CN102567964A CN102567964A CN2011104123844A CN201110412384A CN102567964A CN 102567964 A CN102567964 A CN 102567964A CN 2011104123844 A CN2011104123844 A CN 2011104123844A CN 201110412384 A CN201110412384 A CN 201110412384A CN 102567964 A CN102567964 A CN 102567964A
- Authority
- CN
- China
- Prior art keywords
- point
- noise
- filtering
- image
- parallax
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention relates to a filtering method for a stereoscopic vision parallax image. The filtering method is applied to the filtering treatment for parallax data restored from an image and is used for removing noise points, thereby being beneficial to the subsequent application, such as three-dimensional reconstruction, scene analysis and path planning. The filtering method comprises the following steps: treating the parallax data and converting the parallax data into a continuous and integer image data; designing a treating method and solving a gradient of the parallax image according to the distribution feature of the parallax image; automatically cutting a gradient image and identifying a noise seed point therein; taking the noise seed point as a starting point, and filtering a noise area communicated with the gradient image; continuously filtering the parallax image after being subjected to noise filtering, and perfecting a parallax image effect; and lastly, restoring the parallax data according to the parallax image after being subjected to noise filtering. According to the filtering method provided by the invention, the noise points in the parallax data are reduced and the usability of the parallax data is increased.
Description
Technical field
The invention belongs to field of machine vision, particularly a kind of filtering method that is used for the stereoscopic vision disparity map.
Background technology
In existing theories of vision and the vision technique, have a variety ofly to the method for raising stereoscopic vision matching effect, the filtering method of disparity map is also had a lot of researchs.But most methods still on general aspect, are not directed against the method for parallax characteristic and noise special consideration thereof.
In the document research of matched filtering method " in the stereoscopic vision mistake ", two kinds of filtering methods have been proposed: based on the filter method of parallax average with based on the filter method at true reference mark to the disparity map of dense matching.Wherein first method is that the point that surpasses the parallax average in the wicket is filtered, second method be filter through the sparse match point after relaxative iteration and the minimum intermediate value quadratic method right.In the document " Study on pretreatment that the binocular solid matching image is right ", adopted general Gauss's template, smooth template and intermediate value template that disparity map has been carried out filtering.Introduce a kind of reliable polygon intersection filtering method in the document " Reliability-aware Cross Multilateral Filtering for Robust DisparityMap Refinement ", utilized the function of left and right sides image forward and reverse coupling parallax value to carry out the filtering of parallax.
Summary of the invention
The technical matters that the present invention will solve is: the deficiency that overcomes prior art; A kind of filtering method of stereoscopic vision output parallax data is provided; Effectively reduce the noise in the parallax data, improved the availability of parallax data, satisfied the demand of 3-d recovery and planning.
Technical solution of the present invention may further comprise the steps: a kind of filtering method that is used for the stereoscopic vision disparity map, and performing step is following:
The first step; Convert the parallax data of stereoscopic vision output into the integer view data; The parallax data of said stereoscopic vision output comprises Null Spot and available point, and the Null Spot gray scale is 0, and available point carry out shapingization according to maximum disparity and minimum parallax; Disparity point gray scale maximum after accomplishing is 255, and minimum disparity point gray scale is 1;
Second step, to the view data that the first step obtains, ask the gray scale difference value of neighborhood in its pixel, and reject the influence of Null Spot, calculate the gradient map of disparity map;
The 3rd step, adopt adaptive dividing method to cut apart gradient map, the fetched that graded is violent is the noise seed points;
The 4th step; With said noise seed points is starting point, carries out traversal search up and down aspect four in the noise seed points, and consecutive point that find and noise seed points gray scale difference value are thought the noise connected region less than the zone of setting threshold; Reject, obtain the image behind the noise filtering;
The 5th step, the image behind the noise filtering is carried out connected domain filtering, noise is rejected the area of absence that forms fill up, obtain new more level and smooth continuous disparity map;
In the 6th step, recover parallax data according to the disparity map after filling up.
The computation process of the said second step gradient map is:
(21) with the current point A in the view data
0Be centre coordinate, get its 8 consecutive point A on every side
1A
8, wherein setting intermediate value m is 1, then A
1=I
D(i-m, j-m), A
2=I
D(i-m, j), A
3=I
D(i-m, j+m), A
4=I
D(i, j-m), A
5=I
D(i, j+m), A
6=I
D(i+m, j-m), A
7=I
D(i+m, j), A
8=I
D(i+m, j+m), I
D(i, (i, j) the parallax gray-scale value of this point add ups gray scale in 8 consecutive point and are the total num of zero point in j) expression; I representes the pixel horizontal ordinate of this point, and j representes the pixel ordinate of this point;
(22) if total num greater than the threshold value V of the number of setting
ZNThen forward (23) to, otherwise calculate the gradient T of current point
D=| A
0* (8-num)-A
1-A
2-A
3-A
4-A
5-A
6-A
7-A
8|;
(23) with A
0The current consecutive point of point are outwards expanded a circle, and m increases by 1, in order to reject the influence of Null Spot, upgrades new A
1A
8With total num, if the number of turns of expansion is less than number of turns preset threshold V
QN, then forward (22) to, otherwise forward (24) to;
(24) with the Grad T of current point
DValue is made as 255;
(25) image is traveled through, calculate the gradient map of disparity map.
Said the 3rd step specifically is embodied as:
(31) gradient map with disparity map is divided into impartial four zones up and down;
(32) in each zone, adding up gray scale respectively is 1～255 number N (i); I=1～255; The calculating pixel number accounts for ratio
i=1～255 of the total pixel of entire image; Width is the pixel wide of image, and Height is the pixels tall of image;
(33) setting the segmentation threshold of cutting apart target prospect and background is t, calculates intermediate variable
With
Choose the criterion of cutting apart of Fisher,
when J (i) is maximum corresponding t value be optimal segmenting threshold;
(34) according to segmentation threshold gradient image is cut apart, thought the point that graded is violent, be made as 255 greater than the point of segmentation threshold, zero less than being made as of segmentation threshold.Gray scale is that 255 point is the noise seed points.
Described concrete realization of the 4th step:
(41) the noise seed points is labeled as finds a little, all the other points are labeled as and do not find a little;
(42) begin to search for from the noise seed points, if the grey value difference of more following gray-scale value on this direction and noise seed points is less than the noise threshold V that sets to its left, right-hand, above and below four direction
TNAnd be labeled as and do not find a little, think that then this point is the noise connected region, this point is labeled as finds a little, and find point coordinate to upgrade the seed points position with this; The noise seed points finds the gray scale on the disparity map of a correspondence to be made as 0 with this, from disparity map, rejects;
(43) repeating step (42) is traveled through until entire image, obtains the image behind the noise filtering.
In described the 5th step; Disparity map to filtering noise adopts medium filtering to carry out smoothly; Medium filtering is the data area of in image, extracting the filter window size, and the intensity profile in the zone is arranged, and the intermediate value of getting arrangement replaces the gray-scale value of this window center.
Advantage of the present invention is:
(1) the present invention converts the parallax value that stereoscopic vision calculates into anaglyph; And consider the different qualities of disparity map intensity profile; The form of cutting apart with gradient is extracted regional noise, and parallax data has been carried out effective filtering, has reduced the noise in the parallax data; Improved the availability of parallax data, solved to a certain extent and utilized the stereoscopic vision data to carry out the difficult problem of environment sensing.
(2) the present invention has taken into full account the discrete block distortion that occurs in the disparity map, gets rid of Null Spot and participates in calculating, and enlarges the searching scope, to guarantee the accuracy of gradient calculation, also can effectively remove the independent noise piece in the inactive area.
Description of drawings
Fig. 1 is realization flow figure of the present invention;
The original image that Fig. 2 takes for the camera among the present invention;
Fig. 3 is the disparity map that is formed by original image coupling parallax data among the present invention;
Fig. 4 is a parallax design sketch behind the noise filtering among the present invention;
Fig. 5 is a parallax design sketch after the continuity filtering among the present invention.
Embodiment
As shown in Figure 1, the present invention is concrete to be realized as follows:
The first step, the stereo-picture that left and right sides camera is taken carries out intrinsic parameter and polar curve correction to (left camera image is as shown in Figure 2, is 8 gray level images); Image to after proofreading and correct matees and interpolation; (i, j), maximum parallax is D to obtain each parallax data Df that puts floating type
Max, the minimum parallax except that zero is D
MinSetting invalid disparity point gray scale is zero; All the other points carry out integerizations; Formula is following:
i=0～Width-1 j=0～Height-1; Width is the pixel wide of image, and this figure is 256.Height is the pixels tall of image, and this figure is 256.Obtain the integer parallax Di (i of every bit; J); The formation disparity map is as shown in Figure 3; The location of pixels of the corresponding original image of each pixel among the figure, that gray-scale value is represented is the result that this impact point is imaged on the left and right sides magazine lateral coordinates difference integerization, rejects the part of not mating in the corresponding original image in the edge of black.
Second step is with the current point A in the disparity map
0Be centre coordinate, get its 8 consecutive point A on every side
1A
8, setting intermediate quantity m is 1, makes A
1=I
D(i-m, j-m), A
2=I
D(i-m, j), A
3=I
D(i-m, j+m), A
4=I
D(i, j-m), A
5=I
D(i, j+m), A
6=I
D(i+m, j-m), A
7=I
D(i+m, j), A
8=I
D(i+m j+m), adds up gray scale in 8 consecutive point and is the total num of zero point; If num less than setting value 4, then calculates T
D=| A
0* (8-num)-A
1-A
2-A
3-A
4-A
5-A
6-A
7-A
8|, otherwise with A
0The current consecutive point of point are outwards expanded a circle, make m=m+1, upgrade new A
1A
8And num, if the expansion the number of turns less than setting value 10, num also satisfies condition, and then calculates T
D=| A
0* (8-num)-A
1-A
2-A
3-A
4-A
5-A
6-A
7-A
8| (A
1A
8Be the updating value on the current circle), otherwise with the T of current point
DValue is made as 255.
The 3rd step was divided into impartial four zones up and down with gradient map; Gradient map according to 256 * 256 is divided into upper left corner i=0～127 j=0～127; Upper right corner i=128～255 j=0～127, lower left corner i=0～127 j=128～255, lower right corner i=128～255 j=128～255; The statistics gray scale is 1～255 number N (k) in each zone; K=1～255, calculating pixel number account for ratio
k=1～255 of the total pixel of entire image; Calculate intermediate quantity
According to formula
Calculate maximal value, corresponding t value is designated as D during with J (t) maximum
ThAccording to D
ThGradient image is cut apart, greater than D
ThBe made as 255, be made as zero less than segmentation threshold.
The 4th step, get the point that gray scale equals 255 on the image after cutting apart, correspond to that the point of this position is made as seed points in the disparity map; Seed points is labeled as finds a little, all the other points are labeled as and do not find a little; Begin to search for from seed points to its left, right-hand, above and below four direction; If the grey value difference of more following gray-scale value on this direction and seed points is less than setting threshold 8 and be labeled as and do not find a little; Then this point is labeled as and finds a little, and upgrade the seed points position with its coordinate; Repeat this process, traveled through until entire image.Obtain parallax design sketch such as Fig. 4 behind the noise filtering; The location of pixels of the corresponding original image of each pixel among the figure; That gray-scale value is represented is the result that this impact point is imaged on the left and right sides magazine lateral coordinates difference integerization, rejects the part of not mating in the corresponding original image in the edge of black.Compare Fig. 3, the point that graded is bigger, the intersection part that comprises white noise piece, the high lower of topographic relief is by filtering.
The 5th step, adopt medium filtering to carry out smoothly to the disparity map of filtering noise, medium filtering is in image, to extract the big or small data area of filter window, and the intensity profile in the zone is arranged, the intermediate value of getting arrangement replaces the gray-scale value of this window center.The parallax effect that obtains medium filtering is as shown in Figure 5; The location of pixels of the corresponding original image of each pixel among the figure; That gray-scale value is represented is the result that this impact point is imaged on the left and right sides magazine lateral coordinates difference integerization, rejects the part of not mating in the corresponding original image in the edge of black.Compare Fig. 4, the area of absence that the noise filtering forms obtains filling, and new disparity map is more level and smooth, and is more continuous.
The 6th step; Parallax to each picture point carries out the conversion of integer to floating-point; Di (i; J) be the integer parallax value of current point, then its floating-point parallax value is
i=0～Width-1 j=0～Height-1.Utilize Df (i; J) and the right parameter of stereoscopic camera carry out the recovery of dimensional topography value; Fundamental formular is
i=0～Width-1 j=0～Height-1;
i=0～Width-1 j=0～Height-1;
i=0～Width-1 j=0～Height-1; Wherein Baseline be stereoscopic camera between baseline; F is a camera focus; Dx is the pixel lateral dimension; Dy is the pixel longitudinal size, and Width and Height are respectively the pixel wide and the height of camera image, get Width=256 and Height=256 here.
Through the present invention, removed large-area noise piece by the dimensional topography that disparity map recovers, more identical with real terrain, data can directly be used for path planning, have solved that there is the noise piece in dimensional topography and the problem that causes planning.
The present invention does not set forth the known technology that part belongs to those skilled in the art in detail.
Claims (5)
1. filtering method that is used for the stereoscopic vision disparity map is characterized in that performing step is following:
The first step; Convert the parallax data of stereoscopic vision output into the integer view data; The parallax data of said stereoscopic vision output comprises Null Spot and available point, and the Null Spot gray scale is 0, and available point carry out shapingization according to maximum disparity and minimum parallax; Disparity point gray scale maximum after accomplishing is 255, and minimum disparity point gray scale is 1;
Second step, to the view data that the first step obtains, ask the gray scale difference value of neighborhood in its pixel, and reject the influence of Null Spot, calculate the gradient map of disparity map;
The 3rd step, adopt adaptive dividing method to cut apart said gradient map, the fetched that graded is violent in the gradient map is the noise seed points;
The 4th step; With said noise seed points is starting point, carries out traversal search up and down aspect four in the noise seed points, and consecutive point that find and noise seed points gray scale difference value are thought the noise connected region less than the zone of setting threshold; Reject, obtain the image behind the noise filtering;
The 5th step, the image behind the noise filtering is carried out connected domain filtering, noise is rejected the area of absence that forms fill up, obtain new more level and smooth continuous disparity map;
In the 6th step, recover parallax data according to the disparity map after filling up.
2. according to the filtering method that is used for the stereoscopic vision disparity map of claim 1, it is characterized in that: the gradient map process that calculates disparity map in said second step is:
(21) with the current point A in the view data
0Be centre coordinate, get A
08 consecutive point A on every side
1A
8, wherein setting intermediate value m is 1, then A
1=I
D(i-m, j-m), A
2=I
D(i-m, j), A
3=I
D(i-m, j+m), A
4=I
D(i, j-m), A
5=I
D(i, j+m), A
6=I
D(i+m, j-m), A
7=I
D(i+m, j), A
8=I
D(i+m, j+m), I
D(i, (i, j) the parallax gray-scale value of this point add ups gray scale in 8 consecutive point and are the total num of zero point in j) expression; I representes the pixel horizontal ordinate of this point, and j representes the pixel ordinate of this point;
(22) if total num greater than the threshold value V of the number of setting
ZNThen forward (23) to, otherwise calculate the gradient T of current point
D=| A
0* (8-num)-A
1-A
2-A
3-A
4-A
5-A
6-A
7-A
8|;
(23) with A
0The current consecutive point of point are outwards expanded a circle, and m increases by 1, in order to reject the influence of Null Spot, upgrades new A
1A
8With total num, if the number of turns of expansion is less than number of turns preset threshold V
QN, then forward (22) to, otherwise forward (24) to;
(24) with the Grad T of current point
DValue is made as 255;
(25) image is traveled through, calculate the gradient map of disparity map.
3. according to the filtering method that is used for the stereoscopic vision disparity map of claim 1, it is characterized in that: said the 3rd step specifically is embodied as:
(31) gradient map with disparity map is divided into impartial four zones up and down;
(32) in each zone, adding up gray scale respectively is 1～255 number N (i); I=1～255; The calculating pixel number accounts for ratio
i=1～255 of the total pixel of entire image; Width is the pixel wide of image, and Height is the pixels tall of image;
(33) setting the segmentation threshold of cutting apart target prospect and background is t; Calculate intermediate variable
and
and choose the criterion of cutting apart of Fisher,
when J (t) is maximum the t value of correspondence be optimal segmenting threshold;
(34) according to segmentation threshold gradient image is cut apart, is thought the point that graded is violent, be made as gray scale 255 greater than the point of segmentation threshold, less than segmentation threshold be made as zero, gray scale is that 255 point is the noise seed points.
4. according to the filtering method that is used for the stereoscopic vision disparity map of claim 1, it is characterized in that: described concrete realization of the 4th step:
(41) the noise seed points is labeled as finds a little, all the other points are labeled as and do not find a little;
(42) begin to search for from the noise seed points, if the grey value difference of more following gray-scale value on this direction and noise seed points is less than the noise threshold V that sets to its left, right-hand, above and below four direction
TNAnd be labeled as and do not find a little, think that then this point is the noise connected region, this point is labeled as finds a little, and find point coordinate to upgrade the seed points position with this; The noise seed points finds the gray scale on the disparity map of a correspondence to be made as 0 with this, from disparity map, rejects;
(43) repeating step (42) is traveled through until entire image, obtains the image behind the noise filtering.
5. according to the filtering method that is used for the stereoscopic vision disparity map of claim 1; It is characterized in that: in described the 5th step; Disparity map to filtering noise adopts medium filtering to carry out smoothly; Medium filtering is the data area of in image, extracting the filter window size, and the intensity profile in the zone is arranged, and the intermediate value of getting arrangement replaces the gray-scale value of this window center.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110412384.4A CN102567964B (en) | 2011-12-08 | 2011-12-08 | Filtering method for stereoscopic vision parallax image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110412384.4A CN102567964B (en) | 2011-12-08 | 2011-12-08 | Filtering method for stereoscopic vision parallax image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102567964A true CN102567964A (en) | 2012-07-11 |
| CN102567964B CN102567964B (en) | 2014-08-27 |
Family
ID=46413316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110412384.4A Active CN102567964B (en) | 2011-12-08 | 2011-12-08 | Filtering method for stereoscopic vision parallax image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102567964B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123715A (en) * | 2013-04-27 | 2014-10-29 | 株式会社理光 | Method and system for configuring parallax value |
| CN104915943A (en) * | 2014-03-12 | 2015-09-16 | 株式会社理光 | Method and apparatus for determining main disparity value in disparity map |
| CN104574342B (en) * | 2013-10-14 | 2017-06-23 | 株式会社理光 | The noise recognizing method and Noise Identification device of parallax depth image |
| CN108182666A (en) * | 2017-12-27 | 2018-06-19 | 海信集团有限公司 | A kind of parallax correction method, apparatus and terminal |
| CN109427043A (en) * | 2017-08-25 | 2019-03-05 | 国家测绘地理信息局卫星测绘应用中心 | A kind of matched smooth item calculation method of parameters of stereopsis global optimization and equipment |
| CN110110645A (en) * | 2019-04-30 | 2019-08-09 | 北京控制工程研究所 | A kind of obstacle method for quickly identifying and system suitable for low signal-to-noise ratio (SNR) images |
| CN110455502A (en) * | 2019-08-15 | 2019-11-15 | 广东海洋大学 | The method of lens and lens group focus and image point position is judged based on image parallax |
-
2011
- 2011-12-08 CN CN201110412384.4A patent/CN102567964B/en active Active
Non-Patent Citations (6)
| Title |
|---|
| FRANCISCO ROVIRA-MÁS ET AL: "Noise Reduction in Stereo Disparity Images based on Spectral Analysis", 《2009 ASABE ANNUAL INTERNATIONAL MEETING》 * |
| JORN JACHALSKY ET AL: "Reliability-aware cross multilateral filtering for robust disparity map refinement", 《3DTV-CONFERENCE: THE TRUE VISION - CAPTURE, TRANSMISSION AND DISPLAY OF 3D VIDEO》 * |
| LI LI ET AL: "Stereo Matching Algorithm Based on a Generalized Bilateral Filter Model", 《JOURNAL OF SOFTWARE》 * |
| 刘庆华 等: "双目立体匹配图像对的预处理研究", 《计算机工程与设计》 * |
| 苏永芝: "立体视觉中大视差图像误匹配滤波研究", 《物联网》 * |
| 高宏伟 等: "立体视觉中误匹配滤波方法的研究", 《计算机工程》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123715A (en) * | 2013-04-27 | 2014-10-29 | 株式会社理光 | Method and system for configuring parallax value |
| CN104123715B (en) * | 2013-04-27 | 2017-12-05 | 株式会社理光 | Configure the method and system of parallax value |
| CN104574342B (en) * | 2013-10-14 | 2017-06-23 | 株式会社理光 | The noise recognizing method and Noise Identification device of parallax depth image |
| CN104915943A (en) * | 2014-03-12 | 2015-09-16 | 株式会社理光 | Method and apparatus for determining main disparity value in disparity map |
| CN104915943B (en) * | 2014-03-12 | 2018-03-06 | 株式会社理光 | Method and apparatus for determining main parallax value in disparity map |
| CN109427043A (en) * | 2017-08-25 | 2019-03-05 | 国家测绘地理信息局卫星测绘应用中心 | A kind of matched smooth item calculation method of parameters of stereopsis global optimization and equipment |
| CN108182666A (en) * | 2017-12-27 | 2018-06-19 | 海信集团有限公司 | A kind of parallax correction method, apparatus and terminal |
| CN108182666B (en) * | 2017-12-27 | 2021-11-30 | 海信集团有限公司 | Parallax correction method, device and terminal |
| CN110110645A (en) * | 2019-04-30 | 2019-08-09 | 北京控制工程研究所 | A kind of obstacle method for quickly identifying and system suitable for low signal-to-noise ratio (SNR) images |
| CN110455502A (en) * | 2019-08-15 | 2019-11-15 | 广东海洋大学 | The method of lens and lens group focus and image point position is judged based on image parallax |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102567964B (en) | 2014-08-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102567964B (en) | Filtering method for stereoscopic vision parallax image | |
| CN101901343B (en) | Remote sensing image road extracting method based on stereo constraint | |
| CN103985108B (en) | Method for multi-focus image fusion through boundary detection and multi-scale morphology definition measurement | |
| CN111832655B (en) | Multi-scale three-dimensional target detection method based on characteristic pyramid network | |
| CN102074014B (en) | Stereo matching method by utilizing graph theory-based image segmentation algorithm | |
| CN103455991B (en) | A kind of multi-focus image fusing method | |
| CN102930530B (en) | Stereo matching method of double-viewpoint image | |
| CN104036479B (en) | Multi-focus image fusion method based on non-negative matrix factorization | |
| CN101556693B (en) | Division method for extracted watershed SAR image with threshold method and marking | |
| CN102609917B (en) | Image edge fitting B spline generating method based on clustering algorithm | |
| CN101765019B (en) | Stereo matching algorithm for motion blur and illumination change image | |
| CN112801022A (en) | Method for rapidly detecting and updating road boundary of unmanned mine card operation area | |
| CN104850850A (en) | Binocular stereoscopic vision image feature extraction method combining shape and color | |
| CN106022259A (en) | Laser-point cloud based method for extracting mountainous road by use of three-dimensional characteristic description model | |
| CN104820991A (en) | Multi-soft-constraint stereo matching method based on cost matrix | |
| CN104021368A (en) | Method and system for estimating road height shape | |
| CN103364410A (en) | Crack detection method of hydraulic concrete structure underwater surface based on template search | |
| CN107240073A (en) | A kind of 3 d video images restorative procedure merged based on gradient with clustering | |
| CN103106658A (en) | Island or reef coastline rapid obtaining method | |
| KR20130120730A (en) | Method for processing disparity space image | |
| Maltezos et al. | Automatic detection of building points from LiDAR and dense image matching point clouds | |
| CN111640128A (en) | Cell image segmentation method based on U-Net network | |
| CN105809673A (en) | SURF (Speeded-Up Robust Features) algorithm and maximal similarity region merging based video foreground segmentation method | |
| CN105023263B (en) | A kind of method of occlusion detection and parallax correction based on region growing | |
| CN103955945A (en) | Self-adaption color image segmentation method based on binocular parallax and movable outline |
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 |