CN110443823A - A kind of floater foreground segmentation method - Google Patents
A kind of floater foreground segmentation method Download PDFInfo
- Publication number
- CN110443823A CN110443823A CN201810412292.8A CN201810412292A CN110443823A CN 110443823 A CN110443823 A CN 110443823A CN 201810412292 A CN201810412292 A CN 201810412292A CN 110443823 A CN110443823 A CN 110443823A
- Authority
- CN
- China
- Prior art keywords
- image
- segmentation
- value
- foreground
- floater
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of floater foreground segmentation methods, it is characterized in that, it mainly include image preprocessing part, bank body partitioning portion, prospect pre-segmentation part and prospect correct part, its foreground segmentation process for being mainly used for floater, unitary of illumination processing is carried out to the Surface Picture under different illumination conditions, eliminate influence of the illumination to foreground segmentation, it is responded simultaneously from water-surface areas in the different of frequency domain using bank body, bank body region is separated with water-surface areas, eliminate influence of the bank body to floating material foreground segmentation, simultaneously, based on adjacent similar priori conditions, to belong to same category of image-region distribution with identical class label, then body segmentation in opposite bank is undesirable, water surface building and isolated image region are handled, correct the segmentation result of foreground target.
Description
Technical field
The present invention relates to a kind of floater foreground segmentation methods of view-based access control model method, can be in complicated Surface Picture
It is partitioned into out the foreground area of non-waters part, belongs to field of visual inspection.
Background technique
With the development of industry and improvement of living standard, the floating on water in various regions waters various rubbish, to Natural Water
Quality guarantee shield, shipping, power generation etc. all cause different degrees of influence.Therefore, salvaging cleaning is carried out to garbage on water, society can be promoted
It can benefit and economic benefit.
Cleaning floater method be usually passive type setting block bleaching device, make its natural packing to a certain extent after
Manpower salvaging is carried out again.Although this method is simple, consumption human resources are more, operating efficiency is low, consuming time is long, even
Easily cause safety accident.
Also have simultaneously and salvaging cleaning carried out to garbage on water by the method for teleoperated vehicle, but this method for
The technical level of operator has high requirement, and operator is faced with heavy job task and complicated waters ring
Border may not be able to carry out prune job, the rubbish prune job in especially a wide range of waters well.
At abroad, the Seabin water surface floating dustbin in Europe, makes garbage on water certainly by water surface wave and tidal action
It is dynamic to enter in bucket, but it still needs manually to be timed it cleaning.Japanese " BayClean " is for clearing up sea rubbish
With recycling oil spilling, but it is main or identification of rubbish target is carried out by people, and then manpower completes prune job.The sea in South Africa
Ocean cleaning robot " WasteShark " is gone on patrol in harbour automatically, is encountered rubbish and is then cleaned, but this robot traverses
Entire operation waters, in the lesser waters of rubbish distribution density, working efficiency is extremely inefficient.
At home, the garbage on water automation cleaning ship of Ningbo Rui Chuan Environmental Protection Technology Co., Ltd is mainly used in city river
The refuse on water surface in road is salvaged after being gathered the certain rubbish in front by swing arm, still manpower is relied on to be grasped
Make.The water rubbish clean-up vessel that Luoyang City's rivers and canals maintenance center introduces realizes the cutting of rubbish, for the floating of flood season river surface
The bulky refuses such as trunk can also be salvaged easily, but its work flow is still and is realized by artificial remote operating mode.
With the development of machine vision technique, the unmanned fishing operation of garbage on water can help to solve clear under vision guide
The problem of managing floating material.Therefore, it is necessary in conjunction with the relevant technologies of machine vision and image procossing, from new angle to water surface rubbish
Rubbish carries out intelligent recognition judgement, so that Work robot is independently carried out the salvaging cleaning of garbage on water, to improving the ecological environment and mention
High social benefit, is of great significance.
Summary of the invention
To solve the above-mentioned problems, the present invention provides a kind of floater foreground segmentation method, and this method is to acquiring
To waters image handled, and the floating target that will not belong to background is split from the image of waters, in order to it
It is identified and is positioned, examined, there is preferable floater recognition effect.It mainly include image preprocessing part, bank body
Partitioning portion, prospect pre-segmentation part and prospect correct part.
Technical scheme is as follows: a kind of floater foreground segmentation method, this method main working process packet
Containing following steps:
(1) Image Acquisition and pretreatment: waters image is acquired by the camera being placed in water surface cleaning equipment, then to acquisition
Obtained image carries out unitary of illumination processing, avoids the different illumination conditions under different weather situation to floater prospect
Segmentation interferes；
(2) bank body is divided: for step (1) treated image information, using the image of water-surface areas it is a wide range of on tend to one
It causes and this mixed and disorderly feature of the image information on bank body is obtained from separating bank body region from the image of waters on frequency domain
Effective water-surface areas region in image；
(3) prospect pre-segmentation: for obtained effective water-surface areas image information, we are based on similar adjacent theory and carry out to it
Subdivision and merging treatment, are assigned to same category for similar image-region, utilize the difference of floating material and waters part on the image
The opposite sex, the floating material foreground target that will not belong to water surface background come out from background saliency；
(4) prospect is corrected: due to merging again after image subdivision, it may appear that certain isolated thins, but these parts
Being not as it is floater, but ambiguousness when due to Isolated interferers or merging, still carried out simply connected
Processing；Meanwhile we are based on a priori knowledge: what is be connected on a large scale with image boundary is background area, will be divided due to bank body
From be not thorough, heavy construction in the water surface inverted image, cross waters pontic or sluice etc. assign background again from foreground target
Label.
Detailed description of the invention
Fig. 1 is a kind of overview flow chart of floater foreground segmentation method of the present invention.
Specific embodiment
The technical scheme of the present invention will be explained in further detail with reference to the accompanying drawings and detailed description.
Shown in referring to Fig.1, a kind of floater foreground segmentation method, mainly comprising image preprocessing part, bank body point
Cut part, prospect pre-segmentation part and prospect amendment part.
(1) image preprocessing
It is extremely crucial for image quality under the different weather situation in same place due to the distribution site multiplicity of floater
Illumination factor be also not quite similar, therefore it is particularly important to the foreground segmentation of floater how to eliminate illumination, adopts in the present invention
It takes the method for unitary of illumination, concrete methods of realizing is as follows:
1.: the Surface Picture collected is transformed into CIE Lab color space from RGB color；
2.: it averages to the channel L of Surface Picture in CIE Lab space, and subtracts average value pixel-by-pixel, it is residual to obtain illumination
Value, if the illumination residual value of current point is negative, takes its absolute value；
3.: the enhancing of logarithmetics contrast is carried out to the illumination residual value of Surface Picture, i.e., illumination residual value is taken into logarithmetics, formula is, in formula,For step 2. in illumination residual value；
4.: by the original channel a, b of Surface Picture in CIE Lab space and through step 1., 2., 3. handle after obtained L port number
According to RGB color is transformed into together, the Surface Picture after unitary of illumination is obtained.
Since in CIE Lab space, the channel L is the luminance information of image, the atmospheric window of same waters image is close, but
The light characteristics of itself lead to its light tone, and reactions vary, we can not directly remove the Lighting information in the channel L, meeting in this way
Its light characteristics is lost, is unfavorable for come out different from the floating material Target Segmentation of water-surface areas light tone characteristic, to it in the present invention
It is average to carry out the overall situation, and point seeks its residual error with global mean value pixel-by-pixel, it can be by the image inconsistent with general image
Region significantly comes out, but the brightness value of some regions and global mean value are close, and it is larger by residual error to will lead to its light tone information
Area flooding, still again to illumination residual value carry out the enhancing of logarithmetics contrast, promoted background and foreground target contrast,
And inhibit unusual light tone region, it is come out conducive to by prospect floating material target from background saliency.
(2) bank body is divided
We carry out foreground segmentation to the floating material of the water surface, and bank body region inevitably comes across the water-surface areas figure of acquisition
As in, the diversity factor of this part and water-surface areas is much bigger relative to the diversity factor of floater and water-surface areas, therefore extremely
It is easy to cause foreground segmentation mistake or floater can not be split from the background of water-surface areas, therefore in the present invention, In
Before carrying out foreground segmentation, first bank body region is separated from the image of water-surface areas, since water-surface areas is in color, bright
It is close consistent on degree, form, therefore it is closer in a frequency domain, and due to being dispersed with building, trees and other forms on bank body
Different uncertain object, it is also more at random in frequency domain, still to the water-surface areas image after unitary of illumination
The wavelet transform process for carrying out frequency domain carries out multiple small echo orthogonal transformation to image, by high-frequency information region abundant from image
In separate, i.e., bank body region is separated with water-surface areas, obtains effective water-surface areas image information.
(3) prospect pre-segmentation
For our human body subjective visions, we can easily say in water-surface areas that belonging to same mesh target area is considered as very much
The same area, among these it is understood that a particularly important priori knowledge be that adjacent pixel is identical or phase in same target
Seemingly, or referred to as adjacent principle of similarity, it is just based on this priori knowledge in the present invention, to the different mesh in the image of water-surface areas
Mark is divided, and the floating material foreground target that will not belong to background is split from the image of waters, and specific implementation step is such as
Under:
1.: be by waters image cuttingFritter, the size of each image block is according to the prospect floating material mesh that need to divide
Depending on mark, in general, m, n are no more than 16；
2.: seek the central point P of each image block；It averages, obtains average about pixel column to every row pixel of image block
The column vector of value seeks intermediate value to the column vector, obtains the characteristic value R of the block of pixels in the row direction；Similarly to the every of image block
Column pixel is averaged, and the row vector about pixel column average value is obtained, and seeks intermediate value to the row vector, in a column direction should be obtained
The characteristic value C of block of pixels；
3.: judging whether two block of pixels are adjacent by the Euclidean distance between the central point of image block (has common edge and public top
Point image block be all adjacent), between adjacent image block, calculate image block characteristics value R and characteristic value C Euclid away from
From the image block by Euclidean distance less than threshold value is considered as same category；
4.: it repeats step 3., divides equally until all image blocks and be equipped with corresponding class label.
In order to better conform to the broken edge of either foreground target or background area, need to cut image block size
Get sufficiently small, form target by sufficient amount of image block, just can preferably be portrayed the shape of target in this way,
Also it is more accurate Target Segmentation to be obtained.
(4) prospect is corrected
By above-mentioned steps, we can be distributed for the different target in the image of waters with different class labels, but for
Bank body is divided not accurate enough in step (2), can make in step (3) remaining relative to bank body region, meanwhile, removed in waters
Bank is external, also deposits the regions such as the bridge, sluice, power station for crossing waters, although this part and non-water surface background parts,
But for certain intelligent cleaning equipments, cleaning work can not be undoubtedly carried out, then we also need to carry out foreground target
It corrects, in the present invention, we are based on a priori knowledge: the target area overwhelming majority being connected with image boundary is background mesh
Mark, then, we redistribute the part being connected with image boundary with background label, even if there are certain floating material prospect mesh
Mark is exactly in image boundary, but during prune job, with the variation in the image capture device visual field, can gradually be taken off
From image boundary, prospect label can be assigned at that time, and by its from background it is separated.
At the same time, for being distributed different target with different class labels based on adjacent principle of similarity in step (3),
But in assorting process, it is understood that there may be a certain image block is not assigned to the classification mark of any one target area of surrounding
Label, become one isolated unusual piece, characteristic value R, C that then we seek the image block again belong to different mesh from adjacent
Mark region image block characteristic value R, C between Euclidean distance, by this isolate block distribution with Europe between its characteristic value it is several in
Obtain the class label apart from the smallest target area.
The foregoing is only a preferred embodiment of the present invention, the application scope of application of the invention is without being limited thereto, appoint
What those familiar with the art within the technical scope of the present disclosure, the technical solution that can be become apparent to
Simple change or equivalence replacement each fall in the application scope of application of the invention.
Claims (5)
1. a kind of floater foreground segmentation method characterized by comprising S1, image preprocessing；S2, the segmentation of bank body；
S3, prospect pre-segmentation；S4, prospect amendment.
2. a kind of floater foreground segmentation method according to claim 1, which is characterized in that in the step S1,
Carry out the pretreatment of unitary of illumination, the specific steps of the preprocess method are as follows: S11, by the Surface Picture collected from RGB
Color space conversion is to CIE Lab color space；S12, it averages to the channel L of Surface Picture in CIE Lab space, and by
Pixel subtracts average value, obtains illumination residual value, if the illumination residual value of current point is negative, takes its absolute value；S13, to water surface figure
The illumination residual value of picture carries out the enhancing of logarithmetics contrast, i.e., illumination residual value is taken logarithmetics, formula are as follows:
In formula,For the illumination residual value in step S12；S14, by the original channel a, b of Surface Picture in CIE Lab space and warp
Step S11, the L channel data obtained after S12, S13 processing is transformed into RGB color together, obtains by unitary of illumination
Surface Picture later.
3. a kind of floater foreground segmentation method according to claim 1, which is characterized in that in the step S2,
Bank body region and water-surface areas are separated in the different responses of frequency domain by bank body region and water-surface areas image information, had
Body mode are as follows: the small echo orthogonal transformation that multiple frequency domain is carried out to image separates in high-frequency information region abundant from image
Come.
4. a kind of floater foreground segmentation method according to claim 1, which is characterized in that in the step S3,
Pass through adjacent similar priori knowledge and carry out foreground object segmentation, the specific implementation step of the dividing method are as follows: S41, by waters
Image cutting isFritter, for the size of each image block depending on the prospect floating material target that need to divide, m, n are equal
No more than 16；S42, the central point P for seeking each image block；It averages, is obtained about pixel to every row pixel of image block
The column vector of row average value seeks intermediate value to the column vector, obtains the characteristic value R of the block of pixels in the row direction；Similarly to image
The each column pixel of block is averaged, and the row vector about pixel column average value is obtained, and is sought intermediate value to the row vector, is obtained in column side
The characteristic value C of the upward block of pixels；S43, judged by the Euclidean distance between the central point of image block two block of pixels whether phase
Adjacent (image block of common edge and public vertex is all adjacent), between adjacent image block, calculate image block characteristics value R and
The image block that Euclidean distance is less than threshold value is considered as same category by the Euclidean distance of characteristic value C；S44, step is repeated
S43 divides equally until all image blocks and is equipped with corresponding class label.
5. a kind of floater foreground segmentation method according to claim 1, which is characterized in that in the step S4,
Whether it is connected with boundary the image-region that segmentation effect is modified, and eliminates isolated by the foreground target of pre-segmentation,
Enhance the connectivity of foreground target, specific implementation method are as follows: S51, redistribute the part being connected with image boundary with background
Label；S52, for isolated unusual piece, processing method of the present invention are as follows: seek again characteristic value R, C of the image block with it is adjacent
The image block for belonging to different target region characteristic value R, C between Euclidean distance, by this isolate block distribution with its spy
The class label of the smallest target area of Euclidean distance between value indicative.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810412292.8A CN110443823A (en) | 2018-05-03 | 2018-05-03 | A kind of floater foreground segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810412292.8A CN110443823A (en) | 2018-05-03 | 2018-05-03 | A kind of floater foreground segmentation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110443823A true CN110443823A (en) | 2019-11-12 |
Family
ID=68428052
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810412292.8A Pending CN110443823A (en) | 2018-05-03 | 2018-05-03 | A kind of floater foreground segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443823A (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101526994A (en) * | 2009-04-03 | 2009-09-09 | 山东大学 | Fingerprint image segmentation method irrelevant to collecting device |
| CN104933714A (en) * | 2015-06-12 | 2015-09-23 | 北京理工大学 | SAR remote sensing scene oil-spilling segmentation and detection method based on context relation |
| CN105139415A (en) * | 2015-09-29 | 2015-12-09 | 小米科技有限责任公司 | Foreground and background segmentation method and apparatus of image, and terminal |
| US20160092746A1 (en) * | 2014-09-25 | 2016-03-31 | Aricent Technologies Luxembourg s.a.r.l. | Intelligent background selection and image segmentation |
| CN105512666A (en) * | 2015-12-16 | 2016-04-20 | 天津天地伟业数码科技有限公司 | River garbage identification method based on videos |
| CN105913419A (en) * | 2016-04-07 | 2016-08-31 | 南京汇川图像视觉技术有限公司 | TFT-LCD mura defect detection method based on ICA learning and multichannel fusion |
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN106022288A (en) * | 2016-05-30 | 2016-10-12 | 电子科技大学 | Marine oil spill information identification and extraction method based on SAR image |
| CN106228545A (en) * | 2016-07-14 | 2016-12-14 | 西北民族大学 | A kind of automatic division method of figure of buddha class Tangka's concordance damaged area |
| CN106384344A (en) * | 2016-08-30 | 2017-02-08 | 中国科学院长春光学精密机械与物理研究所 | Sea-surface ship object detecting and extracting method of optical remote sensing image |
| CN107492076A (en) * | 2017-08-03 | 2017-12-19 | 重庆大学 | A kind of freeway tunnel scene vehicle shadow disturbance restraining method |
| CN107622231A (en) * | 2017-09-08 | 2018-01-23 | 内蒙古大学 | A kind of water day integrally intelligent floating object collection system and its collection method |
-
2018
- 2018-05-03 CN CN201810412292.8A patent/CN110443823A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101526994A (en) * | 2009-04-03 | 2009-09-09 | 山东大学 | Fingerprint image segmentation method irrelevant to collecting device |
| US20160092746A1 (en) * | 2014-09-25 | 2016-03-31 | Aricent Technologies Luxembourg s.a.r.l. | Intelligent background selection and image segmentation |
| CN104933714A (en) * | 2015-06-12 | 2015-09-23 | 北京理工大学 | SAR remote sensing scene oil-spilling segmentation and detection method based on context relation |
| CN105139415A (en) * | 2015-09-29 | 2015-12-09 | 小米科技有限责任公司 | Foreground and background segmentation method and apparatus of image, and terminal |
| CN105512666A (en) * | 2015-12-16 | 2016-04-20 | 天津天地伟业数码科技有限公司 | River garbage identification method based on videos |
| CN105913419A (en) * | 2016-04-07 | 2016-08-31 | 南京汇川图像视觉技术有限公司 | TFT-LCD mura defect detection method based on ICA learning and multichannel fusion |
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN106022288A (en) * | 2016-05-30 | 2016-10-12 | 电子科技大学 | Marine oil spill information identification and extraction method based on SAR image |
| CN106228545A (en) * | 2016-07-14 | 2016-12-14 | 西北民族大学 | A kind of automatic division method of figure of buddha class Tangka's concordance damaged area |
| CN106384344A (en) * | 2016-08-30 | 2017-02-08 | 中国科学院长春光学精密机械与物理研究所 | Sea-surface ship object detecting and extracting method of optical remote sensing image |
| CN107492076A (en) * | 2017-08-03 | 2017-12-19 | 重庆大学 | A kind of freeway tunnel scene vehicle shadow disturbance restraining method |
| CN107622231A (en) * | 2017-09-08 | 2018-01-23 | 内蒙古大学 | A kind of water day integrally intelligent floating object collection system and its collection method |
Non-Patent Citations (5)
| Title |
|---|
| ETHAN D. MONTAG等: "Psychophysical Evaluation of Gamut Mapping Techniques Using Simple Rendered Images and Artificial Gamut Boundaries", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 张燕红等: "基于小波变换和频域滤波的指纹增强方法", 《图像处理》 * |
| 肖利平等: "复杂海地背景下的舰船目标检测", 《光电工程》 * |
| 薛萍: "基于超像素特征表示的图像前景背景分割算法", 《西安科技大学学报》 * |
| 詹曙等: "边缘增强的多字典学习图像超分辨率重建算法", 《光电工程》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107622231B (en) | A kind of intelligent floating material collection system of water day one and its collection method | |
| CN109978869A (en) | A kind of sea horizon detection method and system based on gray level co-occurrence matrixes and Hough transform | |
| CN109472200B (en) | Intelligent sea surface rubbish detection method, system and storage medium | |
| CN108647648A (en) | A kind of Ship Recognition system and method under visible light conditions based on convolutional neural networks | |
| CN108229342B (en) | Automatic sea surface ship target detection method | |
| CN102663726B (en) | Method and device for material segmentation of material accumulation images | |
| CN106127728B (en) | A kind of infrared image sea horizon under sea and sky background is connected to area detecting method | |
| CN114708464B (en) | Municipal sanitation cleaning garbage truck cleaning method based on road garbage classification | |
| CN110751075A (en) | Remote sensing image culture pond detection method based on example segmentation | |
| CN1623171A (en) | Method for producing cloud free and cloud-shadow free images | |
| CN104700071B (en) | A kind of extracting method of panorama sketch road profile | |
| CN106650728B (en) | A kind of shade Binarization of License Plate Images | |
| CN114881869A (en) | Inspection video image preprocessing method | |
| CN110348442B (en) | Shipborne radar image offshore oil film identification method based on support vector machine | |
| CN103824078A (en) | Complex scene multi-license plate positioning method | |
| CN110443823A (en) | A kind of floater foreground segmentation method | |
| Gusmawati et al. | Aquaculture Pond Precise Mapping in Perancak Estuary, Bali, Indonesia | |
| Sheta et al. | Detection of oil spills in SAR images using threshold segmentation algorithms | |
| CN109145820B (en) | River channel position marking method based on video dynamic images | |
| CN117058018A (en) | Method for repairing suspended impurity vision shielding area facing underwater structure detection | |
| CN109859220B (en) | Linear image segmentation method | |
| CN108520498B (en) | Efficient crystal shadow noise removal method in crystal crystallization process monitoring | |
| CN116597164A (en) | Port ship waterline detection method based on computer vision | |
| CN110751150A (en) | FPGA-based binary neural network license plate recognition method and system | |
| CN101719273A (en) | On-line self-adaptation extraction method of metallurgy strip surface defect based on one-dimension information entropy |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191112 |
|
| WD01 | Invention patent application deemed withdrawn after publication |