CN108932722A - A kind of method of determining target single pixel profile - Google Patents
A kind of method of determining target single pixel profile Download PDFInfo
- Publication number
- CN108932722A CN108932722A CN201810765658.XA CN201810765658A CN108932722A CN 108932722 A CN108932722 A CN 108932722A CN 201810765658 A CN201810765658 A CN 201810765658A CN 108932722 A CN108932722 A CN 108932722A
- Authority
- CN
- China
- Prior art keywords
- pixel
- denoted
- point
- profile
- image
- 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.)
- Withdrawn
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
- G06T7/13—Edge detection
Abstract
The invention discloses a kind of methods of determining target single pixel profile, include the following steps: step 1: reading in the image for needing to detect vehicle, be denoted as I1, wherein the height of image is height, width width；Step 2: by I1It is input in full convolutional neural networks, obtains profile probability graph I2；Step 3: finding I2In the maximum pixel of grey scale pixel value be denoted as PL, gray value is denoted as PL(xL,yL), (xL,yL) indicate pixel PLCoordinate；The beneficial effects of the present invention are: the extraction operation of single pixel profile is carried out to target by using method of the invention, it is possible to prevente effectively from the interference for the background that target proximity edge is connected, and due to the characteristic of full convolutional network, the contour extraction of objects method is not influenced by color of object and illumination, target intrinsic colour and background close in the case where can also carry out locations of contours well, good contour extraction of objects effect can be obtained under complex environment.
Description
Technical field
The present invention relates to computer vision and machine learning field, it is specifically a kind of by combine full convolutional neural networks with
And the method that local configuration maximum probability calculates to determine target single pixel profile.
Background technique
With the development of image processing techniques, the contours extract of target object is increasingly by the concern of researchers, target
The extraction of profile plays an important role for the identification and positioning of target object, secondly, with the extensive hair of neural network
The acquisition of exhibition, the training sample of the network of Pixel-level classification is still a big problem, the single pixel profile of accurate objective contour
Extracting method can save manpower and the time required for frame sampling sheet.
It is current to propose different contour extraction of objects methods there are many scholar, it can be mainly divided into based on video frame method
With the objective contour localization method of morphological method, profile and border extracting method neural network based and be based on active contour
Contour extraction of objects method of model etc..
The technical solution being wherein closer to the present invention are as follows: document (Wang Rong, Yang Ning, Li Jin great waves single pixel human body contour outline
Study on Extraction Method [J] science and technology and engineering, 2014,14 (24): 252-255.) it proposes to use gray level image Background difference
Human region image is obtained, repair process then is carried out to human region image using the method for mathematical morphology, finally, utilizing
The human body contour outline of edge pixel neighborhood information extraction single pixel.(Wang Pei, Zhou Xin, Peng Rongkun wait jointing edge and region to document
Movable contour model SAR image contour extraction of objects [J] Journal of Image and Graphics, 2014,19 (7): 1095-1103.)
It proposes the New activity skeleton pattern of a kind of jointing edge movable contour model and regional activity skeleton pattern, carries out SAR image mesh
Target contours extract.(Cai's document gallops, Song Xiaoxiao, what is into flourish ox face contours extract algorithm based on computer vision and realization
[J] Journal of Agricultural Engineering, 2017,33 (11): 171-177..) it proposes to position ox face position with adaptive cascade detectors, it uses
Count the method that iterative model extracts ox face profile.This method acquires ox face direct picture, orients ox with tandem type detector
The position of face, and supervised gradient descent algorithm (supervised descent method, SDM), local binary is respectively adopted
Algorithm (local binary features, LBF) and active appearance models algorithm (fast active appearance
Model, FAAM) 3 kinds of algorithms be used to extract ox face profile.Document (Xie S, Tu Z.Holistically-Nested Edge
Detection [J] .International Conference on Computer Vision.2015:1395-1403.) it proposes
A method of based on convolutional neural networks.The detection for proposing the profile of convolutional neural networks progress end to end, has used more rulers
The method of degree carries out feature learning, directly exports the contour images of target.
In conclusion current objective contour determines method, there are the following shortcomings: 1) Clutter edge is more under complex background,
It is easy to cause the identification difficulty of objective contour；2) accurate positioning of the information meeting jamming target in target proximity region；3) it does not fill
Divide the information of itself using profile.
Summary of the invention
For the above problem present in existing objective contour localization method, the invention proposes one kind to pass through nerve net
The method that network and computer vision technique determine target single pixel profile.
Technical scheme is as follows:
A kind of method of determining target single pixel profile, which comprises the steps of:
Step 1: reading in the image for needing to detect vehicle, be denoted as I1, wherein the height of image is height, and width is
width；
Step 2: by I1It is input in full convolutional neural networks, obtains profile probability graph I2；
Step 3: finding I2In the maximum pixel of grey scale pixel value be denoted as PL, gray value is denoted as PL(xL,yL), (xL,
yL) indicate pixel PLCoordinate；
Step 4: finding central point PC；
Step 5: by image I2Edge point set be denoted as E={ Ei| i=1,2 ..., m, m=2width+2height },
Connection central point PC and Ei obtain straight-line segment, and the pixel collection on line segment is represented by Li={ (xj,yj) | j=1,
2,...,ni,ni=dist (PC,Ei), wherein dist indicates the distance of two o'clock, and calculation formula is such as shown in (4):
Wherein, (xa,ya) be a point coordinate and ordinate, (xb,yb) be respectively b point coordinate, niIndicate LiIn element
Number；
Step 6: traversal arbitrary collection Li, i=1,2 ..., m, if finding, to meet the pixel of formula (5) be marginal point, note
For Pi=(xi,yi), then it adds it in set Cont；The set Cont that all marginal points then found are constituted is represented by
Cont={ Pi| i=1,2 ..., Q }, wherein Q is the number of marginal point:
Wherein, α, β are the conditional coefficient being previously set, and meet alpha+beta=1；Value by formula (6) count
It obtains；
ΔDa=| dist (PC,PL)-dist(PC,a)| (6)
Step 7: removal image I3In be not belonging to pixel in edge point set Cont, the single pixel wheel of target can be obtained
Wide image I4。
A kind of method of determining target single pixel profile, which is characterized in that the step 4 specifically:
Step 4.1: according to formula (1) and (2) to profile probability graph I2Middle carry out binaryzation, and obtained binary map is denoted as
I3；
Wherein, I2(x, y) is I2The gray value of middle pixel (x, y), I3(x, y) is I3The gray scale of middle pixel (x, y)
Value；
Step 4.2: to I3In pixel value be 255 point position counted, obtain set be denoted as R={ (xi,yi)|i
=1,2 ..., n }, n is the number for the point that pixel value is 255；
Step 4.3: according to formula (3), obtaining central point PC；
The beneficial effects of the present invention are: being grasped by using the extraction that method of the invention carries out single pixel profile to target
Make, it is possible to prevente effectively from the interference for the background that target proximity edge is connected, and due to the characteristic of full convolutional network, the target
Contour extraction method is not influenced by color of object and illumination, target intrinsic colour and background close in the case where also can be very
Good carry out locations of contours, can obtain good contour extraction of objects effect under complex environment.
Detailed description of the invention
Fig. 1 is the vehicle image that the present invention implements that sample is chosen；
Fig. 2 is the profile probability distribution graph of the invention obtained after step 2 carries out full convolutional neural networks processing；
Fig. 3 is the bianry image obtained after filtering noise by step 4 in the embodiment of the present invention；
Fig. 4 is the relational graph between each point in the embodiment of the present invention in step 6；
Fig. 5 is to pass through the finally obtained vehicle single pixel contour images of step 7 in the embodiment of the present invention.
Specific embodiment
The single pixel contour method of the invention that sets the goal really is elaborated below with reference to embodiment.It should be appreciated that herein
Described specific embodiment is used only for explaining the present invention, is not intended to limit the present invention.
A kind of method of determining target single pixel profile of the invention, includes the following steps:
Step 1: reading in the image for needing to detect vehicle, be denoted as I1, in the present embodiment, Fig. 1 is I1, wherein scheming
The height of picture is height, width width；
Step 2: by I1It is input in full convolutional neural networks, obtains profile probability graph I2；In the present embodiment, Fig. 2 is
I2。
Step 3: finding the maximum pixel of the grey scale pixel value in Fig. 2 and be denoted as PL, gray value is denoted as PL(xL,yL)；
Step 4: finding central point PC, specifically:
Step 4.1: according to formula (1) and (2) to profile probability graph I2Middle carry out binaryzation, and obtained binary map is denoted as
I3, in the present embodiment, Fig. 3 is I3。
Wherein, I2(x, y) is the gray value of pixel (x, y) in Fig. 2, I3(x, y) is the gray scale of pixel (x, y) in Fig. 3
Value；
Step 4.2: the point position for being 255 to the pixel value in Fig. 3 counts, and obtains set and is denoted as R={ (xi,yi)|i
=1,2 ..., n }, n is the number for the point that pixel value is 255；
Step 4.3: according to formula (3), obtaining central point PC；
Step 5: the edge point set of Fig. 2 is denoted as E={ Ei| i=1,2 ..., m, m=2width+2height }, even
Meet central point PCWith EiStraight-line segment is obtained, the pixel collection on line segment is represented by Li={ (xj,yj) | j=1,2 ...,
ni,ni=dist (PC,Ei), wherein dist indicates the distance of two o'clock, and calculation formula is such as shown in (4):
Wherein, (xa,ya) be a point coordinate and ordinate, (xb,yb) be respectively b point coordinate, niIndicate LiIn element
Number.
Step 6: traversal arbitrary collection Li, i=1,2 ..., m, if finding, to meet the pixel of formula (5) be marginal point, note
For Pi=(xi,yi), then it adds it in set Cont；The set Cont that all marginal points then found are constituted is represented by
Cont={ Pi| i=1,2 ..., Q }, wherein Q is the number of marginal point:
Wherein, α, β are the conditional coefficient being previously set, and meet alpha+beta=1, in the present embodiment, α=0.7, β=0.3；Value be calculated by formula (6)；
ΔDa=| dist (PC,PL)-dist(PC,a)| (6)
Step 7: the single pixel profile of target can be obtained in the pixel being not belonging in edge point set Cont in removal Fig. 3
Image I4, in the present embodiment, the relational graph of each point is as shown in Figure 4.
In the present embodiment, by handling above, it can be seen that the single pixel profile of the target in Fig. 5 is accurately positioned out
Come.
Claims (2)
1. a kind of method of determining target single pixel profile, which comprises the steps of:
Step 1: reading in the image for needing to detect vehicle, be denoted as I1, wherein the height of image is height, and width is
width；
Step 2: by I1It is input in full convolutional neural networks, obtains profile probability graph I2；
Step 3: finding I2In the maximum pixel of grey scale pixel value be denoted as PL, gray value is denoted as PL(xL,yL), (xL,yL) table
Show pixel PLCoordinate；
Step 4: finding central point PC；
Step 5: by image I2Edge point set be denoted as E={ Ei| i=1,2 ..., m, m=2width+2height }, connection
Central point PCWith EiStraight-line segment is obtained, the pixel collection on line segment is represented by Li={ (xj,yj) | j=1,2 ..., ni,
ni=dist (PC,Ei), wherein dist indicates the distance of two o'clock, and calculation formula is such as shown in (4):
Wherein, (xa,ya) be a point coordinate and ordinate, (xb,yb) be respectively b point coordinate, niIndicate LiIn element
Number；
Step 6: traversal arbitrary collection Li, i=1,2 ..., m, if finding, to meet the pixel of formula (5) be marginal point, is denoted as Pi
=(xi,yi), then it adds it in set Cont；The set Cont that all marginal points then found are constituted is represented by Cont
={ Pi| i=1,2 ..., Q }, wherein Q is the number of marginal point:
Wherein, α, β are the conditional coefficient being previously set, and meet alpha+beta=1；Value calculated by formula (6)
It arrives；
ΔDa=| dist (PC,PL)-dist(PC,a)| (6)
Step 7: removal image I3In be not belonging to pixel in edge point set Cont, the single pixel profile diagram of target can be obtained
As I4。
2. a kind of method of determining target single pixel profile according to claim 1, which is characterized in that step 4 tool
Body are as follows:
Step 4.1: according to formula (1) and (2) to profile probability graph I2Middle carry out binaryzation, and obtained binary map is denoted as I3；
Wherein, I2(x, y) is I2The gray value of middle pixel (x, y), I3(x, y) is I3The gray value of middle pixel (x, y)；
Step 4.2: to I3In pixel value be 255 point position counted, obtain set be denoted as R={ (xi,yi) | i=1,
2 ..., n }, n is the number for the point that pixel value is 255；
Step 4.3: according to formula (3), obtaining central point PC；
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810765658.XA CN108932722A (en) | 2018-07-12 | 2018-07-12 | A kind of method of determining target single pixel profile |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810765658.XA CN108932722A (en) | 2018-07-12 | 2018-07-12 | A kind of method of determining target single pixel profile |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108932722A true CN108932722A (en) | 2018-12-04 |
Family
ID=64447115
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810765658.XA Withdrawn CN108932722A (en) | 2018-07-12 | 2018-07-12 | A kind of method of determining target single pixel profile |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108932722A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436255A (en) * | 2008-11-28 | 2009-05-20 | 华中科技大学 | Method for extracting remarkable configuration in complicated image |
| CN107273905A (en) * | 2017-06-14 | 2017-10-20 | 电子科技大学 | A kind of target active contour tracing method of combination movable information |
-
2018
- 2018-07-12 CN CN201810765658.XA patent/CN108932722A/en not_active Withdrawn
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436255A (en) * | 2008-11-28 | 2009-05-20 | 华中科技大学 | Method for extracting remarkable configuration in complicated image |
| CN107273905A (en) * | 2017-06-14 | 2017-10-20 | 电子科技大学 | A kind of target active contour tracing method of combination movable information |
Non-Patent Citations (1)
| Title |
|---|
| FEI GAO等: "A Two-stage Vehicle Type Recognition Method", 《 2018 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS(IJCNN)》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105740945B (en) | A kind of people counting method based on video analysis | |
| CN104299260B (en) | Contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration | |
| CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
| CN102214355B (en) | Cutout method of clothing display material | |
| CN105320917B (en) | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network | |
| CN106991686B (en) | A kind of level set contour tracing method based on super-pixel optical flow field | |
| CN109446922B (en) | Real-time robust face detection method | |
| CN107230203A (en) | Casting defect recognition methods based on human eye vision attention mechanism | |
| CN104077775A (en) | Shape matching method and device combined with framework feature points and shape contexts | |
| CN107895379A (en) | The innovatory algorithm of foreground extraction in a kind of video monitoring | |
| CN105389581A (en) | Germinated rice germ integrity intelligent identification system and identification method thereof | |
| CN109784216B (en) | Vehicle-mounted thermal imaging pedestrian detection Rois extraction method based on probability map | |
| CN112396655B (en) | Point cloud data-based ship target 6D pose estimation method | |
| CN107292896A (en) | Contour extraction method based on Snake models | |
| CN109886271A (en) | It merges deep learning network and improves the image Accurate Segmentation method of edge detection | |
| CN115797813B (en) | Water environment pollution detection method based on aerial image | |
| CN109781737A (en) | A kind of detection method and its detection system of hose surface defect | |
| CN108805924A (en) | A kind of lily picking independent positioning method and system | |
| CN104050674B (en) | Salient region detection method and device | |
| CN106709432B (en) | Human head detection counting method based on binocular stereo vision | |
| CN107729863B (en) | Human finger vein recognition method | |
| Shire et al. | A review paper on: agricultural plant leaf disease detection using image processing | |
| CN116763295B (en) | Livestock scale measuring method, electronic equipment and storage medium | |
| CN111815640B (en) | Memristor-based RBF neural network medical image segmentation algorithm | |
| CN105139452B (en) | A kind of Geologic Curve method for reconstructing based on image segmentation |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20181204 |