CN106327494B - A kind of pavement crack image automatic testing method - Google Patents
A kind of pavement crack image automatic testing method Download PDFInfo
- Publication number
- CN106327494B CN106327494B CN201610717481.7A CN201610717481A CN106327494B CN 106327494 B CN106327494 B CN 106327494B CN 201610717481 A CN201610717481 A CN 201610717481A CN 106327494 B CN106327494 B CN 106327494B
- Authority
- CN
- China
- Prior art keywords
- sub
- pictures
- image
- crack image
- crack
- 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
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- 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
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30184—Infrastructure
Abstract
The invention discloses a kind of pavement crack image automatic testing methods, including every width pavement crack image and its sub-pictures are associated with corresponding label, each sub-pictures have its feature description vectors, multiple sub-pictures, which are separated into, by sub-pictures forms pyramid structure, and characteristics of image description vectors in crack are finally formed according to pyramid pattern, feature description vectors are transformed into Fourier space, and calculate crack image in the Oscillation Amplitude of Fourier space；It is filtered using controllable filter, then looks for Weak Classifier, multiple crack image Weak Classifiers are combined to obtain strong classifier by its weight, finally detected automatically with crannied pavement image.It is compared with the traditional method, the crack characteristics of image description vectors obtained in the present invention can characterize the texture speciality of crack image, and be detected using improved Adaboost fracture image, be significantly improved in detection accuracy and speed.
Description
Technical field
The present invention relates to pavement crack field of image detection, more particularly to a kind of pavement crack image side of detection automatically
Method.
Background technique
In Crack Detection field, some research work depend on unsupervised segmentation: by statistical analysis, structural analysis,
Space filtering, the analysis based on model and classification method neural network based or K- means clustering method, people like always
With the simple algorithm of non-formaldehyde finishing, the step of these do not learn.Now, many of real-time grading technology is applied,
And these classifications prepare, and do not propose so far in crack field also, in addition, each method can be for the overall situation,
Can be the region for image local, but when with global description, we can from the shape of fractured model and it is several where
Face considers, but most of the time, crack accounts for the seldom ratio of entire image, may be missed by global method detection
Crack, it also illustrates such a fact, thus the characteristics of being directed to crack image be difficult it is high-precision by global analysis mode
The detection crack image of degree.
Therefore, in the present invention in order to improve local testing result, we have proposed changing based on multiple classifiers combinations
Into Adaboost algorithm, each sub-pictures have its feature description vectors, are separated into multiple sub-pictures by sub-pictures and form golden word
Tower structure, and feature description vectors are finally formed according to pyramid pattern, feature description vectors can describe the line of crack image
Feature is managed, and image is denoised and filtered by Fourier transform and controllable filter, finally by multiple weak typings
Device combines, and ultimately forms strong classifier, has achieved the purpose that detect crack image, in the present invention feature that obtains describe to
Amount can characterize the texture speciality of crack image, and be detected using improved Adaboost road pavement crack image, examine
It surveys in accuracy and speed and is significantly improved, several classification methods are the analyses based on part, and partial analysis method identifies
Crack is a very useful method.Moreover, local classification makes good use of the extraordinary classifier of proposition: structure classifier
With the classifier based on model.
Summary of the invention
The present invention in view of the above-mentioned problems existing in the prior art, proposes a kind of pavement crack image automatic testing method.
The specific technical solution of the present invention be achieved in that comprising the following specific steps
Step 1: every width pavement crack image is associated with corresponding label, and every width crack image is divided into n sub-pictures,
Sub-pictures can be sub-divided into sub-pictures, form pyramid structure, and finally form feature description vectors according to pyramid pattern,
Each sub-pictures are associated with corresponding label, labelCorresponding to thering is crack to be 0 in sub-pictures and free from flaw is 1,
Each sub-pictures have its feature description vectors, it can be the series connection of a describer or several describers.
Step 2: sub-pictures are converted into Fourier space, first amplitude of the calculating crack image in Fourier space, grid
Size selectionThis size can enhance signal by inhibiting noise, finally calculate each crack image subgraph
The amplitude of piece；
Step 3: being filtered using controllable filter, and controllable filter uses three linear group of basis filters
Conjunction obtains, in addition, controllable filter can assess the relative angle of each pixel according to pixel distribution, in filtered image
Each pixel both correspond to the minimum value in controllable filter window, variance parameter determines the size of controllable filter,
Controllable filter is made of three parameters, adjusts controllable filter and then to state modulator, utilizes controllable filter
Wave device fracture image is handled.
Step 4: the Weak Classifier that selection is passed through to filtered crack image dataFind suitable weight
It is preferably minimized error rate:
Weight is calculated:
The test errors rate on an independent database repeats step if error rate can reduce in n times test
Four, finally obtain strong classifier:
Wherein indicate strong classifier, indicate the t times it is trained when the i-th width picture weight,It indicates i-th in x picture
Corresponding j-th of Weak Classifier, parameter in width pictureParameter value when being trained the t times,It is the label letter of the i-th width picture
Breath value,Indicate the t times it is trained when standard value, strong classifier can automatically detect out with crannied pavement image.
Preferably, feature description vectors described in step 1 include by the histogram calculation to sub-pictures or being based on
Structural element is applied to be obtained in the morphological transformation of sub-pictures.
Preferably, further include initialization weight in step 3:
It preferably, further include cross validation in step 4, cross validation is statistics partition data, i.e., in sub-pictures, to complete
Portion's image is trained, although subset finally remains preliminary analysis data, initial data is divided into 8 groups, cross validation weight
It is eight times multiple.
Compared with prior art, the present invention is based on the improvement Adaboost algorithms of multiple classifiers combinations, he relies on structure
Analysis, multiple Weak Classifiers are combined, and ultimately form strong classifier, and pass through Fourier transform and controllable filter
Image is denoised and is filtered, has achieved the purpose that detect crack image, overall structure is simple, and at low cost, deployment is fast；At any time
It uses everywhere, more light, use is more flexible and convenient.
Detailed description of the invention
Fig. 1 is that crack image is divided into sub-pictures and the Principle of Process figure as vector.
Fig. 2 is crack image gray-scale transformation figure.
Fig. 3 (a) is primary fissure seam image, is (b) image after denoising, is (c) controllable filter treated crack pattern
Picture.
Specific embodiment
Specific implementation case of the invention is illustrated with reference to the accompanying drawing:
As shown in Figure 1, crack image is divided into multiple subgraphs, finally the feature vector of image and subgraph is accumulated
One vector indicates crack pattern picture, and the texture structure of crack image can show, by every width pavement crack image with it is right
Label is answered to be associated with, and every width crack image is divided into n sub-pictures, sub-pictures can be sub-divided into sub-pictures, form pyramid knot
Structure, and feature description vectors are finally formed according to pyramid pattern, each sub-pictures are associated with corresponding label, labelCorresponding to having crack to be 0 in sub-pictures and free from flaw is 1, each sub-pictures have its feature description vectors, it
It can be the series connection of a describer or several describers.
Show that the textural characteristics description figure of a width crack image of selection, t indicate threshold value in Fig. 2, T indicates optimal threshold, p
Indicate crack pixel value size, p (z) illustrates that pixel value distribution curve, q (z) indicate each section accounting, and sub-pictures are converted into Fu
In vane space, first calculating Fourier's amplitude image, mean amplitude of tide is in gridThis size can be by inhibiting to make an uproar
Sound enhances signal, finally calculates the logarithmic mean value of each sub-pictures.
It being filtered using controllable filter, controllable filter is obtained using three basis filters linear combinations,
In addition, controllable filter can assess the relative angle of each pixel, each of filtered image according to pixel distribution
Pixel both corresponds to the minimum value in controllable filter window, and variance parameter determines the size of controllable filter, can control
Filter be made of three parameters, adjust controllable filter and then to state modulator.
The Weak Classifier that selection is passed through to filtered crack image dataFind suitable weightMake error rate
It is preferably minimized:
Weight is calculated:
The test errors rate on an independent database repeats step if error rate can reduce in n times test
Four, finally obtain strong classifier:
Wherein indicate strong classifier,Indicate the t times it is trained when the i-th width picture weight,It indicates in x picture
Corresponding j-th of Weak Classifier, parameter in i-th width pictureParameter value when being trained the t times,It is the label of the i-th width picture
The value of information,Indicate the t times it is trained when standard value, strong classifier can automatically detect out with crannied pavement image.
Fig. 3 (a) is the pavement crack original image taken, and Fig. 3 (b) indicates the crack pattern after slightly denoising
Picture, Fig. 3 (c) indicate the crack image after controlling filter filtering.
Preferably, feature description vectors described in step 1 include by the histogram calculation to sub-pictures or being based on
Structural element is applied to be obtained in the morphological transformation of sub-pictures.
Preferably, further include initialization weight in step 3:
It preferably, further include cross validation in step 4, cross validation is statistics partition data, i.e., in sub-pictures, to complete
Portion's image is trained, although subset finally remains preliminary analysis data, initial data is divided into 8 groups, cross validation weight
It is eight times multiple.
Compared with prior art, the present invention is based on the improvement Adaboost algorithms of multiple classifiers combinations, he relies on structure
Analysis, multiple Weak Classifiers are combined, and ultimately form strong classifier, and pass through Fourier transform and controllable filter
Image is denoised and is filtered, has achieved the purpose that detect crack image, overall structure is simple, and at low cost, deployment is fast；At any time
It uses everywhere, more light, use is more flexible and convenient.
It limits the scope of implementation of the present invention, i.e., generally according to letter made by scope of the present invention patent and invention description content
Single equivalent changes and modifications, all still remain within the scope of the patent.
Claims (4)
1. a kind of pavement crack image automatic testing method, which is characterized in that comprising the following specific steps
Step 1: every width pavement crack image is associated with corresponding label, and every width crack image is divided into n sub-pictures, subgraph
Piece is sub-divided into sub-pictures, forms pyramid structure, and finally forms feature description vectors, each subgraph according to pyramid pattern
Piece is associated with corresponding label, 1 ∈ of label { 0,1 }, and corresponding to having crack to be 0 in sub-pictures and free from flaw is 1, each sub-pictures have
Its feature description vectors, it is the series connection of a describer or several describers；
Step 2: sub-pictures are converted into Fourier space, first amplitude of the calculating crack image in Fourier space, size of mesh opening
Selection 17 × 17, this size enhance signal by inhibiting noise, finally calculate the amplitude of each crack image sub-pictures；
Step 3: being filtered using controllable filter, and controllable filter is obtained using three basis filters linear combinations
It arrives, in addition, controllable filter assesses the relative angle of each pixel, each picture in filtered image according to pixel distribution
Element both corresponds to the minimum value in controllable filter window, and variance parameter determines the size of controllable filter, controllable
Filter is made of three parameters, adjusts controllable filter and then to state modulator, utilizes controllable filter counterincision
Seam image is handled；
Step 4: pass through the Weak Classifier h of selection to filtered crack image dataj(xi) find suitable weightMake mistake
Rate is preferably minimized:
Weight is calculated:
The test errors rate on an independent database repeats step 4 if error rate can reduce in n times test, most
After obtain strong classifier:
Wherein,Indicate the t times it is trained when the i-th width picture weight, hj(xi) indicate in x picture corresponding the in the i-th width picture
J Weak Classifier, ZtIndicate the t times it is trained when standard value, strong classifier can automatically detect out with crannied road surface figure
Picture.
2. a kind of pavement crack image automatic testing method according to claim 1, it is characterised in that: described in step 1
Feature description vectors include applying the form in sub-pictures by the histogram calculation to sub-pictures or based on structural element
Transformation is learned to obtain.
3. a kind of pavement crack image automatic testing method according to claim 2, it is characterised in that: also wrapped in step 3
Include initialization weight:
4. a kind of pavement crack image automatic testing method according to claim 2, it is characterised in that: also wrapped in step 4
Cross validation is included, cross validation is that statistics partition data is trained all images that is, in sub-pictures, although subset is most
After remain preliminary analysis data, initial data are divided into 8 groups, cross validation is repeated eight times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610717481.7A CN106327494B (en) | 2016-08-24 | 2016-08-24 | A kind of pavement crack image automatic testing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610717481.7A CN106327494B (en) | 2016-08-24 | 2016-08-24 | A kind of pavement crack image automatic testing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106327494A CN106327494A (en) | 2017-01-11 |
| CN106327494B true CN106327494B (en) | 2019-02-15 |
Family
ID=57790570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610717481.7A Active CN106327494B (en) | 2016-08-24 | 2016-08-24 | A kind of pavement crack image automatic testing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106327494B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222609A (en) * | 2019-05-24 | 2019-09-10 | 江西理工大学 | A kind of wall body slit intelligent identification Method based on image procossing |
| CN110473187B (en) * | 2019-08-08 | 2022-03-22 | 武汉光谷卓越科技股份有限公司 | Object-oriented line scanning three-dimensional pavement crack extraction method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945548A (en) * | 2012-11-20 | 2013-02-27 | 成都晶石石油科技有限公司 | Directional pyramid filtering-based image processing method and device |
| CN105320962A (en) * | 2015-10-21 | 2016-02-10 | 东南大学 | Pavement damage type identification method based on classifier ensemble |
| WO2016092783A1 (en) * | 2014-12-12 | 2016-06-16 | Canon Kabushiki Kaisha | Information processing apparatus, method for processing information, discriminator generating apparatus, method for generating discriminator, and program |
-
2016
- 2016-08-24 CN CN201610717481.7A patent/CN106327494B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945548A (en) * | 2012-11-20 | 2013-02-27 | 成都晶石石油科技有限公司 | Directional pyramid filtering-based image processing method and device |
| WO2016092783A1 (en) * | 2014-12-12 | 2016-06-16 | Canon Kabushiki Kaisha | Information processing apparatus, method for processing information, discriminator generating apparatus, method for generating discriminator, and program |
| CN105320962A (en) * | 2015-10-21 | 2016-02-10 | 东南大学 | Pavement damage type identification method based on classifier ensemble |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106327494A (en) | 2017-01-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102117413B (en) | Method for automatically filtering defective image based on multilayer feature | |
| CN100458831C (en) | Human face model training module and method, human face real-time certification system and method | |
| CN107103317A (en) | Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution | |
| CN101866427A (en) | Method for detecting and classifying fabric defects | |
| CN107742099A (en) | A kind of crowd density estimation based on full convolutional network, the method for demographics | |
| CN103218831A (en) | Video moving target classification and identification method based on outline constraint | |
| WO2021228088A1 (en) | Method for recognizing haze concentration in haze image | |
| CN103440476A (en) | Locating method for pupil in face video | |
| CN103942577A (en) | Identity identification method based on self-established sample library and composite characters in video monitoring | |
| CN104182772A (en) | Gesture recognition method based on deep learning | |
| CN106250936A (en) | Multiple features multithreading safety check contraband automatic identifying method based on machine learning | |
| CN107368778A (en) | Method for catching, device and the storage device of human face expression | |
| Salve et al. | Iris recognition using SVM and ANN | |
| CN106653032A (en) | Animal sound detecting method based on multiband energy distribution in low signal-to-noise-ratio environment | |
| CN105893946A (en) | Front face image detection method | |
| CN105718889A (en) | Human face identity recognition method based on GB(2D)2PCANet depth convolution model | |
| CN103886589A (en) | Goal-oriented automatic high-precision edge extraction method | |
| CN110503613A (en) | Based on the empty convolutional neural networks of cascade towards removing rain based on single image method | |
| CN104504721A (en) | Unstructured road detecting method based on Gabor wavelet transformation texture description | |
| CN110728185B (en) | Detection method for judging existence of handheld mobile phone conversation behavior of driver | |
| CN105894018B (en) | Classification of Polarimetric SAR Image method based on depth multi-instance learning | |
| CN110415260A (en) | Smog image segmentation and recognition methods based on dictionary and BP neural network | |
| CN109241814A (en) | Pedestrian detection method based on YOLO neural network | |
| CN102236786B (en) | Light adaptation human skin colour detection method | |
| CN109472733A (en) | Image latent writing analysis method based on convolutional neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20190104 Address after: 430000 No. 5 Hanyang Erqiao Road, Wuhan City, Hubei Province Applicant after: Hubei Highway and Waterway Engineering Consulting and Supervision Company Address before: 262700 Weifang science and Technology College, 1299 Jin Guang street, Shouguang City, Weifang, Shandong Applicant before: Zhang Jiali |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |