CN103048331B - Printing defect detection method based on flexible template registration - Google Patents
Printing defect detection method based on flexible template registration Download PDFInfo
- Publication number
- CN103048331B CN103048331B CN201210558063.XA CN201210558063A CN103048331B CN 103048331 B CN103048331 B CN 103048331B CN 201210558063 A CN201210558063 A CN 201210558063A CN 103048331 B CN103048331 B CN 103048331B
- Authority
- CN
- China
- Prior art keywords
- image
- registration
- flexible
- pixel
- template
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A printing defect detection method based on flexible template registration adopts rigid registration to carry out initial positioning, takes the overall similarity measure and the coordinate difference degree of all pixels of a training sample after the rigid initial registration and a standard image as optimization targets, obtains a flexible registration image from the training sample to the standard image, obtains a high-value template image and a low-value template image by learning a training sample set after the flexible registration, and finally respectively judges whether the gray scale and/or the color of each pixel in an image to be detected exceed the range of the high-value template and the low-value template, thereby obtaining a defect image. The invention ensures the image detection precision and simultaneously maintains the tolerance capability to unstable factors such as deformation and the like, thereby not only reducing the false alarm probability of machine vision detection, but also reducing the waste leakage risk and being greatly helpful for improving the detection precision.
Description
Technical field
The present invention relates to the printing defects detection method that a kind of misinformation probability is low, the useless risk of leakage is little, belong to printing technology.
Background technology
Existing printing defects detection system mainly adopts rigid template registration technology.The method of this quasi-tradition, by carrying out multi-level images search location to current image to be checked, is chosen specific image block as search locating template image, is realized search location, thus determine defective locations through translation.The advantage of these class methods is that data processing calculates simply, and detection speed is fast, when image stabilization, has good detection sensitivity.But the latent instability factor of printed matter in manufacture and testing process is difficult to avoid, and these factors all can cause the free geometric deformation of image.Labile factor is mainly derived from three aspects: the relative displacement 1) between many impressions; 2) product medium is being produced and waving in transmitting procedure; 3) inherent error of printed matter itself and imaging lens distortion correction.And in most cases, these of printed matter are inconsistent is in the allowed band of product quality, does not namely form defect.But due to the employing of conventional template registration technology is Rigid Registration pattern, more weak to the geometry inconsistency tolerance of image, therefore inevitable generation in testing process is reported by mistake in a large number, and its consequence causes the huge waste of product and human resources, and finally cause production efficiency low.
Summary of the invention
The object of the invention is to for prior art drawback, a kind of printing defects detection method based on flexible template registration is provided, to improve accuracy of detection, while useless risk is leaked in reduction, reduce misinformation probability.
Problem of the present invention realizes with following technical proposals:
A kind of printing defects detection method based on flexible template registration, described method adopts Rigid Registration to carry out initial alignment, and the overall similarity of the entire pixels of the training sample after rigidity initial registration and standard picture is estimated with coordinate difference degree as optimization aim, try to achieve the flexible registering images of training sample to standard picture, then by the training sample set after the flexible registration of study, high level template image and low value template image is obtained; Described high level template image and low value template image are determined by following principle:
Add up several specification product images, get the pixel that same position brightness is the highest, the image be made up of these pixels is high level template image;
Add up several specification product images, get the pixel that same position brightness is minimum, the image be made up of these pixels is low value template image;
Finally judge whether gray scale and/or the color of each pixel in image to be checked exceed high level template and low value template scope respectively, be off-limitsly defect image.
The above-mentioned printing defects detection method based on flexible template registration, the concrete detecting step of printing defects is as follows:
A. screen illumination, mass colour is normal, print face is smooth, without the product of knuckle, fold and chromatography defect as standard picture;
B. Rigid Registration technology is adopted to realize the initial alignment of image, the position that namely search is maximum with standard picture similarity in strain image;
C. training sample is to the flexible registration of standard picture:
For any one pixel of the training sample completed after rigidity initial registration, near the same coordinate of standard picture, find a corresponding pixel to form pixel pair with it, record each pixel between coordinate difference, overall similarity right for entire pixels is estimated and coordinate difference degree;
The similarity measure of described two objects refers to the degree of their mutual vicinities in Euclidean space, and similarity measure calculates by Euclidean distance, related coefficient or information entropy mode;
Described coordinate difference degree be pixel between the summation of distance in two-dimensional space coordinate system in units of pixel;
As optimization aim, solve this optimization problem, obtain the flexible registering images of training sample to standard picture;
D. learn the training sample set after flexible registration, obtain high low value template:
Collect a series of normal printed matter image as training sample set, choose wherein one as standard picture, other all samples are carried out deformation normalization as benchmark, then the dynamic range of the gray scale concentrated of statistical sample and color respectively in all pixels, obtains high level template image and low value template image;
E. high low value template is to the flexible registration of image to be checked:
For an image to be checked, using this image as benchmark, high low value template image is carried out flexible registration;
F. to each pixel in image to be checked, within judging whether its gray scale and/or color are positioned at the marginal range defined by the high low value template image after being out of shape respectively, if exceed this tolerance limit, then this pixel is judged as defect.
The above-mentioned printing defects detection method based on flexible template registration, the optimization object function that training sample adopts to the flexible registration of standard picture is:
C = -C
similarity+C
deformation，
Wherein C
similarityfor similar energies item, C
deformationfor deformation energy quantifier;
Similarity measure adopts Euclidean distance formula:
，
Wherein
for image subject to registration,
for target image,
for the number of pixels that image subject to registration and target image comprise;
for the gray-scale value of i-th pixel in target image,
for the gray-scale value of i-th pixel in image subject to registration.
The present invention carries out image flexible registration to the free deformation of image, high low value template image after flexible registration had both maintained the due grain details of image, contain again brightness and the color dynamic range information of each pixel of training sample, while ensureing image detection accuracy, maintain the tolerance to labile factors such as deformation.The present invention not only reduces the misinformation probability of Machine Vision Detection, and reduces Lou useless risk, has great help to raising accuracy of detection.
Accompanying drawing explanation
Below in conjunction with accompanying drawing, the invention will be further described.
Fig. 1 is image subject to registration (left side) and target image (right side) (having small deformation between 3 pixels);
Fig. 2 is Rigid Registration process between image subject to registration and target image in Fig. 1;
Fig. 3 is flexible registration process between image subject to registration and target image in Fig. 1;
Fig. 4 is the flexible registration of training sample to standard picture;
Fig. 5 is the acquisition process of high low value template image;
Fig. 6 is the flexible registration of low value template to image to be checked;
Fig. 7 is the flexible registration of high level template to image to be checked;
Fig. 8 is defect inspection process schematic diagram;
Fig. 9 is the high low value template that Rigid Registration training obtains;
Figure 10 is the high low value template of training acquisition after flexible registration;
Figure 11 is the comparison of employing two kinds of algorithm errors extraction effects: (left side) defect image; (in) defect of traditional method for extracting; The defect that (right side) this method is extracted.
In literary composition, each list of reference numerals is: C is optimization object function, C
similarityfor similar energies item, C
deformationfor deformation energy quantifier,
for Euclidean distance,
for image subject to registration,
for target image,
for the number of pixels that image subject to registration and target image comprise,
for the gray-scale value of i-th pixel in target image,
for the gray-scale value of i-th pixel in image subject to registration.
Embodiment
The present invention adopts Rigid Registration to carry out initial alignment, for completing the image subject to registration after rigidity initial registration, wherein any one pixel, the pixel that one corresponding can be found near target image (standard picture, template) same coordinate, form pixel pair with it, similarity measure between them (similarity measure of two objects refers to the degree of their mutual vicinities in Euclidean space, and conventional similarity measure computing method have Euclidean distance, related coefficient or information entropy etc.) is higher.Record each pixel between coordinate difference.Overall similarity right for entire pixels is estimated and coordinate difference degree as optimization object function:
C = -C
similarity+C
deformation ，
Wherein C
similarityfor similar energies item, C
deformationfor deformation energy quantifier;
Similarity measure adopts Euclidean distance formula:
，
Wherein
,
be respectively image subject to registration and target image, comprising number of pixels is
;
with
be respectively the grey scale pixel value in image subject to registration and target image, span is between 0 to 255.
Solve this optimization problem, the flexible registering images to target image subject to registration can be obtained.
Fig. 1 has the image subject to registration (left side) of miniature deformation and target image (right side) between 3 pixels, Fig. 2 is Rigid Registration process between image subject to registration and target image in Fig. 1, as can be seen from Figure 2, pixel C and C ' is because the incomplete registration of rigid translation.Fig. 3 is flexible registration process between image subject to registration and target image, 3 pixels registration all completely.
This detection method comprises the steps:
1, selection standard image
Screen an illumination, mass colour is normal, print face is smooth, without the product of the flaws such as knuckle, fold and chromatography defect as standard picture.Standard picture is using the reference as training sample initial alignment, flexible registration.
2, rigid image initial registration
Before carrying out flexible image registration, first to utilize the image rigid registration method based on grey similarity that strain image and standard picture are carried out initial alignment, i.e. search and the maximum position of standard picture similarity in strain image, object be the deformation distance of pixel is adjusted to suitable scope (as
pixel) namely within deformation range, reduce the room and time complexity of follow-up flexible image registration process.
3, training sample is to the flexible registration of standard picture
For the training sample completed after rigidity initial registration, wherein any one pixel, can find the pixel that corresponding near standard picture same coordinate, form pixel pair with it, the similarity measure (absolute value error, error sum of squares, mutual information etc.) between them is higher.Record each pixel between coordinate difference.Overall similarity right for entire pixels is estimated and coordinate difference degree as optimization aim, solve this optimization problem, the flexible registering images of training sample to standard picture can be obtained.
4, high low value template is learnt
Collect a series of normal printed matter image as training sample set, choose wherein one as standard picture, other all samples are carried out deformation normalization as benchmark, then the dynamic range of the gray scale concentrated of statistical sample and color respectively in all pixels, namely obtains high level template image and low value template image.This two width image is using as detecting the whether qualified reference frame of image pixel to be checked.
5, high low value template is to the flexible registration of image to be checked
For an image to be checked, using this image as benchmark, high low value template image is carried out flexible registration.The reason done like this may there is printing defects in image to be checked, the error of calculation or information dropout may be there is in the process of the data processings such as deformation reduction, in order to avoid these situations have influence on defect itself, thus reduction system sensitivity, therefore we are by the geometric shape of template deformation to image to be checked.
6, defects detection
To each pixel in image to be checked, judge whether its gray scale and/or color are positioned at by within being out of shape marginal range that rear high low value template image defines, if exceed this tolerance limit, then this pixel are judged as defect respectively.
By flexible template registration technology, new detection method under the condition of keeping system sensitivity, tool
There is stronger deformation tolerance, improve the robustness of printing quality checking system.On the other hand, because training sample is registrated to standard picture by flexibility, maintain the grain details of high low value template image, reduce the leakage brought because height template is fuzzy and to give up risk.As can be seen from Fig. 9 and Figure 10, the high low value template of flexible registration maintains meticulousr grain details.
Claims (2)
1. the printing defects detection method based on flexible template registration, it is characterized in that, described method adopts Rigid Registration to carry out initial alignment, and the overall similarity of the entire pixels of the training sample after rigidity initial registration and standard picture is estimated with coordinate difference degree as optimization aim, try to achieve the flexible registering images of training sample to standard picture, then by the training sample set after the flexible registration of study, high level template image and low value template image is obtained; Described high level template image and low value template image are determined by following principle:
Add up several specification product images, get the pixel that same position brightness is the highest, the image be made up of these pixels is high level template image;
Add up several specification product images, get the pixel that same position brightness is minimum, the image be made up of these pixels is low value template image;
Finally judge whether gray scale and/or the color of each pixel in image to be checked exceed high level template and low value template scope respectively, be off-limitsly defect image;
The concrete detecting step of printing defects is as follows:
A. screen illumination, mass colour is normal, print face is smooth, without the product of knuckle, fold and chromatography defect as standard picture;
B. adopt Rigid Registration technology to realize the initial alignment of image, the position that namely search is maximum with standard picture similarity in strain image, so-called strain image refers to the image that the geometric configuration caused because of the product medium factor of waving is distorted herein;
C. training sample is to the flexible registration of standard picture:
For the training sample completed after rigidity initial registration, any one pixel after training sample is out of shape according to the outward appearance of standard picture, near the same coordinate of standard picture, find a corresponding pixel to form pixel pair with it, record each pixel between coordinate difference, overall similarity right for entire pixels is estimated and coordinate difference degree as optimization aim, solve this optimization problem, obtain the flexible registering images of training sample to standard picture;
D. learn the training sample set after flexible registration, obtain high level template and low value template:
Collect a series of normal printed matter image as training sample set, choose wherein one as standard picture, other all samples are carried out deformation normalization as benchmark, then the dynamic range of the gray scale concentrated of statistical sample and color respectively in all pixels, obtains high level template image and low value template image;
E. high low value template is to the flexible registration of image to be checked:
For an image to be checked, using this image as benchmark, high level template image, low value template image are carried out flexible registration, make high level template image, low value template image is out of shape according to the outward appearance of image to be checked;
F. to each pixel in image to be checked, within judging whether its gray scale and/or color are positioned at the marginal range defined by the high level template image after being out of shape and low value template image respectively, if exceed this tolerance limit, then this pixel is judged as defect.
2. a kind of printing defects detection method based on flexible template registration according to claim 1, is characterized in that, the optimization object function that training sample adopts to the flexible registration of standard picture is:
C = -C
similarity+C
deformation，
Wherein C
similarityfor similar energies item, C
deformationfor deformation energy quantifier;
Similarity measure adopts Euclidean distance formula:
Wherein
for image subject to registration,
for target image,
for the number of pixels that image subject to registration and target image comprise;
for the gray-scale value of i-th pixel in target image,
for the gray-scale value of i-th pixel in image subject to registration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210558063.XA CN103048331B (en) | 2012-12-20 | 2012-12-20 | Printing defect detection method based on flexible template registration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210558063.XA CN103048331B (en) | 2012-12-20 | 2012-12-20 | Printing defect detection method based on flexible template registration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103048331A CN103048331A (en) | 2013-04-17 |
| CN103048331B true CN103048331B (en) | 2015-09-23 |
Family
ID=48061046
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210558063.XA Active CN103048331B (en) | 2012-12-20 | 2012-12-20 | Printing defect detection method based on flexible template registration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103048331B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103674962B (en) * | 2013-09-27 | 2016-04-27 | 北京中钞钞券设计制版有限公司 | Plate quality detection system and method |
| CN104820987B (en) * | 2015-04-30 | 2017-11-17 | 中国电子科技集团公司第四十一研究所 | A kind of method based on optical imagery and microwave imagery detection target scattering performance deficiency |
| CN105241894B (en) * | 2015-08-28 | 2018-01-19 | 北京大恒图像视觉有限公司 | A kind of template method for registering surveyed for the product examine of multiple operation flexible printing |
| CN107278318B (en) * | 2017-02-07 | 2019-10-29 | 深圳前海达闼云端智能科技有限公司 | A kind of blind guiding system |
| CN106918602B (en) * | 2017-04-13 | 2020-01-31 | 北京海风智能科技有限责任公司 | robust surface defect detection method based on machine vision |
| CN108037142B (en) * | 2017-12-04 | 2021-01-19 | 江苏维普光电科技有限公司 | Mask optical defect detection method based on image gray value |
| CN107945172A (en) * | 2017-12-08 | 2018-04-20 | 博众精工科技股份有限公司 | A kind of character detection method and system |
| CN111197958A (en) * | 2018-11-19 | 2020-05-26 | 北新集团建材股份有限公司 | Online detection system and method for plate appearance |
| CN109712186B (en) * | 2018-12-11 | 2021-10-22 | 上海联影医疗科技股份有限公司 | Method, computer device and storage medium for delineating a region of interest in an image |
| CN109916913A (en) * | 2019-04-04 | 2019-06-21 | 哈尔滨理工大学 | A kind of intelligence manufacture product identification and detection method based on machine vision |
| CN112085707B (en) * | 2020-08-18 | 2024-09-06 | 北京华夏视科技术股份有限公司 | Printed matter color detection method, device, equipment and medium |
| CN112489002B (en) * | 2020-11-24 | 2024-05-31 | 广东省电子技术研究所 | Defect detection method, device, equipment and storage medium |
| CN114742835B (en) * | 2022-06-13 | 2022-09-02 | 新乡职业技术学院 | Test equipment for performance of liquid crystal elastomer material array |
| CN114882037B (en) * | 2022-07-11 | 2022-10-21 | 北京中科慧眼科技有限公司 | Image defect detection method and system based on dynamic printing mask |
| CN115311266B (en) * | 2022-10-10 | 2023-04-14 | 南通富兰妮纺织品有限公司 | Printing and dyeing identification method in textile printing and dyeing process |
| CN116754566B (en) * | 2023-08-17 | 2023-10-31 | 绍兴旭源新材料科技有限公司 | Flexible folding screen protective film detection method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102236897A (en) * | 2011-05-09 | 2011-11-09 | 哈尔滨工业大学 | Method for detecting old and new as well as deformity of paper currency by sorting machine based on paper currency degeneration energy function |
-
2012
- 2012-12-20 CN CN201210558063.XA patent/CN103048331B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102236897A (en) * | 2011-05-09 | 2011-11-09 | 哈尔滨工业大学 | Method for detecting old and new as well as deformity of paper currency by sorting machine based on paper currency degeneration energy function |
Non-Patent Citations (2)
| Title |
|---|
| A Generic Framework for Non-rigid Registration Based on Non-uniform Multi-level Free-Form Deformations;Julia A. Schnabel 等;《MICCAI 2001, LNCS 2208》;20011231;573－581 * |
| 自动检测精细印刷品缺陷的初步方案;章毓晋 等;《中国体视学与图像分析》;20010630;第6卷(第2期);109－116 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103048331A (en) | 2013-04-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103048331B (en) | Printing defect detection method based on flexible template registration | |
| CN111640157B (en) | Checkerboard corner detection method based on neural network and application thereof | |
| CN111784778B (en) | Binocular camera external parameter calibration method and system based on linear solving and nonlinear optimization | |
| CN105957015B (en) | A kind of 360 degree of panorama mosaic methods of threaded barrel inner wall image and system | |
| CN111429533B (en) | Camera lens distortion parameter estimation device and method | |
| CN103345755B (en) | A kind of Chessboard angular point sub-pixel extraction based on Harris operator | |
| US8306323B2 (en) | Method and apparatus for correcting depth image | |
| CN108596867A (en) | A kind of picture bearing calibration and system based on ORB algorithms | |
| CN102800096B (en) | Robustness estimation algorithm of camera parameter | |
| CN102521816A (en) | Real-time wide-scene monitoring synthesis method for cloud data center room | |
| CN110009690A (en) | Binocular stereo vision image measuring method based on polar curve correction | |
| CN105160652A (en) | Handset casing testing apparatus and method based on computer vision | |
| US11315229B2 (en) | Method for training defect detector | |
| Yang et al. | An image-based intelligent system for pointer instrument reading | |
| CN110223355B (en) | Feature mark point matching method based on dual epipolar constraint | |
| CN112348869B (en) | Method for recovering monocular SLAM scale through detection and calibration | |
| CN105989588A (en) | Irregular-shaped material cutting image correction method and system | |
| CN115546716A (en) | Binocular vision-based method for positioning fire source around power transmission line | |
| CN117557565B (en) | Detection method and device for lithium battery pole piece | |
| CN114612418A (en) | Method, device and system for detecting surface defects of mouse shell and electronic equipment | |
| CN117671299A (en) | Loop detection method, device, equipment and storage medium | |
| CN117253062A (en) | Relay contact image characteristic quick matching method under any gesture | |
| CN108898585B (en) | Shaft part detection method and device | |
| CN112562008B (en) | Target point matching method in local binocular vision measurement | |
| CN114897990A (en) | Camera distortion calibration method and system based on neural network and storage medium |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Li Hanzhong Inventor after: Wen Xingche Inventor after: Wang Xi Inventor after: Fan Fuyu Inventor after: Cao Baohua Inventor after: Meng Wenxi Inventor after: Qu Shaohua Inventor after: Sun Xiaogang Inventor after: Wei Jun Inventor after: Zhang Pengyi Inventor after: Jia Ke Inventor after: Zhang Liang Inventor after: Zhang Jian Inventor before: Li Hanzhong Inventor before: Wen Xinche Inventor before: Wang Xi Inventor before: Fan Fuyu Inventor before: Cao Baohua Inventor before: Meng Wenxi Inventor before: Qu Shaohua Inventor before: Sun Xiaogang Inventor before: Wei Jun Inventor before: Zhang Pengji Inventor before: Jia Ke Inventor before: Zhang Liang Inventor before: Zhang Jian |
|
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 050031 No.37 Zhaiying South Street, Shijiazhuang City, Hebei Province Patentee after: SHIJIAZHUANG BANKNOTE PRINTING Co.,Ltd. Patentee after: China Banknote Printing and Minting Group Co., Ltd Address before: 050031 No.37 Zhaiying South Street, Shijiazhuang City, Hebei Province Patentee before: SHIJIAZHUANG BANKNOTE PRINTING Co.,Ltd. Patentee before: China Banknote Printing and mint General Company |