CN106447656A - Rendering flawed image detection method based on image recognition - Google Patents
Rendering flawed image detection method based on image recognition Download PDFInfo
- Publication number
- CN106447656A CN106447656A CN201610842103.1A CN201610842103A CN106447656A CN 106447656 A CN106447656 A CN 106447656A CN 201610842103 A CN201610842103 A CN 201610842103A CN 106447656 A CN106447656 A CN 106447656A
- Authority
- CN
- China
- Prior art keywords
- matrix
- pixel
- image
- template
- neighborhood
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/005—General purpose rendering architectures
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a rendering flawed image detection method based on image recognition. The method comprises the following steps of inputting a rendering flawed image to be detected; calculating a differential matrix of the rendering image; determining the dimension of a detection template according to the rendering image change rate; calculating the neighbourhood minimum pixel difference of each pixel point in each frame of image according to the detection template; judging the located position of a broken frame image and flaws according to the minimum pixel differences. By aiming at the problems existing in the prior art, the invention aims at providing a video flaw detection method; flaw parts in the rendering result video image are fast detected.
Description
Technical field
The present invention designs one kind and renders image Defect Detection technology, especially design a kind of for rendering image Defect Detection
Method.
Background technology
Due to house ornamentation and video display render computing resource and storage resource demands are larger, completed using large-scale cloud computing technique
Such it is rendered to as mainstream technology.Thus provide the system rendering in the face of numerous different types of rendering tasks, load in resource
During there is certain probability so that part calculate node cannot ensure the whole normal load of all resources, thus can cause part
Rendering node renders not normal.When rendering result image file collects, those render scenes are screened by manual detection and is adjacent
The result images that image has differences, and define such image for abnormal two field picture, it is rendered again to revise mistake.Although
This abnormal image belongs to contingencies in rendering result, and typically isolated exist, and occurs continuous in the case of only a few
Two frame Rendering errors images occur, but this kind of mistake leads to whole render process heavy dependence manually to post-process misarrangement, greatly
Limit the automatic flow rendering and efficiency, for this reason, this patent designs a class algorithm automatic identification renders abnormal two field picture, with
Lifting renders process automation level, lifts rendering efficiency.
More than render its file size of abnormal image, form is had no with normal rendering result and is clearly distinguished from, using conventional
It is difficult to screen during information based on file format, design a kind of method based on machine vision for this present invention, using sequential chart
As the bad frame image in the local characteristicses automatic detection rendering result of change, its concrete content of the invention is as follows.
Content of the invention
The present invention is directed to the problem that prior art exists, and its object is to provide a kind of video flaw detection method, to wash with watercolours
Flaw part in dye result video image is used for quickly detecting.
For achieving the above object, the invention provides a kind of render flaw image detecting method based on image recognition, should
Method comprises the following steps：Input render video image to be detected；Calculate the difference matrix rendering image；According to rendering image
Rate of change determines detection template size；Each pixel neighborhood of a point minimum pixel in each two field picture is calculated according to detection template
Difference；Bad frame image and flaw present position are judged according to neighborhood minimum pixel difference.
As a further improvement on the present invention, methods described is further comprising the steps of：The rendering result image of input is entered
Row serializing is processed, and is read in RGB form and treats result images, and is arranged in chronological order.Obtain four-matrix data M
[X, Y, Z, T], wherein X, Y are video image location of pixels, and Z is RGB tri- color, and T is time shaft.
As a further improvement on the present invention, methods described is further comprising the steps of：Change according to rendering result image
Speed determines single pixel neighborhood detection template ST, because rendering the special nature of image, when rendering contexts are identical, rendering result
The numerical value relative constancy of each pixel in image.To the adjacent two field pictures sorting in temporal sequence, if prospect or background
On the premise of ohject displacement is relatively little, then necessarily can find completely corresponding with current pixel point another in closing on two field picture
One pixel, and this corresponding two pixel has neighborhood similarity in position.Therefore according to display foreground or background position
The speed moved can determine a neighborhood template so that each pixel can be with this neighborhood template as model in current frame image
Enclose, in a neighborhood closing on this pixel correspondence position of frame, find same pixel, neighborhood size is the template of ST
Size.In actual application, this template size is typically obtained by artificial experience.
As a further improvement on the present invention, methods described is further comprising the steps of：To seasonal effect in time series rendering result figure
As countershaft order carries out difference processing on time, difference processing is divided into two steps, i.e. matrix displacement and Difference Calculation：
M '=Mt-Mt+1T=[1...T-1]
Matrix displacement is to adjacent two field pictures matrix MtWith Mt+1Middle MtMatrix carries out shifting processing.First according to ST mould
The shape of plate, determines unit's point position in template, except each pixel in unit's point exterior sheathing determines a MtMovement (Δ x, Δ y)
Operation.MtTranslated in units of pixel according to the size of Δ x, Δ y, original matrix MtIn each pixel mobile after obtain new picture
Plain (x+ Δ x, y+ Δ y).If removing matrix M after pixel is mobilet+1Partly directly the giving up of scope；And if pixel occurs after moving
In matrix Mt+1In scope and original pixel is not in matrix MtIn scope, then need this new pixel to be carried out initialize assignment, assignment is big
Little for twice single pixel maximum pmax(pmax=max (m) m ∈ Mt).
Matrix M after displacement is calculated according to single pixel neighborhood detection template after shifting every timei tWith Mt+1Matrix difference matrix Δ Mi t
Once (i ∈ { ST-0 }), after all differences, matrix seeks vectorial 2 norms by RGB dimension, obtains differential mode matrix | Δ Mi t|.Finally to institute
Each corresponding pixel points in differential mode matrix are had to minimize matrix Δ Mmin t(ΔMmin t=mini{|ΔMi t|,i∈{ST-0}}).
As a further improvement on the present invention, methods described is further comprising the steps of：To matrix Δ Mmin tPost-processed,
Because there is identical pixel, matrix Δ M in most of pixel neighborhood in consecutive framemin tBig portion in theory
Subregion is 0.But due to there is the impact of noise, Δ M in imagemin tIn individually non-zero isolated point occurs, using Mathematical Morphology
Method can be to Δ Mmin tIn isolated abnormity point denoising.Concrete operations are the template determining Mathematical Morphology Method first
SE (using disc template, template size is | XY |/20000), then to Δ Mmin tCarry out closing operation of mathematical morphology, and result is entered
Row morphology opening operation, obtains correction matrix sequence Δ M 'min t.
As a further improvement on the present invention, methods described is further comprising the steps of：To differential mode correction matrix sequence Δ
M’min tDiscriminant analysis.Because based on similar principle in neighborhood of pixels in rendering result image, Δ M ' in matrix sequencemin tGreatly
Part is 0 matrix, therefore to matrix sequence Δ M 'min tEach of matrix carry out non-zero pixels statisticses, when wherein non-zero pixel count
Then can be determined that in the corresponding rendering result of current time t there is bad frame during surge, and check that this rendering result image is carried out point
Distinguish.
Beneficial effect：
Context of methods is significant in the industrial production, can achieve the bad frame automatic detection work(rendering image using this method
Can, the abnormal results during automatic detection renders under conditions of without manual intervention, and be marked.Render in extensive outdoor scene
In task, automatic bad frame detection method can assist examination bad frame to render image, improves the quality of rendering result, reduces identification bad frame
Cost of labor.
Brief description
Fig. 1 present invention renders image Defect Detection flow chart.
Fig. 2 present invention renders bad frame instance graph.
The difference matrix non-zero region figure of Fig. 3 present invention.
The difference matrix of Fig. 4 present invention calculates schematic diagram.
The bad frame detection curve figure of Fig. 5 present invention.
The bad frame region labeling figure of Fig. 6 present invention.
Specific embodiment
Describe the present invention below with reference to each embodiment shown in the drawings.
It show preferable system flow chart of the present invention refering to Fig. 1, be divided into input picture and carry out serializing process, determine list
Neighborhood of pixels template, sequence image matrix is shifted and Difference Calculation, difference modular matrix is carried out post-process, statistic mixed-state
Five steps such as bad frame image.
It is normal rendering result picture refering to left figure shown in Fig. 2, right figure is to render abnormal results picture.Because material loads
, it is evident that lower right corner meadow part renders exception in right figure, there is notable difference with left figure corresponding region in problem.This two frame
, there is large area non-zero region on correspondence position in image subtraction.In other words this adjacent two field pictures is on this position
Each pixel all can not find corresponding same pixel in template ST neighborhood.The difference result image of this two field pictures is as schemed
3, the wherein visible continuous non-zero region of flaw location, and other scatterplot are two field pictures prospect and the mobile caused difference of background
Dissimilarity.Because prospect and background translational speed are unhappy, so such discrepancy is isolated zonule, monomer area is little.
It show the difference matrix asking for t frame refering to Fig. 4.As it was noted above,It is in t two field picture MtOn the basis of
Translate the set of several subgraphs obtaining, the number of subgraph is relevant with the template size selected, template bigger needs translation
Number of times is more, and its subgraph is more.As shown in Fig. 4 (1),In MtOn the basis of horizontal direction be shifted x pixel of Δ, Vertical Square
To being shifted y pixel of Δ, (Δ x, Δ y) need to travel through the whole pixels except 0 point in template ST.One is had in translation motion
Subregion removes Mt, also can move in some region simultaneouslyFor immigrationThis subregion need initial to it
Change assignment, that is, image grey area, assignment size is twice single pixel maximum pmax(pmax=max (m) m ∈
Mt).
Several image collections obtaining after translationThrough with after an image Mt+1Phase
Subtract, obtain difference matrix
It show the numbering containing abnormal frame in this section of consecutive image asked for using the present invention refering to Fig. 5.As Fig. 3 institute
Show, in image, marked the scatterplot region that the continuous non-zero region being caused by bad frame and prospect, background movement cause.Using form
Method first carries out closed operation to image, then carries out morphology opening operation to result, you can protected with removing scatterplot region simultaneously
Stay the continuum that bad frame causes.The elemental area of the remaining regions in image after the computation of morphology operations can be drawn
Block diagram as Fig. 5.
It show using the present invention in the abnormal frame shown in Fig. 2 refering to Fig. 6, the accurate location of the unusual part asked for,
I.e. lower right corner black region in Fig. 6.Can be told also by remaining regions in previous step are carried out with interpretation or threshold value screening
The picture number of bad frame, can get bad frame as shown in Figure 6 by being marked to non-zero region continuous in this numbering image
Area image.
Certainly above-described embodiment only technology design to illustrate the invention and feature, its object is to allow and is familiar with technique
People will appreciate that present disclosure and implement according to this, can not be limited the scope of the invention with this.All according to this
The modification that the Spirit Essence of bright main technical schemes is done, all should be included within the scope of the present invention.
Claims (7)
1. flaw image detecting method rendered it is characterised in that the method comprises the following steps based on image recognition：
Step one, input render video image to be detected；
Step 2, calculating render the difference matrix of image；
Step 3, detection template size is determined according to rendering result image change speed；
Step 4, according to detection template calculate each two field picture in each pixel neighborhood of a point minimum pixel poor；
Step 5, according to neighborhood minimum pixel difference judge bad frame image and flaw present position.
2. according to claim 1 render flaw image detecting method it is characterised in that：According to rendering result in step 3
Image change speed determines single pixel neighborhood detection template ST, because rendering the special nature of image, when rendering contexts are identical,
The numerical value relative constancy of each pixel in rendering result image；To the adjacent two field pictures sorting in temporal sequence, if front
Scape or background object displacement relatively less on the premise of, then necessarily can find complete with current pixel point in closing on two field picture
Corresponding one other pixel point, and this corresponding two pixel has neighborhood similarity in position；Therefore according to display foreground
Or the speed of background displacement can determine a neighborhood template so that each pixel can be with this neighborhood in current frame image
Template is scope, finds same pixel, neighborhood size is in a neighborhood closing on this pixel correspondence position of frame
The template size of ST.
3. according to claim 2 render flaw image detecting method it is characterised in that：To seasonal effect in time series rendering result
Countershaft order carries out difference processing to image on time, and difference processing is divided into two steps, i.e. matrix displacement and Difference Calculation：
M '=Mt-Mt+1T=[1...T-1]
Matrix displacement is to adjacent two field pictures matrix MtWith Mt+1Middle MtMatrix carries out shifting processing：First according to ST template
Shape, determines unit's point position in template, except each pixel in unit's point exterior sheathing determines a MtMovement (Δ x, Δ y) grasp
Make；MtTranslated in units of pixel according to the size of Δ x, Δ y, original matrix MtIn each pixel mobile after obtain new pixel
(x+ Δ x, y+ Δ y)；If removing matrix M after pixel is mobilet+1Partly directly the giving up of scope；And if pixel occurs in after moving
Matrix Mt+1In scope and original pixel is not in matrix MtIn scope, then need this new pixel to be carried out initialize assignment, assignment size
For twice single pixel maximum pmax(pmax=max (m) m ∈ Mt)；
Matrix M after displacement is calculated according to single pixel neighborhood detection template after shifting every timei tWith Mt+1Matrix difference matrix Δ Mi tOnce (i
∈ { ST-0 }), after all differences, matrix seeks vectorial 2 norms by RGB dimension, obtains differential mode matrix | Δ Mi t|；Finally to all differential modes
In matrix, each corresponding pixel points is minimized matrix Δ Mmin t(ΔMmin t=mini{|ΔMi t|,i∈{ST-0}}).
4. according to claim 3 render flaw image detecting method it is characterised in that：To matrix Δ Mmin tLocate after carrying out
, because there is identical pixel, matrix Δ M in most of pixel neighborhood in consecutive frame in reasonmin tIn theory
Most of region is 0；But due to there is the impact of noise, Δ M in imagemin tIn individually non-zero isolated point occurs, using mathematics
Morphological method can be to Δ Mmin tIn isolated abnormity point denoising.
5. according to claim 4 render flaw image detecting method it is characterised in that described abnormity point denoising tool
Gymnastics conduct：Determine template SE of Mathematical Morphology Method first, then to Δ Mmin tCarry out closing operation of mathematical morphology, and by result
Carry out morphology opening operation, obtain differential mode correction matrix sequence Δ M 'min t.
6. according to claim 5 render flaw image detecting method it is characterised in that：Described template SE uses disk mould
Plate, template size is | XY |/20000.
7. according to claim 5 render flaw image detecting method it is characterised in that to differential mode correction matrix sequence
ΔM’min tDiscriminant analysis：Because based on similar principle in neighborhood of pixels in rendering result image, Δ M ' in matrix sequencemin t
Most of is 0 matrix, therefore to matrix sequence Δ M 'min tEach of matrix carry out non-zero pixels statisticses, when wherein non-zero pixel
Number then can be determined that in the corresponding rendering result of current time t there is bad frame when increasing sharply, and checks that this rendering result image is carried out
Differentiate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610842103.1A CN106447656B (en) | 2016-09-22 | 2016-09-22 | Rendering flaw image detecting method based on image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610842103.1A CN106447656B (en) | 2016-09-22 | 2016-09-22 | Rendering flaw image detecting method based on image recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106447656A true CN106447656A (en) | 2017-02-22 |
| CN106447656B CN106447656B (en) | 2019-02-15 |
Family
ID=58166855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610842103.1A Active CN106447656B (en) | 2016-09-22 | 2016-09-22 | Rendering flaw image detecting method based on image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106447656B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107197325A (en) * | 2017-06-13 | 2017-09-22 | 深圳市瑞云科技有限公司 | A kind of bad frame automatic testing method |
| CN111160160A (en) * | 2019-12-18 | 2020-05-15 | 河海大学 | Neighborhood box-based block mass acquisition computing method |
| CN111300983A (en) * | 2018-11-26 | 2020-06-19 | 海德堡印刷机械股份公司 | Fast image correction for image inspection |
| CN115661145A (en) * | 2022-12-23 | 2023-01-31 | 海马云（天津）信息技术有限公司 | Cloud application bad frame detection method and device, electronic equipment and storage medium |
| CN115941914A (en) * | 2023-01-06 | 2023-04-07 | 湖南马栏山视频先进技术研究院有限公司 | Video rendering system based on video frame analysis |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101604458A (en) * | 2008-06-11 | 2009-12-16 | 美国西门子医疗解决公司 | The method that is used for the computer aided diagnosis results of display of pre-rendered |
| CN102903120A (en) * | 2012-07-19 | 2013-01-30 | 中国人民解放军国防科学技术大学 | Time-space condition information based moving object detection method |
| CN103945228A (en) * | 2014-03-28 | 2014-07-23 | 上海交通大学 | Video intra-frame copy-move tampering detection method based on space and time relevance |
| CN104539942A (en) * | 2014-12-26 | 2015-04-22 | 赞奇科技发展有限公司 | Video shot switching detection method and device based on frame difference cluster |
-
2016
- 2016-09-22 CN CN201610842103.1A patent/CN106447656B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101604458A (en) * | 2008-06-11 | 2009-12-16 | 美国西门子医疗解决公司 | The method that is used for the computer aided diagnosis results of display of pre-rendered |
| CN102903120A (en) * | 2012-07-19 | 2013-01-30 | 中国人民解放军国防科学技术大学 | Time-space condition information based moving object detection method |
| CN103945228A (en) * | 2014-03-28 | 2014-07-23 | 上海交通大学 | Video intra-frame copy-move tampering detection method based on space and time relevance |
| CN104539942A (en) * | 2014-12-26 | 2015-04-22 | 赞奇科技发展有限公司 | Video shot switching detection method and device based on frame difference cluster |
Non-Patent Citations (2)
| Title |
|---|
| RONG-CHI CHANG等: "Photo Defect Detection for Image Inpainting", 《PROCEEDINGS OF THE SEVENTH IEEE INTERNATIONAL SYMPOSIUM ON MULTIMEDIA(ISM"05)》 * |
| 刘谦雷 等: "用于视频镜头突变切换检测的二次差分法和像素点匹配二次差分法", 《中国图象图形学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107197325A (en) * | 2017-06-13 | 2017-09-22 | 深圳市瑞云科技有限公司 | A kind of bad frame automatic testing method |
| CN111300983A (en) * | 2018-11-26 | 2020-06-19 | 海德堡印刷机械股份公司 | Fast image correction for image inspection |
| CN111160160A (en) * | 2019-12-18 | 2020-05-15 | 河海大学 | Neighborhood box-based block mass acquisition computing method |
| CN111160160B (en) * | 2019-12-18 | 2022-08-05 | 河海大学 | Neighborhood box-based block mass acquisition computing method |
| CN115661145A (en) * | 2022-12-23 | 2023-01-31 | 海马云（天津）信息技术有限公司 | Cloud application bad frame detection method and device, electronic equipment and storage medium |
| CN115941914A (en) * | 2023-01-06 | 2023-04-07 | 湖南马栏山视频先进技术研究院有限公司 | Video rendering system based on video frame analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106447656B (en) | 2019-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106447656A (en) | Rendering flawed image detection method based on image recognition | |
| CN101453575B (en) | Video subtitle information extracting method | |
| EP1683105B1 (en) | Object detection in images | |
| CN102830958B (en) | A kind of method and system for obtaining interface control information | |
| CN111126115B (en) | Violent sorting behavior identification method and device | |
| CN107784301A (en) | Method and apparatus for identifying character area in image | |
| CN110766017B (en) | Mobile terminal text recognition method and system based on deep learning | |
| CN113283395B (en) | Video detection method for blocking foreign matters at transfer position of coal conveying belt | |
| CN111597941B (en) | Target detection method for dam defect image | |
| CN111414938B (en) | Target detection method for bubbles in plate heat exchanger | |
| CN110807763A (en) | Method and system for detecting ceramic tile surface bulge | |
| JP5006479B1 (en) | Moving image region determination apparatus or method thereof | |
| CN110458809B (en) | Yarn evenness detection method based on sub-pixel edge detection | |
| CN114419006A (en) | Method and system for removing watermark of gray level video characters changing along with background | |
| CN114862642A (en) | Method for removing short video visible watermark and computer readable storage medium | |
| CN117115171B (en) | Slight bright point defect detection method applied to subway LCD display screen | |
| CN116823775A (en) | Display screen defect detection method based on deep learning | |
| CN111429437A (en) | Image non-reference definition quality detection method for target detection | |
| CN105913440A (en) | Bimodal discrimination based fabric material surface defect segmenting method | |
| CN116229355A (en) | Image detection method and device, electronic equipment and storage medium | |
| CN110674778B (en) | High-resolution video image target detection method and device | |
| CN114067097A (en) | Image blocking target detection method, system and medium based on deep learning | |
| CN111028245A (en) | Multi-mode composite high-definition high-speed video background modeling method | |
| CN113066075B (en) | Multi-image fusion denim flaw detection method and device | |
| CN113887430B (en) | Method and system for locating polling video text |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20191113 Address after: 200030 room 249, zone a, 2f, no.420 Fenglin Road, Xuhui District, Shanghai Patentee after: Shanghai Zanqi Culture Technology Co., Ltd Address before: 23, building 9-2, 213022 East Taihu Road, Jiangsu, Changzhou Patentee before: JIANGSU CUDATEC CO., LTD. |