CN110246114A - Solve the problems, such as the photometric stereo vision data driving global optimization approach of BRDF nonlinearity - Google Patents
Solve the problems, such as the photometric stereo vision data driving global optimization approach of BRDF nonlinearity Download PDFInfo
- Publication number
- CN110246114A CN110246114A CN201910239574.7A CN201910239574A CN110246114A CN 110246114 A CN110246114 A CN 110246114A CN 201910239574 A CN201910239574 A CN 201910239574A CN 110246114 A CN110246114 A CN 110246114A
- Authority
- CN
- China
- Prior art keywords
- brdf
- reflex
- nonlinearity
- mapping relations
- photometric stereo
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005457 optimization Methods 0.000 title claims abstract description 9
- 238000013459 approach Methods 0.000 title claims abstract 3
- 238000000034 method Methods 0.000 claims abstract description 25
- 238000013507 mapping Methods 0.000 claims abstract description 13
- 230000011514 reflex Effects 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000013178 mathematical model Methods 0.000 claims abstract description 6
- 238000005070 sampling Methods 0.000 claims abstract description 4
- 238000013461 design Methods 0.000 claims abstract description 3
- 244000062793 Sorghum vulgare Species 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 235000019713 millet Nutrition 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 4
- 230000005855 radiation Effects 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- XCWPUUGSGHNIDZ-UHFFFAOYSA-N Oxypertine Chemical compound C1=2C=C(OC)C(OC)=CC=2NC(C)=C1CCN(CC1)CCN1C1=CC=CC=C1 XCWPUUGSGHNIDZ-UHFFFAOYSA-N 0.000 claims description 2
- 230000003287 optical effect Effects 0.000 claims description 2
- 230000004044 response Effects 0.000 claims description 2
- 239000000463 material Substances 0.000 abstract description 18
- 238000002474 experimental method Methods 0.000 abstract description 8
- 230000007547 defect Effects 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 abstract description 2
- 239000013598 vector Substances 0.000 description 15
- 238000009877 rendering Methods 0.000 description 6
- 241000283973 Oryctolagus cuniculus Species 0.000 description 3
- 229910000906 Bronze Inorganic materials 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 2
- 239000010974 bronze Substances 0.000 description 2
- 238000005315 distribution function Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 229910000838 Al alloy Inorganic materials 0.000 description 1
- VYZAMTAEIAYCRO-UHFFFAOYSA-N Chromium Chemical compound [Cr] VYZAMTAEIAYCRO-UHFFFAOYSA-N 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052804 chromium Inorganic materials 0.000 description 1
- 239000011651 chromium Substances 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 150000002730 mercury Chemical class 0.000 description 1
- 230000032696 parturition Effects 0.000 description 1
- 238000007747 plating Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
A kind of photometric stereo vision data driving global optimization approach solving the problems, such as BRDF nonlinearity in Machine Vision Recognition Technology field, propose a kind of for describing the mapping relations of non-lambertian reflex, which realizes the decoupling of reflex and surface normal while describing reflex.This patent is directed to proposed mapping relations, establishes corresponding continuous mathematical model using Gaussian process, and design training precision and predetermined speed that sampling policy guarantees model.It can better solve BRDF nonlinearity problem, and possess higher computational efficiency.Emulation experiment and true experiment based on MERL database all demonstrate the superiority of this method.This method has preferable application prospect in terms of the defects detection on high-volume homogenous material surface.
Description
Technical field
The present invention relates to a kind of technology in Machine Vision Recognition field, specifically a kind of solution BRDF nonlinearity
The photometric stereo vision data of problem drives global optimization (Data-driven Photometric Stereo, DPS) algorithm.
Background technique
One of photometric stereo vision technique difficult point is fast and effeciently to solve the problems, such as the complex reflex of Non Lambert reflector material.
Traditional photometric stereo vision is assumed to estimate body surface normal vector based on lambert's body.Under the assumptions, pixel record value and object
There are linear relationships between body surface normal.Therefore, for some pixel, it is known that the pixel of three non-coplanar direction of illuminations
Record value, so that it may seek the normal vector that the pixel corresponds to object table millet cake.But in real world, meet lambert's body hypothesis
Material is seldom, and which limits the application spaces of photometric stereo vision.In order to solve this problem, it three has developed during the last ten years perhaps
More algorithms, wherein the solution of mainstream is segmented into two kinds.
First method thinks that most pixel record values meet lambert's body it is assumed that by highlight, cast shadow (cast
Shadow), Attached shadows (attach shadow) etc. are used as Error processing.However for some materials, even eliminating height
The influence of light and shade, the relationship between pixel record value and corresponding normal vector are also unsatisfactory for lambert's body hypothesis.Second method
Using more complicated bidirectional reflectance distribution function (Bidirectional Reflectance Distribution
Function, BRDF) model, or improve normal vector using certain properties of BRDF and restore precision.But above-mentioned model is suitable
Answering property is still limited, and normal vector restores the space that precision is still improved.In addition, the model proposed is often relative complex, or
Person's optimization iterative solution process is cumbersome, and calculating speed is slow.
Summary of the invention
The invention proposes a kind of photometric stereo vision datas for solving the problems, such as BRDF nonlinearity to drive global optimization
(Data-driven Photometric Stereo, DPS) algorithm, it can better solve BRDF nonlinearity problem,
And possess higher computational efficiency.
The technical scheme adopted by the invention is that: this method instructing referring to data to certain material by Gaussian process
Practice, a reasonable global presumption model is obtained, to the other shapes object method with higher under same material same scene
Vector restores precision, comprising the following steps:
First, the mapping relations of non-lambertian reflex are described with following formula, the mapping relations are existing in description reflection
As while realize the decoupling of reflex and surface normal:
Ip=fBRDF(l,n,v)kcIsTnTl
Wherein, kpIt is the linear response coefficient of camera internal, d is diaphragm diameter, and f is camera focus, and α is object table millet cake
With the angle between corresponding pixel points line and optical axis, under hypothesis of the shooting distance much larger than dimension of object, α ≈ 0, F are objects
The ratio of radiation intensity and body surface irradiation intensity that body surface millet cake is reflected to camera direction, IsThe intensity of light source, l be into
Radiation direction is penetrated, T is the time for exposure；
Second, for mentioned mapping relations, establish corresponding continuous mathematical model using Gaussian process, by with it is known
Data relevant Joint Distributions is predictedNumerical value:
WhereinIt is the rank input quantity of n × 4, L=(lp1,lp2,…lpn)T,lp·ForIt is right
The incident ray direction answered, μ (X) are equal functional equations, and K (X, X) is n × n rank covariance matrix, express the correlation between X
Property, n is training points quantity,For the prediction result of BRDF, I is unit matrix,Indicate noise；
Third proposes a kind of data-driven photometric stereo visual global optimization algorithm based on Gaussian process, design sampling
Strategy guarantees the training precision and predetermined speed of model:
WhereinAnd pass throughIt is normalized
Processing.
Compared with prior art, the invention has the following beneficial effects:
First, this patent proposes a kind of for describing the mapping relations of non-lambertian reflex, which is retouching
The decoupling of reflex and surface normal is realized while stating reflex；
Second, this patent is directed to proposed mapping relations, establishes corresponding continuous mathematical model using Gaussian process, and set
Count training precision and predetermined speed that sampling policy guarantees model；
Third, this patent propose that a kind of photometric stereo vision data driving for solving the problems, such as BRDF nonlinearity is global excellent
Change (Data-driven Photometric Stereo, DPS) algorithm；
4th, it is based on the emulation of MERL (Mitsubishi Electric Research Laboratories) database
Experiment and true experiment all demonstrate the superiority of this method.Defects detection side of this method on high-volume homogenous material surface
There is preferable application prospect in face.
Detailed description of the invention
Fig. 1 is inventive algorithm main-process stream；
Fig. 2 is that ball of the embodiment of the present invention and the rendering of rabbit some materials are schemed；
Fig. 3 is that five kinds of material normal vectors of the embodiment of the present invention restore figure (from top to bottom: alum-bronze, color-
Changing-paint2, ipswich-pine-221, pvc and ss440)；
Fig. 4 be five kinds of material error figures of the embodiment of the present invention (from top to bottom: alum-bronze, color-changing-
Paint2, ipswich-pine-221, pvc and ss440)；
Fig. 5 standard law of embodiment of the present invention vectogram and practical photograph；
Fig. 6 is that normal vector of the embodiment of the present invention restores figure and Error Graph (from top to bottom: LS, CBR, DPS).
Specific embodiment
The preferred embodiments of the present invention will be described in detail with reference to the accompanying drawing, so that advantages and features of the invention energy
It is easier to be readily appreciated by one skilled in the art, so as to make a clearer definition of the protection scope of the present invention.
In the description of the present invention it is to be appreciated that the orientation or positional relationship of term " on " "lower" " left side " " right side " instruction
For orientation based on the figure and positional relationship, it is only for facilitate and describes structurally and operationally mode of the invention, without
It is instruction or implies that signified part must have a particular orientation, with the operation of specific orientation, thus should not be understood as pair
Limitation of the invention.
Embodiment
The following further describes the present invention with reference to the drawings:
In general, the BRDF of Non Lambert reflector material is nonlinearity, and the BRDF of different materials is different.Therefore,
Remove to indicate the BRDF of the material with the dense measurement data of certain material, it is more suitable than analytic uniform model.Of the invention is basic
Thought is exactly to obtain reasonable mathematical model according to the measured database of certain material, and then according to the data model to corresponding
Data in material database are inferred.Specifically, the present invention establishes incident direction l, observed direction v, pixel record
Value IpMapping model between BRDF, and inferred from input data is carried out according to the mapping model.
In order to realize that above-mentioned thought, the present invention utilize Gaussian process founding mathematical models.Gaussian process is a kind of Bayes
Inference pattern is a kind of powerful mathematical tool for handling nonlinearity problem.The vacation of dimension of object is much larger than in shooting distance
It sets, observed direction v may be considered constant vector (0,0,1)T, therefore in derivation process, v is not as input quantity.Such Gauss
Process just establishes incident ray direction l, pixel record valueWithBetween mapping relations, as shown in formula (1).In order to protect
Demonstrate,prove the variation of pixel record value under same pixel difference direction of illumination only as BRDF and light incident direction it is different caused by, because
This needs to eliminate calibration incident ray intensity kcIsT is to pixel record value IpInfluence.The model has just used elimination calibration incident
The pixel record value that light intensity influencesNormal vector solves below for convenience, which has used falling for BRDF
NumberIt is denoted as
GP (X)~N (μ (X), K (X, X)) (1)
WhereinIt is the rank input quantity of n × 4, L=(lp1,lp2,…lpn)T,lp·ForCorresponding incident ray direction, μ (X) are equal functional equations, and K (X, X) is n × n rank covariance matrix, are expressed between X
Correlation, wherein n is training points quantity.
When establishing model, due to not knowing any prior information, so functional equation is considered zero-mean function.For
Any desired prediction Numerical value can be predicted to obtain by Joint Distribution relevant to given data,
WhereinFor BRDF prediction result, shown in expression such as formula (2).
In order to prove the validity of this method and the scalability of training pattern, ball and two kinds of shapes of rabbit are generated respectively herein
The rendering picture of shape.Because just CBR algorithm needs 300 pictures that can obtain relatively good experimental result, in order to preferably
With CBR algorithm comparison, 300 pictures have been rendered herein.Wherein the resolution ratio of ball rendering picture is that rabbit renders the resolution of picture
Rate is 256 × 256.Fig. 2 illustrates the rendering picture of 5 kinds of materials, in order to clearly show rendering picture, the rendering of every kind of material
Data have carried out increasing with multiple.
Present invention employs with most of normal estimation figures, that is, allow three channels r, g, b of color image to respectively represent
Three components of unit normal vector, value range are [- 1,1].In order to facilitate display, these three components are normalized to
[0,1] section, i.e.,Wherein IshowIt is to show normal estimation figure, ItrueIt is true normal estimation figure.In order to
It clearly indicates the evaluated error of normal vector, generates corresponding Error Graph, i.e. Fig. 3 in experiment, which is estimation normal direction
The distribution map of amount and true value normal vector angle.Color is more blue in figure, and angle is smaller, and error is smaller, and the color the partially red, and angle is got over
Greatly, error is bigger.
The present invention has built a photometric stereo visual experiment platform.The platform guarantees dark room conditions, and with 46 20W
Point light source irradiation shooting distance be 1.8 meters of object, shooting object is the aluminium alloy paraboloid of 100mm*100mm.Camera-type
Number be Daheng's camera Mercury series MER-502-79U3C industrial camera, photo resolution 2448*2048.Present invention assumes that should
Paraboloidal mismachining tolerance very little, paraboloidal true value normal vector can be calculated by corresponding mathematics model.Automatically it is giving birth to
After mask artwork, standard law vectogram can be obtained by by interpolation, Fig. 4 illustrates standard law vectogram and wherein a Zhang Zhenshi
Photo.
It is put into chromium plating steel ball Calibrating source incident direction in the scene, is subsequently placed into standard white plate, carries out intensity of light source mark
It is fixed.What it is due to this experiment is color camera, so being the intensity of light source value in tri- channels r, g, b obtained in experiment.Prepare
When training data, the pixel record value in each channel eliminates the influence of the intensity of light source divided by the intensity of light source value in the channel, then
Median filtering is carried out to each channel, color image is then switched into black and white picture.When selecting training points, 10 are first uniformly adopted
Then pixel uniformly selects 5 non-zero pixels record values to each pixel.It according to above-mentioned data training and predicts, obtains method
Vector estimation figure and Error Graph are as shown in Figure 5.It can be seen from the figure that the error ratio LS and CBR of DPS algorithm is small.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (1)
1. a kind of photometric stereo vision data driving global optimization approach for solving the problems, such as BRDF nonlinearity, it is characterised in that
The following steps are included:
First, the mapping relations of non-lambertian reflex are described with following formula, the mapping relations are in description reflex
The decoupling of reflex and surface normal is realized simultaneously:
Ip=fBRDF(l,n,v)kcIsTnTl
Wherein, kpThe linear response coefficient of camera internal, d is diaphragm diameter, and f is camera focus, α be object table millet cake with it is right
The angle between pixel line and optical axis is answered, under hypothesis of the shooting distance much larger than dimension of object, α ≈ 0, F are object tables
The ratio of radiation intensity and body surface irradiation intensity that millet cake is reflected to camera direction, IsIt is the intensity of light source, l is incident ray
Direction, T are the time for exposure；
Second, for mentioned mapping relations, establish corresponding continuous mathematical model using Gaussian process, by with given data
Relevant Joint Distribution is predictedNumerical value:
WhereinIt is the rank input quantity of n × 4, L=(lp1,lp2,…lpn)T,lp·ForIt is corresponding
Incident ray direction, μ (X) are equal functional equations, and K (X, X) is n × n rank covariance matrix, express the correlation between X, n
It is training points quantity,For the prediction result of BRDF, I is unit matrix,Indicate noise；
Third proposes a kind of data-driven photometric stereo visual global optimization algorithm based on Gaussian process, designs sampling policy
Guarantee the training precision and predetermined speed of model:
WhereinAnd pass throughIt is normalized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910239574.7A CN110246114A (en) | 2019-03-27 | 2019-03-27 | Solve the problems, such as the photometric stereo vision data driving global optimization approach of BRDF nonlinearity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910239574.7A CN110246114A (en) | 2019-03-27 | 2019-03-27 | Solve the problems, such as the photometric stereo vision data driving global optimization approach of BRDF nonlinearity |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110246114A true CN110246114A (en) | 2019-09-17 |
Family
ID=67883055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910239574.7A Pending CN110246114A (en) | 2019-03-27 | 2019-03-27 | Solve the problems, such as the photometric stereo vision data driving global optimization approach of BRDF nonlinearity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110246114A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112925351A (en) * | 2019-12-06 | 2021-06-08 | 杭州萤石软件有限公司 | Method and device for controlling light source of vision machine |
| CN113870342A (en) * | 2021-08-19 | 2021-12-31 | 广州超音速自动化科技股份有限公司 | Appearance defect detection method, intelligent terminal and storage device |
-
2019
- 2019-03-27 CN CN201910239574.7A patent/CN110246114A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112925351A (en) * | 2019-12-06 | 2021-06-08 | 杭州萤石软件有限公司 | Method and device for controlling light source of vision machine |
| CN112925351B (en) * | 2019-12-06 | 2022-08-02 | 杭州萤石软件有限公司 | Method and device for controlling light source of vision machine |
| CN113870342A (en) * | 2021-08-19 | 2021-12-31 | 广州超音速自动化科技股份有限公司 | Appearance defect detection method, intelligent terminal and storage device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109644224B (en) | System and method for capturing digital images | |
| US7253832B2 (en) | Shape extraction system and 3-D (three dimension) information acquisition system using the same | |
| US9639947B2 (en) | Method and optical system for determining a depth map of an image | |
| Raskar et al. | Glare aware photography: 4D ray sampling for reducing glare effects of camera lenses | |
| CN109792478A (en) | System and method based on focus target information adjustment focus | |
| KR20170005009A (en) | Generation and use of a 3d radon image | |
| WO2019105261A1 (en) | Background blurring method and apparatus, and device | |
| CN102984448A (en) | Method of controlling an action, such as a sharpness modification, using a colour digital image | |
| CN112634156A (en) | Method for estimating material reflection parameter based on portable equipment collected image | |
| WO2019019904A1 (en) | White balance processing method and apparatus, and terminal | |
| CN105163047A (en) | HDR (High Dynamic Range) image generation method and system based on color space conversion and shooting terminal | |
| JP7378219B2 (en) | Imaging device, image processing device, control method, and program | |
| CN110246114A (en) | Solve the problems, such as the photometric stereo vision data driving global optimization approach of BRDF nonlinearity | |
| CN104952048B (en) | A kind of focus storehouse picture synthesis method based on as volume reconstruction | |
| CN108427961A (en) | Synthetic aperture focusing imaging depth appraisal procedure based on convolutional neural networks | |
| CN117061868A (en) | Automatic photographing device based on image recognition | |
| CN106375675B (en) | A kind of more exposure image fusion methods of aerial camera | |
| Li et al. | Three-dimensional measurement for specular reflection surface based on deep learning and phase measuring profilometry | |
| CN110290313B (en) | Method for guiding automatic focusing equipment to be out of focus | |
| Hasinoff | Variable-aperture photography | |
| JP5610137B2 (en) | Photo measurement target and photo measurement method | |
| TW202240273A (en) | Infrared light-guided portrait relighting | |
| CN114529458A (en) | Method and device for realizing large-range continuous optical zooming based on fixed-focus lens | |
| TWI468772B (en) | Device and method for taking photographs | |
| Amano | Manipulation of material perception with light-field projection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190917 |
|
| WD01 | Invention patent application deemed withdrawn after publication |