CN110163817A - A kind of phase main value extracting method based on full convolutional neural networks - Google Patents
A kind of phase main value extracting method based on full convolutional neural networks Download PDFInfo
- Publication number
- CN110163817A CN110163817A CN201910347403.6A CN201910347403A CN110163817A CN 110163817 A CN110163817 A CN 110163817A CN 201910347403 A CN201910347403 A CN 201910347403A CN 110163817 A CN110163817 A CN 110163817A
- Authority
- CN
- China
- Prior art keywords
- layer
- phase
- main value
- layers
- phase main
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 7
- 230000004913 activation Effects 0.000 claims description 19
- 238000002310 reflectometry Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 6
- 230000003287 optical effect Effects 0.000 claims description 5
- 238000013461 design Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 description 8
- 230000010363 phase shift Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000001314 profilometry Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
- G01B11/25—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures by projecting a pattern, e.g. one or more lines, moiré fringes on the object
- G01B11/254—Projection of a pattern, viewing through a pattern, e.g. moiré
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/97—Determining parameters from multiple pictures
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Length Measuring Devices By Optical Means (AREA)
- Image Analysis (AREA)
Abstract
A kind of phase main value extracting method based on full convolutional neural networks, the following steps are included: 1) on computers by the bar graph by Sine distribution needed for preparatory coding, the bar graph encoded in advance is projected to determinand, and acquires determinand bar graph using industrial camera;2) full convolutional neural networks model is constructed, training parameter and loss function are set, 1) picture obtained in is input in neural network, neural network is run, obtains required phase main value;3) solution is carried out to phase main value obtained in 2) using the method being oriented to based on Quality Map to twine, obtain accurate phase value.The quantity that the present invention provides a kind of Image Acquisition is few, without the higher phase main value extracting method based on full convolutional neural networks of training dataset and training process, precision.
Description
Technical field
The present invention relates to a kind of image processing method, especially a kind of phase main value based on full convolutional neural networks is extracted
Method.
Background technique
With the diversification of rapid development and the social demand of information technology, the space three-dimensional of contour of object is measured in industry
The various fields such as automatic detection, control of product quality, reverse engineer, biomedicine, virtual reality, the reproduction of the cultural relics, anthropological measuring
In be widely applied.Especially with the three-dimensional measurement of optical means, due to being non-cpntact measurement, measurement accuracy height, obtaining
It takes data volume big, light, mechanical, electrical integration is easily realized under computer and is greatly developed in last decade.
Phase measuring profilometer (phase measuring profilometry, abbreviation PMP) is thrown using sinusoidal grating
A kind of method for three-dimensional measurement that shadow and phase-shifting technique combine.Its basic thought is: when a sinusoidal grating graphic projection to three
When tieing up diffusing reflection body surface, the deforming stripe modulated by body surface face shape can be obtained from imaging system, utilizes discrete phase
Shifting technology obtains N amplitude variation shape light field image, calculates phase distribution further according to N step phase shift algorithm.The phase main value of mainstream at present
Extraction algorithm includes three step phase shift methods, four-stepped switching policy, these algorithms need at least three width images to extract phase main value, and are
Reduce noise, background influence and guarantee that solution twines precision, often need to shoot different stripeds frequencies in actual measurement respectively
The image of rate just needs to shoot 16 width pictures to guarantee precision, such as using the four-stepped switching policy progress one-shot measurement of 4 kinds of frequencies,
Excessive picture shooting quantity significantly reduces the speed of three-dimensional reconstruction.And the routine use condition poor in experiment condition
Under, the case where mobile situation of measured object occurs, measurement accuracy is caused to decline often is had in the shooting process of more than ten pictures.
Summary of the invention
When in order to overcome the excessive required picture number in actual measurement of conventional images phase main value extracting method, shooting
Between too long, the lower deficiency of precision, in order to extract accurate phase main value in single frames stripe pattern, based on such
Thinking, the present invention propose a kind of higher phase main value extracting method based on full convolutional neural networks of precision.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of phase main value extracting method based on full convolutional neural networks, comprising the following steps:
1) on computers by the bar graph by Sine distribution needed for preparatory coding, the bar graph encoded in advance is projected
Determinand bar graph is acquired to determinand, and using industrial camera;
2) full convolutional neural networks model is constructed, training parameter and loss function are set, 1) picture obtained in is inputted
Into neural network, neural network is run, required phase main value is obtained;
3) solution is carried out to phase main value obtained in 2) using the method being oriented to based on Quality Map to twine, obtain accurate phase
Value.
Further, in the step 1), acquisition bar graph process the following steps are included:
1.1) striped precoding is carried out on computers, constructs the required bar graph being distributed by sinusoidal rule, striped
It is encoded according to following formula:
Wherein I (x, y) is gray value of image, and x is abscissa;
1.2) design shooting optical path, and industrial camera, DLP projector and determinand are placed by designed optical path;
1.3) stripe pattern of high definition is successively acquired using industrial camera.
Further, in the step 2), extract phase main value process the following steps are included:
2.1) neural network constructs
When projector projects the bar graph encoded in advance to determinand surface, due to by body surface height
Modulation, the deforming stripe obtained by CCD camera indicate are as follows:
Wherein a (x, y) and b (x, y) reflects the variation of bias light and surface reflectivity respectively; It is to be calculated
Relative phase values, also referred to as phase main value, it reflects elevation information in object corresponding points;
A full convolutional neural networks model is constructed, the layer of entire full convolutional neural networks includes ten parts, wherein the
One layer is coding layer to layer 5, and each layer is made of two convolutional layers and two BN layers, wherein the convolution kernel of the first convolutional layer
Size is 3 × 3, and quantity is 128, stride 2, and the convolution kernel size of the second convolutional layer is 3 × 3, and quantity is 128, stride
Be 1, the image of input first passes around the first convolutional layer, BN layers and Leaky ReLU activation primitive, using the second convolutional layer,
BN layers and Leaky ReLU activation primitive processing;
Layer 6 is decoding layer to the tenth layer, and each layer is by two convolutional layers, three BN layers and a up-sampling layer group
At, the convolution kernel size of two of them convolutional layer is 3 × 3, and quantity is 128, and stride 1, input feature vector successively passes through BN layers,
First convolutional layer, BN layers and Leaky ReLU activation primitive;Successively pass through the second convolutional layer, BN layers and Leaky ReLU again
Activation primitive processing up-samples feature by bilinear interpolation or closest method finally by up-sampling layer;
In addition, first layer and the tenth layer, the second layer and the 9th layer, third layer and the 8th layer, the 4th layer and layer 7, the 5th
Skip connections is also added between layer and layer 6, skip connections is by convolutional layer, BN layers and Leaky
ReLU activation primitive is constituted, and wherein the convolution kernel size of convolutional layer is 1 × 1, and convolution nuclear volume is 4;
The image that size is 512 × 512 × 1 is inputted into neural network, successively by first layer to layer 6, the spy of output
Levying size is 16 × 16 × 128, then successively by layer 6 to the tenth layer, the feature sizes of output are 512 × 512 × 128;
The convolution kernel and RELU activation primitive that the last one convolutional layer is 1 × 1 by 1 size are handled, and output feature is big
Small is 512 × 512 × 1;
2.2) neural network is run
The input of neural network is the three width sizes Uniform noise image being randomly generated identical with 1) middle shooting image, point
It is not set to be fitted to background a (x, y), surface reflectivity b (x, y) and phase main value φ (x, y), wherein to use two methods
To construct input picture: the first, generating random number in the section of [0,0.1] to fill whole image;The second, it uses
Meshgrid function generates the grid in [0,1] section as input;
The parameter for determining full convolutional neural networks is added after being fitted above-mentioned three width image input neural network,
Energy function is that shooting obtains the root mean square MSE minimum of image in image after being added and step 1), and the MSE is defined as follows:
Wherein yiTo shoot obtained image in step 1),For three width input picture images after being added;
Using AdamOptimizer optimizer to iteration is optimized, optimal solution is obtained after iteration is multiple, at this time three it is defeated
Required phase main value, surface reflectivity and background light intensity will be fitted to respectively by entering image.
Further, in the step 3), using the method for the phase unwrapping based on Quality Map guiding method, steps are as follows:
3.1) then the pixel high from quality checks 4 pixels near this pixel point, then to this 4
A pixel carries out phase unwrapping, that is, phase unwrapping, wherein the formula of phase unwrapping are as follows:
3.2) pixel (not carrying out phase unwrapping) closed on around pixel for having carried out phase unwrapping is stored into " adjacent
In column ";
3.3) according to the Quality Map of phase, the high pixel of quality is selected from " adjacent column " carry out solution and twine, and update this
Column;
3.4) repeat 3.1), 3.2) step, until all phase unwrappings finish;
Wherein mass M is defined as:
D is second differnce, is defined as:
Wherein, V=unwrap (A (i, j-1)-A (i, j))-unwrap (A (i, j)-A (i, j+1))
H=unwrap (A (i-1, j)-A (i, j))-unwrap (A (i, j)-A (i+1, j))
D1=unwrap (A (i-1, j-1)-A (i, j))-unwrap (A (i, j)-A (i+1, j+1))
D2=unwrap (A (i-1, j+1)-A (i, j))-unwrap (A (i, j)-A (i+1, j-1))
Unwrap indicates that solution twines operation i.e. formula (4).
Beneficial effects of the present invention are mainly manifested in: reduce the quantity of Image Acquisition, it must without traditional neural network
The data set and training process of palpus reduce neural network to the requirement of hardware and operation runing time.
Detailed description of the invention
Fig. 1 is a kind of three-dimensional reconstruction system flow chart neural network based of the present invention;
Fig. 2 is a kind of three-dimensional reconstruction system hardware schematic neural network based of the present invention;
Fig. 3 is the structure chart of neural network of the present invention, wherein volume represents convolutional layer, lower to represent down-sampling layer, B represents BN
Layer, R represent Leaky ReLU activation primitive, and upper representative up-samples layer.
Specific embodiment
The present invention is described further with reference to the accompanying drawing:
Referring to Fig.1~Fig. 3, a kind of phase main value extracting method based on full convolutional neural networks, comprising the following steps:
1) referring to fig. 2, acquisition stripe pattern method is to project the bar graph encoded in advance to determinand, and use
Industrial camera acquires determinand bar graph, comprising the following steps:
1.1) striped precoding is carried out on computers, constructs the required bar graph being distributed by sinusoidal rule, striped
It is encoded according to following formula:
Wherein x is abscissa,For initial phase, γ is pre- calibration gamma value;
1.2) the good bar graph of precoding is projected to determinand surface using DLP projector;
1.3) stripe pattern of high definition is successively acquired using industrial camera.
2) full convolutional neural networks model is constructed, training parameter and loss function are set, 1) picture obtained in is inputted
Into neural network, neural network is run, required phase main value is obtained, comprising the following steps:
2.1) neural network constructs
When projector projects the bar graph encoded in advance to determinand surface, due to by body surface height
Modulation, the deforming stripe obtained by CCD camera indicate are as follows:
Wherein a (x, y) and b (x, y) reflects the variation of bias light and surface reflectivity respectively; It is to be calculated
Relative phase values, also referred to as phase main value, it reflects elevation information in object corresponding points;
Referring to Fig. 3, a full convolutional neural networks model is constructed, the layer of entire full convolutional neural networks includes ten portions
Point, wherein first layer to layer 5 is coding layer, and each layer is made of two convolutional layers and two BN layers, wherein the first convolutional layer
Convolution kernel size be 3 × 3, quantity is 128, stride 2, and the convolution kernel size of the second convolutional layer is 3 × 3, quantity 128
A, stride 1, the image of input successively passes through the first convolutional layer, BN layers and Leaky ReLU activation primitive first, then successively
By the second convolutional layer, BN layers and the processing of Leaky ReLU activation primitive;
Layer 6 is decoding layer to the tenth layer, and each layer is by two convolutional layers, three BN layers and a up-sampling layer group
At, the convolution kernel size of two of them convolutional layer is 3 × 3, and quantity is 128, and stride 1, input feature vector successively passes through BN layers,
First convolutional layer, BN layers and Leaky ReLU activation primitive.Successively pass through the second convolutional layer, BN layers and Leaky ReLU again
Activation primitive processing up-samples feature by bilinear interpolation or closest method finally by up-sampling layer;
In addition, first layer and the tenth layer, the second layer and the 9th layer, third layer and the 8th layer, the 4th layer and layer 7, the 5th
Skip connections is also added between layer and layer 6, skip connections is by convolutional layer, BN layers and Leaky
ReLU activation primitive is constituted, and wherein the convolution kernel size of convolutional layer is 1 × 1, and convolution nuclear volume is 4;
The image that size is 512 × 512 × 1 is inputted into neural network, successively by first layer to layer 6, the spy of output
Levying size is 16 × 16 × 128, then successively by layer 6 to the tenth layer, the feature sizes of output are 512 × 512 × 128;
The convolution kernel and RELU activation primitive that the last one convolutional layer is 1 × 1 by 1 size are handled, and output feature is big
Small is 512 × 512 × 1;
2.2) neural network is run
The input of neural network is the three width sizes Uniform noise image being randomly generated identical with 1) middle shooting image, point
It is not set to be fitted to background a (x, y), surface reflectivity b (x, y) and phase main value φ (x, y), wherein to use two methods
Construct input picture, 1, in the section of [0,0.1] generate random number to fill whole image;2, using meshgrid function
The grid in [0,1] section is generated as input;
The parameter for determining full convolutional neural networks is added after being fitted above-mentioned three width image input neural network,
Energy function is that shooting obtains the root mean square MSE minimum of image in image after being added and step 1), and the MSE is defined as follows:
Wherein yiTo shoot obtained image in step 1),For three width input picture images after being added;
Using AdamOptimizer optimizer to iteration is optimized, optimal solution is obtained after iteration is multiple, at this time three it is defeated
Required phase main value, surface reflectivity and background light intensity will be fitted to respectively by entering image.
Further, using the method for the phase unwrapping based on Quality Map guiding method in the step 3), steps are as follows:
3.1) then the pixel high from quality checks 4 pixels near this pixel point, then to this 4
A pixel carries out phase unwrapping, that is, phase unwrapping, wherein the formula of phase unwrapping are as follows:
3.2) pixel (not carrying out phase unwrapping) closed on around pixel for having carried out phase unwrapping is stored into " adjacent
In column ";
3.3) according to the Quality Map of phase, the high pixel of quality is selected from " adjacent column " carry out solution and twine, and update this
Column;
3.4) repeat 3.1), 3.2) step, until all phase unwrappings finish;
Wherein mass M is defined as:
D is second differnce, is defined as:
Wherein, V=unwrap (A (i, j-1)-A (i, j))-unwrap (A (i, j)-A (i, j+1))
H=unwrap (A (i-1, j)-A (i, j))-unwrap (A (i, j)-A (i+1, j))
D1=unwrap (A (i-1, j-1)-A (i, j))-unwrap (A (i, j)-A (i+1, j+1))
D2=unwrap (A (i-1, j+1)-A (i, j))-unwrap (A (i, j)-A (i+1, j-1))
Unwrap indicates that solution twines operation i.e. formula (4).
Claims (4)
1. a kind of phase main value extracting method based on full convolutional neural networks, which is characterized in that the method includes following steps
It is rapid:
1) on computers will in advance coding needed for the bar graph by Sine distribution, the bar graph encoded in advance project to
Object is surveyed, and acquires determinand bar graph using industrial camera;
2) full convolutional neural networks model is constructed, training parameter and loss function are set, 1) picture obtained in is input to mind
Through running neural network, obtaining required phase main value in network;
3) solution is carried out to phase main value obtained in 2) using the method being oriented to based on Quality Map to twine, obtain accurate phase value.
2. a kind of phase main value extracting method based on full convolutional neural networks as described in claim 1, which is characterized in that institute
State in step 1), acquisition bar graph process the following steps are included:
1.1) on computers carry out striped precoding, construct it is required by sinusoidal rule be distributed bar graph, striped according to
Following formula is encoded:
Wherein I (x, y) is gray value of image, and x is abscissa;
1.2) design shooting optical path, and industrial camera, DLP projector and determinand are placed by designed optical path;
1.3) stripe pattern of high definition is successively acquired using industrial camera.
3. a kind of phase main value extracting method based on full convolutional neural networks as claimed in claim 2, which is characterized in that institute
State in step 2), extract phase main value process the following steps are included:
2.1) neural network constructs
When projector projects the bar graph encoded in advance to determinand surface, due to the tune by body surface height
System, the deforming stripe obtained by CCD camera indicate are as follows:
Wherein a (x, y) and b (x, y) reflects the variation of bias light and surface reflectivity respectively; It is to be calculated opposite
Phase value, also referred to as phase main value, it reflects elevation information in object corresponding points;
A full convolutional neural networks model is constructed, the layer of entire full convolutional neural networks includes ten parts, wherein first layer
It is coding layer to layer 5, each layer is made of two convolutional layers and two BN layers, wherein the convolution kernel size of the first convolutional layer
It is 3 × 3, quantity is 128, stride 2, and the convolution kernel size of the second convolutional layer is 3 × 3, and quantity is 128, stride 1,
The image of input first passes around the first convolutional layer, BN layers and Leaky ReLU activation primitive, using the second convolutional layer, BN layers
And Leaky ReLU activation primitive processing;
Layer 6 is decoding layer to the tenth layer, and each layer is made of two convolutional layers, three BN layers and a up-sampling layer,
In the convolution kernel sizes of two convolutional layers be 3 × 3, quantity is 128, and stride 1, input feature vector successively passes through BN layers, first
Convolutional layer, BN layers and Leaky ReLU activation primitive;Again successively by the second convolutional layer, BN layers and Leaky ReLU activation
Function processing up-samples feature by bilinear interpolation or closest method finally by up-sampling layer;
In addition, first layer and the tenth layer, the second layer and the 9th layer, third layer and the 8th layer, the 4th layer and layer 7, layer 5 with
Skip connections is also added between layer 6, skip connections is by convolutional layer, BN layers and Leaky
ReLU activation primitive is constituted, and wherein the convolution kernel size of convolutional layer is 1 × 1, and convolution nuclear volume is 4;
The image that size is 512 × 512 × 1 is inputted into neural network, successively by first layer to layer 6, the feature of output is big
Small is 16 × 16 × 128, then successively by layer 6 to the tenth layer, the feature sizes of output are 512 × 512 × 128;
The convolution kernel and RELU activation primitive that the last one convolutional layer is 1 × 1 by 1 size are handled, and output feature sizes are
512×512×1;
2.2) neural network is run
The input of neural network is the three width sizes Uniform noise image being randomly generated identical with 1) middle shooting image, is made respectively
It is fitted to background a (x, y), surface reflectivity b (x, y) and phase main value φ (x, y), wherein using two methods come structure
It builds input picture: the first, generating random number in the section of [0,0.1] to fill whole image;The second, using meshgrid letter
Number generates the grid in [0,1] section as input;
The parameter for determining full convolutional neural networks is added, energy after being fitted above-mentioned three width image input neural network
Function is that shooting obtains the root mean square MSE minimum of image in image after being added and step 1), and the MSE is defined as follows:
Wherein yiTo shoot obtained image in step 1),For three width input picture images after being added;
Using AdamOptimizer optimizer to iteration is optimized, optimal solution is obtained after iteration is multiple, the figure of three width input at this time
As required phase main value, surface reflectivity and background light intensity will be fitted to respectively.
4. a kind of phase main value extracting method based on full convolutional neural networks as claimed in claim 1 or 2, feature exist
In in the step 3), using the phase unwrapping based on Quality Map guiding method, steps are as follows:
3.1) then the pixel high from quality checks 4 pixels near this pixel point, then to this 4 pictures
Vegetarian refreshments carries out phase unwrapping, that is, phase unwrapping, wherein the formula of phase unwrapping are as follows:
3.2) pixel around pixel of closing on for having carried out phase unwrapping is stored into " adjacent column ";
3.3) according to the Quality Map of phase, the high pixel of quality is selected from " adjacent column " carry out solution and twine, and update this column;
3.4) repeat 3.1), 3.2) step, until all phase unwrappings finish;
Wherein mass M is defined as:
D is second differnce, is defined as:
Wherein, V=unwrap (A (i, j-1)-A (i, j))-unwrap (A (i, j)-A (i, j+1))
H=unwrap (A (i-1, j)-A (i, j))-unwrap (A (i, j)-A (i+1, j))
D1=unwrap (A (i-1, j-1)-A (i, j))-unwrap (A (i, j)-A (i+1, j+1))
D2=unwrap (A (i-1, j+1)-A (i, j))-unwrap (A (i, j)-A (i+1, j-1))
Unwrap indicates that solution twines operation i.e. formula (4).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910347403.6A CN110163817B (en) | 2019-04-28 | 2019-04-28 | Phase principal value extraction method based on full convolution neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910347403.6A CN110163817B (en) | 2019-04-28 | 2019-04-28 | Phase principal value extraction method based on full convolution neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110163817A true CN110163817A (en) | 2019-08-23 |
CN110163817B CN110163817B (en) | 2021-06-18 |
Family
ID=67638752
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910347403.6A Active CN110163817B (en) | 2019-04-28 | 2019-04-28 | Phase principal value extraction method based on full convolution neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110163817B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110487216A (en) * | 2019-09-20 | 2019-11-22 | 西安知象光电科技有限公司 | A kind of fringe projection 3-D scanning method based on convolutional neural networks |
CN110500957A (en) * | 2019-09-10 | 2019-11-26 | 中国科学院苏州纳米技术与纳米仿生研究所 | A kind of active three-D imaging method, device, equipment and storage medium |
CN111189414A (en) * | 2020-01-09 | 2020-05-22 | 西安知象光电科技有限公司 | Real-time single-frame phase extraction method |
CN111461295A (en) * | 2020-03-20 | 2020-07-28 | 南京理工大学 | Single-frame stripe analysis method for generating antagonistic neural network based on multiple scales |
CN111812647A (en) * | 2020-07-11 | 2020-10-23 | 桂林电子科技大学 | Phase unwrapping method for interferometric synthetic aperture radar |
CN113340211A (en) * | 2021-08-03 | 2021-09-03 | 中国工程物理研究院激光聚变研究中心 | Interference image phase demodulation method based on deep learning |
CN113985566A (en) * | 2021-09-10 | 2022-01-28 | 西南科技大学 | Scattered light focusing method based on spatial light modulation and neural network |
CN114152217A (en) * | 2022-02-10 | 2022-03-08 | 南京南暄励和信息技术研发有限公司 | Binocular phase expansion method based on supervised learning |
CN117523344A (en) * | 2024-01-08 | 2024-02-06 | 南京信息工程大学 | Interference phase unwrapping method based on phase quality weighted convolution neural network |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103543453A (en) * | 2013-10-28 | 2014-01-29 | 北京理工大学 | Elevation inversion method for geosynchronous orbit synthetic aperture radar interference |
WO2018107584A1 (en) * | 2016-12-15 | 2018-06-21 | 东南大学 | Error correction method for grating projection three-dimensional measurement system |
CN109253708A (en) * | 2018-09-29 | 2019-01-22 | 南京理工大学 | A kind of fringe projection time phase method of deploying based on deep learning |
CN109459923A (en) * | 2019-01-02 | 2019-03-12 | 西北工业大学 | A kind of holographic reconstruction algorithm based on deep learning |
CN109596227A (en) * | 2018-12-06 | 2019-04-09 | 浙江大学 | A kind of phase recovery detection system of the optical element intermediate frequency error of convolutional neural networks priori enhancing |
-
2019
- 2019-04-28 CN CN201910347403.6A patent/CN110163817B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103543453A (en) * | 2013-10-28 | 2014-01-29 | 北京理工大学 | Elevation inversion method for geosynchronous orbit synthetic aperture radar interference |
WO2018107584A1 (en) * | 2016-12-15 | 2018-06-21 | 东南大学 | Error correction method for grating projection three-dimensional measurement system |
CN109253708A (en) * | 2018-09-29 | 2019-01-22 | 南京理工大学 | A kind of fringe projection time phase method of deploying based on deep learning |
CN109596227A (en) * | 2018-12-06 | 2019-04-09 | 浙江大学 | A kind of phase recovery detection system of the optical element intermediate frequency error of convolutional neural networks priori enhancing |
CN109459923A (en) * | 2019-01-02 | 2019-03-12 | 西北工业大学 | A kind of holographic reconstruction algorithm based on deep learning |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110500957A (en) * | 2019-09-10 | 2019-11-26 | 中国科学院苏州纳米技术与纳米仿生研究所 | A kind of active three-D imaging method, device, equipment and storage medium |
CN110500957B (en) * | 2019-09-10 | 2021-09-14 | 中国科学院苏州纳米技术与纳米仿生研究所 | Active three-dimensional imaging method, device, equipment and storage medium |
CN110487216A (en) * | 2019-09-20 | 2019-11-22 | 西安知象光电科技有限公司 | A kind of fringe projection 3-D scanning method based on convolutional neural networks |
CN110487216B (en) * | 2019-09-20 | 2021-05-25 | 西安知象光电科技有限公司 | Fringe projection three-dimensional scanning method based on convolutional neural network |
CN111189414B (en) * | 2020-01-09 | 2021-09-03 | 西安知象光电科技有限公司 | Real-time single-frame phase extraction method |
CN111189414A (en) * | 2020-01-09 | 2020-05-22 | 西安知象光电科技有限公司 | Real-time single-frame phase extraction method |
CN111461295B (en) * | 2020-03-20 | 2022-08-16 | 南京理工大学 | Single-frame stripe analysis method for generating antagonistic neural network based on multiple scales |
CN111461295A (en) * | 2020-03-20 | 2020-07-28 | 南京理工大学 | Single-frame stripe analysis method for generating antagonistic neural network based on multiple scales |
CN111812647A (en) * | 2020-07-11 | 2020-10-23 | 桂林电子科技大学 | Phase unwrapping method for interferometric synthetic aperture radar |
CN113340211A (en) * | 2021-08-03 | 2021-09-03 | 中国工程物理研究院激光聚变研究中心 | Interference image phase demodulation method based on deep learning |
CN113985566A (en) * | 2021-09-10 | 2022-01-28 | 西南科技大学 | Scattered light focusing method based on spatial light modulation and neural network |
CN113985566B (en) * | 2021-09-10 | 2023-09-12 | 西南科技大学 | Scattered light focusing method based on spatial light modulation and neural network |
CN114152217A (en) * | 2022-02-10 | 2022-03-08 | 南京南暄励和信息技术研发有限公司 | Binocular phase expansion method based on supervised learning |
CN114152217B (en) * | 2022-02-10 | 2022-04-12 | 南京南暄励和信息技术研发有限公司 | Binocular phase expansion method based on supervised learning |
CN117523344A (en) * | 2024-01-08 | 2024-02-06 | 南京信息工程大学 | Interference phase unwrapping method based on phase quality weighted convolution neural network |
CN117523344B (en) * | 2024-01-08 | 2024-03-19 | 南京信息工程大学 | Interference phase unwrapping method based on phase quality weighted convolution neural network |
Also Published As
Publication number | Publication date |
---|---|
CN110163817B (en) | 2021-06-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110163817A (en) | A kind of phase main value extracting method based on full convolutional neural networks | |
Laine et al. | Modular primitives for high-performance differentiable rendering | |
Yariv et al. | Bakedsdf: Meshing neural sdfs for real-time view synthesis | |
US20160307368A1 (en) | Compression and interactive playback of light field pictures | |
JP5830546B2 (en) | Determination of model parameters based on model transformation of objects | |
CN104541127B (en) | Image processing system and image processing method | |
CN109945802B (en) | Structured light three-dimensional measurement method | |
CN106101535B (en) | A kind of video stabilizing method based on part and mass motion disparity compensation | |
US8803902B2 (en) | Computing level of detail for anisotropic filtering | |
CN108955571B (en) | The method for three-dimensional measurement that double frequency heterodyne is combined with phase-shift coding | |
CN106408523A (en) | Denoising filter | |
CN108955574A (en) | A kind of method for three-dimensional measurement and system | |
WO2020169983A1 (en) | Facial shape representation and generation system and method | |
Andersson et al. | Adaptive texture space shading for stochastic rendering | |
CN104025155B (en) | Variable depth compresses | |
CN117011478B (en) | Single image reconstruction method based on deep learning and stripe projection profilometry | |
CN109993701A (en) | A method of the depth map super-resolution rebuilding based on pyramid structure | |
CN113008163A (en) | Encoding and decoding method based on frequency shift stripes in structured light three-dimensional reconstruction system | |
CN117132704A (en) | Three-dimensional reconstruction method of dynamic structured light, system and computing equipment thereof | |
CN112802084A (en) | Three-dimensional topography measuring method, system and storage medium based on deep learning | |
Fu et al. | High Dynamic Range Structured Light 3-D Measurement Based on Region Adaptive Fringe Brightness | |
CN116645466A (en) | Three-dimensional reconstruction method, electronic equipment and storage medium | |
CN107644393A (en) | A kind of Parallel Implementation method of the abundance algorithm for estimating based on GPU | |
Law et al. | Projector placement planning for high quality visualizations on real-world colored objects | |
CN114166150B (en) | Stripe reflection three-dimensional measurement method, system and storage medium |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |