CN206930789U - A kind of subject imaging system based on compressed sensing - Google Patents
A kind of subject imaging system based on compressed sensing Download PDFInfo
- Publication number
- CN206930789U CN206930789U CN201720623572.4U CN201720623572U CN206930789U CN 206930789 U CN206930789 U CN 206930789U CN 201720623572 U CN201720623572 U CN 201720623572U CN 206930789 U CN206930789 U CN 206930789U
- Authority
- CN
- China
- Prior art keywords
- image
- unit
- generation unit
- imaging system
- subject
- 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
Abstract
The utility model discloses a kind of subject imaging system based on compressed sensing, including light source generation unit, filter unit, image generation unit, image acquisition units and the image reconstruction unit being sequentially connected;Light source generation unit generation experiment laser, filter unit filters out High frequency scattering light and forms directional light, the experimental image that image generation unit generation subject image is superimposed with particular measurement matrix, image acquisition units are compressed sampling to the experimental image of generation, and image reconstruction unit is reconstructed to sampled data and recovers subject image.The utility model can greatly reduce image recognition, the data volume recorded in images match, improve the real-time of system, and the parallel processing for machine vision and artificial intelligence provides possibility.
Description
Technical field
It the utility model is related to image imaging, target identification technology field, more particularly to a kind of pair based on compressed sensing
As imaging system.
Background technology
In recent years, the technology of the subjects such as pattern-recognition, computer vision, artificial intelligence is merged one by target identification well
Rise, turn into a very popular problem in vision research field, have broad application prospects.Itd is proposed from target identification method
Since, many experts, researchers have carried out the research of related work.But due to most target identification method
All be reuse after an image is obtained identification technology to image carry out respective handling, the real-time for causing system be not it is fine,
And the image data amount that thus introduces it is big the problem of also turn into the bottleneck for restricting its development.
As imaging and information technology develop rapidly, people are in daily life to electronic product, digital information to needing
Ask also growing day by day.But limited by information transfer rate and information processing rate, it is impossible to disregard the sampling of the raising information of cost
Rate is so as to obtaining the data of high-quality.On the premise of ensuring that quality that information is showed is substantially constant, adopted with a small amount of
Sample obtains the information of similar mass, becomes the task of top priority.
Emerging compressive sensing theory is linearly thrown with the non-self-adapting far below Nyquist sampling rate collection signal
Shadow value, then by solving an optimization problem, primary signal is accurately reconstructed, the data of system acquisition can be substantially reduced
Amount.The a series of target identification scheme based on compressed sensing is suggested in succession, but although these schemes reduce system and adopted
The data volume of collection, but it is cumbersome to still suffer from image information when carrying out further image recognition processing in the later stage, is unfavorable for image
Identification, and these schemes are mostly realized using the digital signaling of electrical domain, the advantages of can not giving full play to parallel processing.
Utility model content
The shortcomings that the purpose of this utility model is to overcome prior art and deficiency, there is provided a kind of pair based on compressed sensing
As imaging system, it can realize compression of images, image sampling and image recognition simultaneously under pure area of light, greatly reduce image knowledge
Not, the data volume recorded in images match, the storage of its mass data and transmission problem are solved, improves the real-time of system
Property, the parallel processing for machine vision and artificial intelligence provides possibility.
The purpose of this utility model is realized by following technical scheme:
A kind of subject imaging system based on compressed sensing, including be sequentially connected light source generation unit, filter unit, figure
As generation unit, image acquisition units and image reconstruction unit;
Light source generation unit generation experiment laser;
The decay of filter unit control laser intensity, filter out High frequency scattering light, form zero-order terms and twin-image, then controlling diffraction
Hot spot forms directional light;
The experimental image that image generation unit generation subject image is superimposed with particular measurement matrix;
Image acquisition units are compressed sampling to experimental image;
Image reconstruction unit is reconstructed to sampled data and recovers subject image.
Preferably, light source generation unit includes laser and can reflect the speculum for changing its direction of propagation.
Preferably, filter unit includes circular adjustable attenuator, pinhole filter and Fu parallel with light path, be sequentially arranged
In leaf lens.
Preferably, image generation unit includes the first polarizer that is parallel with light path, being sequentially arranged, loading particular measurement square
The spatial light modulator and the second polarizer of battle array.
Preferably, image acquisition units include convergent lens and single-photon detector parallel with light path, be sequentially arranged.
Preferably, image reconstruction unit includes computer.
The utility model compared with prior art, has the following advantages that and beneficial effect:
Compared to prior art, the utility model combines emerging compressive sensing theory and mode identification technology, in full light
Compression of images, image sampling and image recognition are realized simultaneously in the environment of domain, i.e., is achieved that target identification in imaging.The technology
The data volume of storage and transmission is greatly reduced, the efficiency of target identification is drastically increased, is machine vision and artificial intelligence
Parallel processing provide possibility.
Brief description of the drawings
Fig. 1 is the theory diagram of the subject imaging system based on compressed sensing in the present embodiment;
Fig. 2 is the structural representation of the subject imaging system based on compressed sensing in the present embodiment;
Fig. 3 is the parts of images for the one group of training sample database established in the present embodiment;
Fig. 4 is the image of the particular measurement matrix that training sample database obtains in the present embodiment;
Fig. 5 is the simulation experiment result figure of the object imaging method based on compressed sensing in the present embodiment.
Embodiment
The utility model is described in further detail with reference to embodiment and accompanying drawing, but implementation of the present utility model
Mode not limited to this.
A kind of subject imaging system based on compressed sensing, as shown in Figure 1, Figure 2, including light source generation unit 11, filter unit
12nd, image generation unit 13, image acquisition units 14, image reconstruction unit 15.
Light source generation unit 11, laser is tested for generating.Filter unit 12, for decay test laser, filter out high frequency
Zero-order terms and twin-image is obtained after scattering light, and it is directional light to adjust diffraction pattern.Image generation unit 13, for realizing object figure
It is superimposed as 16 with the particular measurement matrix 17 obtained after training, generates experimental image.Image acquisition units 14, for gathering reality
Data are tested, and are sent to image reconstruction unit 15.Image reconstruction unit 15, go out subject image for the data reconstruction according to screening
16。
Light source generation unit 11 includes He-Ne laser 111 and strengthens aluminium reflector 112.He-Ne laser 111 sends one
Shu Jiguang, and change its direction of propagation by strengthening the reflection of aluminium reflector 112.
Filter unit 12 includes circular adjustable attenuator 121, pinhole filter 122 and fourier lense 123.It is circular adjustable
Attenuator 121 controls laser intensity decay, and the laser after decay filters out High frequency scattering light by pinhole filter 122, and allows not
The zero order light being scattered forms zero-order terms and twin-image by pin hole, and then diffraction pattern is formed parallel by fourier lense 123
Light.
Image generation unit 13 includes the first polarizer 131, the polarizer 133 of spatial light modulator 132 and second.By filtering
The experiment light irradiation subject image 16 that unit 12 is sent, particular measurement matrix 17 is loaded in spatial light modulator 132, and in fact
Existing subject image 16 is superimposed with particular measurement matrix 17, and the first polarizer 131 and the second polarizer 133 are respectively placed in spatial light
Net amplitude state is at before and after modulator 132.
Image acquisition units 14 include convergent lens 141 and single-photon detector 142.Convergent lens 141 generates image
The experimental image data convergence of unit 13 by the single-point acquiring optical signal of single-photon detector 142 and is converted to afterwards in a bit
Electric signal.
Reconfiguration unit 15 includes computer 151.Computer 151 is used for screening experiment data and optimized by minimum full variation
Algorithm reconstructs subject image 16.
The compression holographic imaging step of the detailed description below system:
Step S1:The laser reflection that light source is sent changes it and propagates light path.
Specifically, beam of laser is sent by He-Ne laser 111;The laser changes by strengthening the reflection of aluminium reflector 112
Become its direction of propagation.
Step S2:Experiment laser is subjected to attenuation processing, obtains zero-order terms and twin-image after filtering out High frequency scattering light, and adjust
Diffraction pattern is directional light.
Specifically, along optical circuit path, a circular adjustable attenuator 121 is set to control laser intensity decay;Can in circle
The rear of controlled attenuator 121 sets pinhole filter 122;High frequency scattering light is filtered out by pinhole filter 122, spread out with obtaining zero level
Penetrate hot spot;One fourier lense 123 is set at the rear of pinhole filter 122, and diffraction pattern forms flat by fourier lense 123
Row light.
Step S3:Directional light is radiated in subject image 16, the lab diagram that generation subject image is superimposed with calculation matrix
Picture.
Specifically, the experiment light irradiation subject image 16 sent by filter unit 12;Add in spatial light modulator 132
Carry particular measurement matrix 17, the transmitted light of subject image 16 and being superimposed for particular measurement matrix 17;First polarizer 131 and second
Polarizer 133 is at net amplitude state before and after being respectively placed in spatial light modulator 132.
In step s3, it is the image training sample set of p × q pixels provided with N sizes according to principal component analytical method
X, each sample form a vectorial X by its grey scale pixel valuei, by the vectorial training sample set X={ X formed1,X2,…,XN,
Training sample set average is calculated first:
Centralization further is carried out to data:
Covariance matrix is asked to the data after centralization:
By covariance matrix R eigenvalue λiArranged according to order from big to small, eigenvalue λiIt is bigger to represent its phenogram
As the ability of information is stronger, therefore, the corresponding characteristic vector u of preceding m characteristic valueiForm principal component matrix U:
U=[u1,u2,…,um]
This m component is just extracted the feature of representative image main information.Particular measurement matrix 17 is used as after being turned order.
It is only right in the scene of complexity so as to realize because the particular measurement matrix 17 includes the information of specific objective subject image 16
Specific objective is imaged, and ignores the background image unrelated with target image.
According to compressed sensing principle, the particular measurement matrix 17 for training to obtain through Sample Storehouse is expressed as:
Φ=UT=[φ1,φ2,…,φm]T(φi=ui)
Step S4:Sampling is compressed to the experimental image of generation, and subject image is gone out according to the data reconstruction.
The output voltage of the highly sensitive photodiode of single-photon detector 142 is expressed as:
Wherein, x ∈ Rp×q, x expression subject images, Represent particular measurement matrix
The n-th dimension, it, which is realized, measures subject image x n-th;
Repeat this process m times, available measured value Y is:
Wherein, Φ ∈ Rm×(p×q)Be training sample database obtain particular measurement matrix, Y ∈ Rm×1It is measured value;
Screen to obtain accurate experimental data using time domain screening and size.Further utilize minimum full Variational Optimal Algorithm
Reconstruct subject image.
Fig. 3 is the parts of images of one group of training sample database.The training sample database is shot by same object under different angle
Obtain, totally 360, size is 800 × 600 pixels.Specifically establishing process is:Training object is positioned over black background
Under circular rotating platform on, by AVT cameras in fixed position shooting image.From 0 degree to 359 degree, 1 degree is often rotated just greyish white
Pattern is shot once, and one is obtained the training sample image of 360 different angles.The image provided in Fig. 3 be followed successively by 0 degree, 20
The image of acquisition is shot under degree, 40 degree ... 180 degree angles.
Fig. 4 is the image for the particular measurement matrix that training sample database obtains;It is respectively take particular measurement matrix the in figure
1st, the image of 101,201,301 row data generation, due to using principal component analytical method, therefore after line number comes image above relatively
The abundant information and pixel value in face are higher.
Fig. 5 is the simulation experiment result figure of the object imaging method based on compressed sensing.Carried by what AVT cameras were shot
For the image of ambient interferences thing respectively as shown in Fig. 5 (a) and 5 (c), size is 800 × 600 pixels.Obtained using training
Particular measurement matrix is compressed sampling to it, and represents only to use using minimum full Variational Optimal Algorithm reconstruct, Fig. 5 (b)
0.075% measurement data is to Fig. 5 (a) reconstruction result, and Fig. 5 (d) expressions are only using 0.075% measurement data to Fig. 5 (c)
Reconstruction result.Simulation results prove that this method can preferably remove the jamming pattern in image, only to special object into
Picture, it is practicable to demonstrate the subject imaging system based on compressed sensing of proposition.
Compared to prior art, the utility model combines emerging compressive sensing theory and mode identification technology, in full light
Compression sampling and image recognition are realized simultaneously in the environment of domain, the data volume of storage and transmission is greatly reduced, drastically increases
The efficiency of target identification, realize the requirement of compressed object imaging.
Above-described embodiment is the preferable embodiment of the utility model, but embodiment of the present utility model is not by above-mentioned
The limitation of embodiment, it is other it is any without departing from Spirit Essence of the present utility model with made under principle change, modify, replace
Generation, combination, simplify, should be equivalent substitute mode, be included within the scope of protection of the utility model.
Claims (6)
- A kind of 1. subject imaging system based on compressed sensing, it is characterised in that including be sequentially connected light source generation unit, filter Ripple unit, image generation unit, image acquisition units and image reconstruction unit;Light source generation unit generation experiment laser;The decay of filter unit control laser intensity, filter out High frequency scattering light, form zero-order terms and twin-image, then controlling diffraction pattern Form directional light;The experimental image that image generation unit generation subject image is superimposed with particular measurement matrix;Image acquisition units are compressed sampling to experimental image;Image reconstruction unit is reconstructed to sampled data and recovers subject image.
- 2. the subject imaging system according to claim 1 based on compressed sensing, it is characterised in that light source generation unit bag Include laser and the speculum for changing its direction of propagation can be reflected.
- 3. the subject imaging system according to claim 1 based on compressed sensing, it is characterised in that filter unit include with Circular adjustable attenuator, pinhole filter and the fourier lense that light path is parallel, is sequentially arranged.
- 4. the subject imaging system according to claim 1 based on compressed sensing, it is characterised in that image generation unit bag Include the first polarizer that is parallel with light path, being sequentially arranged, load the spatial light modulator and the second polarizer of particular measurement matrix.
- 5. the subject imaging system according to claim 1 based on compressed sensing, it is characterised in that image acquisition units bag Include convergent lens and single-photon detector parallel with light path, be sequentially arranged.
- 6. the subject imaging system according to claim 1 based on compressed sensing, it is characterised in that image reconstruction unit bag Include computer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201720623572.4U CN206930789U (en) | 2017-06-01 | 2017-06-01 | A kind of subject imaging system based on compressed sensing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201720623572.4U CN206930789U (en) | 2017-06-01 | 2017-06-01 | A kind of subject imaging system based on compressed sensing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN206930789U true CN206930789U (en) | 2018-01-26 |
Family
ID=61345881
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201720623572.4U Active CN206930789U (en) | 2017-06-01 | 2017-06-01 | A kind of subject imaging system based on compressed sensing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN206930789U (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107121709A (en) * | 2017-06-01 | 2017-09-01 | 华南师范大学 | A kind of subject imaging system and its imaging method based on compressed sensing |
CN115442505A (en) * | 2022-08-30 | 2022-12-06 | 山西大学 | Single photon compression sensing imaging system and method thereof |
-
2017
- 2017-06-01 CN CN201720623572.4U patent/CN206930789U/en active Active
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107121709A (en) * | 2017-06-01 | 2017-09-01 | 华南师范大学 | A kind of subject imaging system and its imaging method based on compressed sensing |
WO2018218974A1 (en) * | 2017-06-01 | 2018-12-06 | 华南师范大学 | Compressed sensing based object imaging system and imaging method thereof |
US11368608B2 (en) | 2017-06-01 | 2022-06-21 | South China Normal University | Compressed sensing based object imaging system and imaging method therefor |
CN115442505A (en) * | 2022-08-30 | 2022-12-06 | 山西大学 | Single photon compression sensing imaging system and method thereof |
CN115442505B (en) * | 2022-08-30 | 2023-07-21 | 山西大学 | Single photon compressed sensing imaging system and method thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107121709A (en) | A kind of subject imaging system and its imaging method based on compressed sensing | |
CN111260584A (en) | Underwater degraded image enhancement method based on GAN network | |
CN106504248A (en) | Vehicle damage method of discrimination based on computer vision | |
CN103971329B (en) | A kind of multisource image anastomosing method based on genetic optimization cell neural network | |
CN109636742A (en) | The SAR image of network and the mode conversion method of visible images are generated based on confrontation | |
CN107966801A (en) | A kind of high speed Fourier lamination imaging device and reconstructing method based on ring illumination | |
CN112116601B (en) | Compressed sensing sampling reconstruction method and system based on generation of countermeasure residual error network | |
CN108844464B (en) | Compressed sensing imaging device and method based on entangled two-photon signals | |
CN206930789U (en) | A kind of subject imaging system based on compressed sensing | |
CN106056067B (en) | Low-resolution face image recognition methods based on corresponding relationship prediction | |
CN111986084A (en) | Multi-camera low-illumination image quality enhancement method based on multi-task fusion | |
Zhai et al. | Foveated ghost imaging based on deep learning | |
CN110225260A (en) | A kind of three-dimensional high dynamic range imaging method based on generation confrontation network | |
Qi et al. | A new heterogeneous neural network model and its application in image enhancement | |
Jung et al. | Fusionnet: Multispectral fusion of RGB and NIR images using two stage convolutional neural networks | |
Jin et al. | Learning to see through reflections | |
CN110650340A (en) | Space-time multiplexing compressed video imaging method | |
CN110319790A (en) | A kind of terrible imaging method of full color calculating based on colored speckle field | |
Li et al. | Single image dehazing with an independent detail-recovery network | |
CN111028302B (en) | Compressed object imaging method and system based on deep learning | |
Wang et al. | Multiscale supervision-guided context aggregation network for single image dehazing | |
Wang et al. | AGCycleGAN: Attention-guided CycleGAN for single underwater image restoration | |
CN113298700A (en) | High-resolution image reconstruction method in scattering scene | |
Nan et al. | An image enhancement method in coal mine underground based on deep retinex network and fusion strategy | |
CN105700135B (en) | A kind of high speed large scene phase recovery data acquisition device of natural lighting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
GR01 | Patent grant | ||
GR01 | Patent grant |