WO2018218974A1 - 一种基于压缩感知的对象成像系统及其成像方法 - Google Patents

一种基于压缩感知的对象成像系统及其成像方法 Download PDF

Info

Publication number
WO2018218974A1
WO2018218974A1 PCT/CN2018/073172 CN2018073172W WO2018218974A1 WO 2018218974 A1 WO2018218974 A1 WO 2018218974A1 CN 2018073172 W CN2018073172 W CN 2018073172W WO 2018218974 A1 WO2018218974 A1 WO 2018218974A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
experimental
measurement matrix
specific measurement
unit
Prior art date
Application number
PCT/CN2018/073172
Other languages
English (en)
French (fr)
Inventor
李军
雷苗
戴晓芳
王尚媛
钟婷
王琛
谢萍
Original Assignee
华南师范大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 华南师范大学 filed Critical 华南师范大学
Priority to US16/618,376 priority Critical patent/US11368608B2/en
Publication of WO2018218974A1 publication Critical patent/WO2018218974A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V8/00—Prospecting or detecting by optical means
    • G01V8/10—Detecting, e.g. by using light barriers
    • G—PHYSICS
    • G02—OPTICS
    • G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
    • G02B5/00—Optical elements other than lenses
    • G02B5/30—Polarising elements
    • G02B5/3025—Polarisers, i.e. arrangements capable of producing a definite output polarisation state from an unpolarised input state
    • G02B5/3033—Polarisers, i.e. arrangements capable of producing a definite output polarisation state from an unpolarised input state in the form of a thin sheet or foil, e.g. Polaroid
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • 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
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
    • H04N23/50—Constructional details
    • H04N23/55—Optical parts specially adapted for electronic image sensors; Mounting thereof
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
    • H04N23/56—Cameras or camera modules comprising electronic image sensors; Control thereof provided with illuminating means
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • the present invention relates to the field of image imaging and target recognition technologies, and in particular, to an object imaging system based on compressed sensing and an imaging method thereof.
  • target recognition has been a good combination of technologies such as pattern recognition, computer vision, and artificial intelligence, which has become a very hot topic in the field of visual research and has broad application prospects. Since the introduction of the target recognition method, many experts and researchers have carried out research on related work. However, since most of the target recognition methods use the recognition technology to process the image after obtaining the image, the real-time performance of the system is not very good, and the problem of the large amount of image data introduced thereby becomes a constraint to its development. The bottleneck.
  • the emerging theory of compressed sensing is a non-adaptive linear projection value that is much lower than the Nyquist sampling rate. Then, by solving an optimization problem and accurately reconstructing the original signal, the amount of data collected by the system can be greatly reduced.
  • a series of target recognition schemes based on compressed sensing have been proposed one after another, but although these schemes reduce the amount of data collected by the system, they still face cumbersome image information when performing further image recognition processing, which is not conducive to image recognition, and most of these schemes It is realized by the digital signal method of the electric field, and the advantages of parallel processing cannot be fully utilized.
  • the main object of the present invention is to overcome the shortcomings and shortcomings of the prior art, and provide an object imaging system based on compressed sensing, which can simultaneously implement image compression, image sampling and image recognition in a pure optical domain, thereby greatly reducing image recognition,
  • the amount of data recorded in image matching solves the problem of storage and transmission of massive data, improves the real-time performance of the system, and provides the possibility of parallel processing of machine vision and artificial intelligence.
  • Another object of the present invention is to provide an imaging method based on the above system.
  • An object imaging system based on compressed sensing comprising a light source generating unit, a filtering unit, an image generating unit, an image collecting unit and an image reconstruction unit connected in sequence;
  • the light source generating unit generates an experimental laser
  • the filtering unit controls the laser light intensity attenuation, filters out the high frequency scattered light, forms a zero-order diffraction spot, and then controls the diffraction spot to form parallel light;
  • the image generation unit generates an experimental image in which the object image is superimposed with the specific measurement matrix
  • the image acquisition unit performs compression sampling on the experimental image
  • the image reconstruction unit reconstructs the sampled data to recover the object image.
  • the light source generating unit includes a laser and a mirror that can reflect and change its propagation direction.
  • the filtering unit comprises a circular adjustable attenuator, a pinhole filter and a Fourier lens arranged in parallel with the optical path.
  • the image generating unit includes a first polarizing plate arranged in parallel with the optical path, a spatial light modulator loaded with a specific measurement matrix, and a second polarizing plate.
  • the image acquisition unit comprises a converging lens and a single photon detector arranged in parallel with the optical path.
  • the image reconstruction unit comprises a computer.
  • Step S1 using a laser light source to emit an experimental laser
  • Step S2 attenuating the experimental laser, filtering out the high-frequency scattered light to obtain a zero-order diffracted spot, and adjusting the diffracted spot to be parallel light;
  • Step S3 illuminating the parallel light on the image of the object, and generating an experimental image superimposed by the object image and the specific measurement matrix through a spatial light modulator loaded with a specific measurement matrix;
  • Step S4 Perform compression sampling on the experimental image, and the image reconstruction unit reconstructs the object image according to the sampled data.
  • the calculation method of the specific measurement matrix in step S3 is: establishing a sample library containing an image of a specific target object, and training the sample image by principal component analysis to obtain a specific measurement matrix.
  • the eigenvalues ⁇ i of the covariance matrix R are arranged in descending order, and the eigenvectors u i corresponding to the first m eigenvalues constitute the principal component matrix U:
  • the specific measurement matrix obtained by training the sample library is expressed as:
  • the experimental measurement data is collected in step S4: the optical signal is collected at a single point and converted into an electrical signal, and the output voltage value is expressed as:
  • x ⁇ R p ⁇ q , x represents an object image, N ⁇ 1,2, ⁇ ,m ⁇ , Represents the nth dimension of a particular measurement matrix that implements the nth measurement of the object image x;
  • ⁇ ⁇ R m ⁇ (p ⁇ q) is a specific measurement matrix
  • Y ⁇ R m ⁇ 1 is a measured value
  • the image reconstruction unit includes a data filtering processing module and an image reconstruction processing module, and the data filtering processing module is configured to perform time domain filtering and size screening on the data collected by the image collecting unit to obtain accurate experimental data;
  • the reconstruction processing module reconstructs the object image using the minimum total variation optimization algorithm.
  • the present invention has the following advantages and beneficial effects:
  • the present invention trains a specific measurement matrix containing specific object information by a machine learning method (ie, a principal component method) and combines compression sensing technology to achieve imaging of a specific target object in a complex scene.
  • a machine learning method ie, a principal component method
  • the present invention combines the emerging compressed sensing theory and pattern recognition technology to simultaneously achieve image compression, image sampling and image recognition in an all-optical environment, that is, target recognition is realized at the time of imaging. This technology greatly reduces the amount of data stored and transmitted, greatly improving the efficiency of target recognition, and provides the possibility of parallel processing of machine vision and artificial intelligence.
  • FIG. 1 is a schematic block diagram of an object imaging system based on compressed sensing in the embodiment
  • FIG. 2 is a schematic structural diagram of an object imaging system based on compression sensing in the embodiment
  • 3 is a partial image of a set of training sample libraries established in the embodiment
  • FIG. 5 is a simulation experiment result diagram of the object imaging method based on the compressed sensing in the present embodiment.
  • An object imaging system based on compressed sensing includes a light source generating unit 11, a filtering unit 12, an image generating unit 13, an image collecting unit 14, and an image reconstructing unit 15.
  • the light source generating unit 11 is configured to generate an experimental laser.
  • the filtering unit 12 is configured to attenuate the experimental laser, filter the high-frequency scattered light to obtain a zero-order diffracted spot, and adjust the diffracted spot to be parallel light.
  • the image generating unit 13 is configured to implement superposition of the object image 16 and the specific measurement matrix 17 obtained after the training to generate an experimental image.
  • the image acquisition unit 14 is configured to collect experimental data and transmit it to the image reconstruction unit 15.
  • the image reconstruction unit 15 is configured to reconstruct the object image 16 based on the filtered data.
  • the light source generating unit 11 includes a krypton laser 111 and a reinforced aluminum mirror 112.
  • the HeNe laser 111 emits a laser beam and changes its direction of propagation by enhancing the reflection of the aluminum mirror 112.
  • the filtering unit 12 includes a circular adjustable attenuator 121, a pinhole filter 122, and a Fourier lens 123.
  • the circular adjustable attenuator 121 controls the laser light intensity attenuation, and the attenuated laser filters the high-frequency scattered light through the pinhole filter 122, and allows the unscattered zero-order light to form a zero-order diffraction spot through the pinhole, and then diffracts.
  • the spot passes through the Fourier lens 123 to form parallel light.
  • the image generating unit 13 includes a first polarizing plate 131, a spatial light modulator 132, and a second polarizing plate 133.
  • the experimental light transmitted by the filtering unit 12 illuminates the object image 16, the spatial light modulator 132 is loaded with a specific measurement matrix 17, and the superposition of the object image 16 with the specific measurement matrix 17 is achieved, the first polarizing plate 131 and the second polarizing plate 133. They are placed in a pure amplitude state before and after being placed in the spatial light modulator 132, respectively.
  • the image acquisition unit 14 includes a condenser lens 141 and a single photon detector 142.
  • the condenser lens 141 converges the experimental image data of the image generating unit 13 at one point, and then collects the optical signal by a single photon detector 142 and converts it into an electrical signal.
  • the reconstruction unit 15 includes a computer 151.
  • Computer 151 is used to screen experimental data and reconstruct object image 16 by a minimum total variation optimization algorithm.
  • Step S1 The laser reflection from the light source changes its propagation optical path.
  • a laser beam is emitted by the krypton laser 111; the laser light is reflected by the reinforced aluminum mirror 112 to change its propagation direction.
  • Step S2 attenuating the experimental laser, filtering out the high-frequency scattered light to obtain a zero-order diffracted spot, and adjusting the diffracted spot to be parallel light.
  • a circular adjustable attenuator 121 is provided to control the laser light intensity attenuation; a pinhole filter 122 is disposed behind the circular adjustable attenuator 121; and the high frequency scattering is filtered by the pinhole filter 122. Light is obtained to obtain a zero-order diffracted spot; a Fourier lens 123 is disposed behind the pinhole filter 122, and the diffracted spot passes through the Fourier lens 123 to form parallel light.
  • Step S3 The parallel light is irradiated on the object image 16, and an experimental image in which the object image and the measurement matrix are superimposed is generated.
  • the experimental light transmitted by the filtering unit 12 illuminates the object image 16; the spatial light modulator 132 is loaded with a specific measurement matrix 17, and the transmitted light of the object image 16 is superimposed with the specific measurement matrix 17; the first polarizing plate 131 and the The two polarizing plates 133 are placed in front of the spatial light modulator 132, respectively, in a pure amplitude state.
  • step S3 N image training sample sets X each having a size of p ⁇ q pixels are provided, and each sample is composed of a vector X i of its pixel gray value, and a training sample composed of vectors.
  • Set X ⁇ X 1 , X 2 , ⁇ , X N ⁇ , first calculate the mean of the training sample set:
  • the eigenvalues ⁇ i of the covariance matrix R are arranged in descending order.
  • Principal component matrix U
  • m components extract features that represent the main information of the image. It is post-ranked as a specific measurement matrix 17. Since the specific measurement matrix 17 contains the specific target object image 16 information, it is possible to image only a specific target in a complicated scene while ignoring the background image unrelated to the target image.
  • the specific measurement matrix 17 obtained by training the sample library is expressed as:
  • Step S4 Perform compression sampling on the generated experimental image, and reconstruct an object image according to the data.
  • the output voltage of the high sensitivity photodiode of single photon detector 142 is expressed as:
  • x ⁇ R p ⁇ q , x represents an object image, N ⁇ 1,2, ⁇ ,m ⁇ , Represents the nth dimension of a particular measurement matrix that implements the nth measurement of the object image x;
  • ⁇ R m ⁇ (p ⁇ q) is a specific measurement matrix obtained by the training sample library
  • Y ⁇ R m ⁇ 1 is a measured value
  • Figure 3 is a partial image of a set of training sample libraries.
  • the training sample library is taken by the same object at different angles, a total of 360 sheets, and the size is 800 ⁇ 600 pixels.
  • the specific establishment process is as follows: the training object is placed on a circular rotating table under a black background, and an image is taken at a fixed position by an AVT camera. From 0 degrees to 359 degrees, every 1 degree of rotation is taken in the gray mode, and a total of 360 different angles of training sample images are obtained.
  • the images given in Fig. 3 are images taken at 0 degree, 20 degrees, 40 degrees, ... 180 degrees, respectively.
  • the 4 is an image of a specific measurement matrix obtained by the training sample library; the images generated by the data of the first, the 101st, the 201st, and the 301th rows of the specific measurement matrix are respectively taken, and the number of rows is ranked in front because the principal component analysis method is adopted.
  • the image is richer than the latter and has a higher pixel value.
  • FIG. 5 is a simulation experiment result diagram of an object imaging method based on compressed sensing.
  • the images with background interferers taken by the AVT camera are shown in Figures 5(a) and 5(c), respectively, and each has a size of 800 x 600 pixels.
  • the specific measurement matrix obtained by training is compressed and sampled, and reconstructed by the minimum total variation optimization algorithm.
  • Figure 5(b) shows the reconstruction result of Fig. 5(a) using only 0.075% of the measured data
  • Fig. 5 (d) shows the reconstruction result of Fig. 5(c) using only 0.075% of the measurement data.
  • the experimental results show that the proposed method can remove the interference background in the image and only image the specific object. It is feasible to verify the proposed object based on compressed sensing.
  • the present invention combines the emerging compression sensing theory and pattern recognition technology to simultaneously implement compressed sampling and image recognition in an all-optical environment, greatly reducing the amount of data stored and transmitted, and greatly improving target recognition.
  • the efficiency achieves the requirements of compressed object imaging.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Optics & Photonics (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Geophysics (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Algebra (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Multimedia (AREA)
  • Image Processing (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)
  • Length Measuring Devices By Optical Means (AREA)
  • Image Analysis (AREA)

Abstract

一种基于压缩感知的对象成像系统及其成像方法，包括光源生成单元（11）、滤波单元（12）、图像生成单元（13）、图像采集单元（14）、图像重构单元（15）；光源生成单元（11）生成实验激光，滤波单元（12）滤除高频散射光并形成平行光，图像生成单元（13）生成物体图像（16）与特定测量矩阵（17）叠加的实验图像，图像采集单元（14）对生成的实验图像进行压缩采样，图像重构单元（15）对采样数据进行重构恢复出物体图像（16）；成像方法包括建立包含特定目标物体图像（16）的样本库，训练样本图像得到特定测量矩阵（17），在全光系统中同时完成图像采样、图像压缩、图像识别三个过程。该系统和方法能够大大地减少图像识别、图像匹配中所记录的数据量，提高了系统的实时性，为机器视觉及人工智能的并行处理提供了可能。

Description

一种基于压缩感知的对象成像系统及其成像方法 技术领域
本发明涉及图像成像、目标识别技术领域，特别涉及一种基于压缩感知的对象成像系统及其成像方法。
背景技术
近年来，目标识别将模式识别、计算机视觉、人工智能等学科的技术很好的融合一起，成为视觉研究领域内一个非常热门的课题，具有广阔的应用前景。自目标识别方法提出以来，很多的专家、研究者们都开展了相关工作的研究。但是，由于大部分的目标识别方法都是在获得图像之后再使用识别技术对图像进行相应处理，导致系统的实时性不是很好，而且由此引入的图像数据量大的问题也成为制约其发展的瓶颈。
随着成像和信息技术的迅速发展，人们在日常生活中对电子产品、数字信息对需求也与日俱增。但受信息传输速度与信息处理速度所限，不能不计代价的提高信息的采样率从而获得高品质的数据。在确保信息所呈现出来的品质大体不变的前提下，用少量的采样来获得相近质量的信息，成为了当务之急。
新兴的压缩感知理论是以远低于奈奎斯特采样速率采集信号的非自适应线性投影值，然后通过求解一个优化问题，精确地重构出原始信号,可以大大降低系统采集的数据量。一系列的基于压缩感知的目标识别方案相继被提出，但是这些方案虽然降低了系统采集的数据量，但是在后期进行进一步图像识别处理时仍然面临图像信息繁琐，不利于图像识别，而且这些方案大都采用电学域的数字信号方式实现，无法充分发挥并行处理的优点。
发明内容
本发明的主要目的在于克服现有技术的缺点与不足，提供一种基于压缩感知的对象成像系统，其可在纯光域下同时实现图像压缩、图像采样和图像识别，大大地减少图像识别、图 像匹配中所记录的数据量，解决其海量数据的存贮与传输问题，提高了系统的实时性，为机器视觉及人工智能的并行处理提供了可能。
本发明的另一目的在于提供一种基于上述系统的成像方法。
本发明的目的通过以下的技术方案实现：
一种基于压缩感知的对象成像系统，包括依次连接的光源生成单元、滤波单元、图像生成单元、图像采集单元和图像重构单元；
光源生成单元生成实验激光；
滤波单元控制激光光强衰减、滤除高频散射光、形成零级衍射光斑，然后控制衍射光斑形成平行光；
图像生成单元生成物体图像与特定测量矩阵叠加的实验图像；
图像采集单元对实验图像进行压缩采样；
图像重构单元对采样数据进行重构恢复出物体图像。
优选的，光源生成单元包括激光器和可以反射改变其传播方向的反射镜。
优选的，滤波单元包括与光路平行、依次布置的圆形可调衰减器、针孔滤波器和傅里叶透镜。
优选的，图像生成单元包括与光路平行、依次布置的第一偏振片、加载特定测量矩阵的空间光调制器和第二偏振片。
优选的，图像采集单元包括与光路平行、依次布置的会聚透镜和单光子探测器。
优选的，图像重构单元包括计算机。
一种基于上述系统的成像方法，包括以下步骤：
步骤S1：利用激光光源发出实验激光；
步骤S2：将实验激光进行衰减处理，滤除高频散射光后获得零级衍射光斑，并调整衍射光斑为平行光；
步骤S3：将平行光照射在物体图像上，透过加载特定测量矩阵的空间光调制器生成物体图像和特定测量矩阵叠加的实验图像；
步骤S4：对实验图像进行压缩采样，图像重构单元根据采样数据重构出物体图像。
优选的，步骤S3中特定测量矩阵的计算方法是：建立包含特定目标物体图像的样本库，利用主成分分析法训练样本图像得到特定测量矩阵。
具体的，根据主成分分析方法，设有N张尺寸均为p×q像素的图像训练样本集X，每个样本由其像素灰度值组成一个向量X i，由向量构成的训练样本集X＝{X 1,X 2,···,X i,···,X N}，首先计算训练样本集均值：
Figure PCTCN2018073172-appb-000001
进一步对数据进行中心化：
Figure PCTCN2018073172-appb-000002
对中心化后的数据求协方差矩阵：
Figure PCTCN2018073172-appb-000003
将协方差矩阵R的特征值λ i按照从大到小的顺序排列，把前m个特征值相对应的特征向量u i构成主成分矩阵U：
U＝[u 1,u 2,···,u m]
这m个分量就提取了代表图像主要信息的特征；
将其转秩后作为特定测量矩阵，根据压缩感知原理，经样本库训练得到的特定测量矩阵表示为：
Φ＝U T＝[φ 1,φ 2,···,φ m] T (φ i＝u i)。
优选的，步骤S4中采集实验测量数据：单点采集光信号并转换为电信号，输出的电压值表示为：
Figure PCTCN2018073172-appb-000004
其中，x∈R p×q，x表示物体图像，
Figure PCTCN2018073172-appb-000005
n∈{1,2,···,m}，
Figure PCTCN2018073172-appb-000006
表示特定测量矩阵的第n维，它实现对物体图像x的第n次测量；
重复这个过程m次，可以得到的测量值Y为：
Figure PCTCN2018073172-appb-000007
其中，Φ∈R m×(p×q)是特定测量矩阵，Y∈R m×1是测量值。
优选的，步骤S4中，图像重构单元包括数据筛选处理模块和图像重构处理模块，数据筛选处理模块用于将图像采集单元采集的数据进行时域筛选和大小筛选得到准确的实验数据；图像重构处理模块利用最小全变分优化算法重构出物体图像。
本发明与现有技术相比，具有如下优点和有益效果：
本发明通过机器学习的方法(即主成分方法)训练包含特定物体信息的特定测量矩阵及结合压缩感知技术实现在复杂场景中完成特定目标对象成像。相比于现有技术，本发明结合新兴的压缩感知理论和模式识别技术，在全光域环境中同时实现图像压缩、图像采样和图像识别，即在成像时就实现了目标识别。该技术大大降低了存储和传输的数据量，极大地提高了目标识别的效率，为机器视觉及人工智能的并行处理提供了可能。
附图说明
图1是本实施例中基于压缩感知的对象成像系统的原理框图；
图2是本实施例中基于压缩感知的对象成像系统的结构示意图；
图3是本实施例中建立的一组训练样本库的部分图像；
图4是本实施例中训练样本库获得的特定测量矩阵的图像；
图5是本实施例中基于压缩感知的对象成像方法的仿真实验结果图。
具体实施方式
下面结合实施例及附图对本发明作进一步详细的描述，但本发明的实施方式不限于此。
一种基于压缩感知的对象成像系统，如图1、图2，包括光源生成单元11、滤波单元12、图像生成单元13、图像采集单元14、图像重构单元15。
光源生成单元11，用于生成实验激光。滤波单元12，用于衰减实验激光，滤除高频散射光后获得零级衍射光斑，并调整衍射光斑为平行光。图像生成单元13，用于实现物体图像16与训练后得到的特定测量矩阵17的叠加，生成实验图像。图像采集单元14，用于采集实验数据，并传送至图像重构单元15。图像重构单元15，用于根据筛选的数据重构出物体图像16。
光源生成单元11包括氦氖激光器111和加强铝反射镜112。氦氖激光器111发出一束激光，并通过加强铝反射镜112反射改变其传播方向。
滤波单元12包括圆形可调衰减器121、针孔滤波器122和傅里叶透镜123。圆形可调衰减器121控制激光光强衰减，衰减后的激光经过针孔滤波器122滤除高频散射光，而让未经散射的零级光通过针孔形成零级衍射光斑，然后衍射光斑经过傅里叶透镜123形成平行光。
图像生成单元13包括第一偏振片131、空间光调制器132和第二偏振片133。由滤波单元12传送过来的实验光照射物体图像16，空间光调制器132上加载特定测量矩阵17，并实现物体图像16与特定测量矩阵17的叠加，第一偏振片131和第二偏振片133分别置于空间光调制器132前后使其处于纯振幅状态。
图像采集单元14包括会聚透镜141和单光子探测器142。会聚透镜141将图像生成单元13的实验图像数据会聚于一点，之后通过单光子探测器142单点采集光信号并转换为电信号。
重构单元15包括计算机151。计算机151用于筛选实验数据并通过最小全变分优化算法重构出物体图像16。
以下具体说明该系统的压缩全息成像步骤：
步骤S1：将光源发出的激光反射改变其传播光路。
具体的，通过氦氖激光器111发出一束激光；该激光经过加强铝反射镜112反射改变其传播方向。
步骤S2：将实验激光进行衰减处理，滤除高频散射光后获得零级衍射光斑，并调整衍射光斑为平行光。
具体的，沿着光路路径，设置一圆形可调衰减器121控制激光光强衰减；在圆形可调衰 减器121后方设置针孔滤波器122；通过针孔滤波器122滤除高频散射光，以获得零级衍射光斑；在针孔滤波器122后方设置一傅里叶透镜123，衍射光斑经过傅里叶透镜123形成平行光。
步骤S3：将平行光照射在物体图像16上，生成物体图像与测量矩阵叠加的实验图像。
具体的，由滤波单元12传送过来的实验光照射物体图像16；空间光调制器132上加载特定测量矩阵17，物体图像16的透射光与特定测量矩阵17的叠加；第一偏振片131和第二偏振片133分别置于空间光调制器132前后使其处于纯振幅状态。
在步骤S3中，根据主成分分析方法，设有N张尺寸均为p×q像素的图像训练样本集X，每个样本由其像素灰度值组成一个向量X i，由向量构成的训练样本集X＝{X 1,X 2,···,X N}，首先计算训练样本集均值：
Figure PCTCN2018073172-appb-000008
进一步对数据进行中心化：
Figure PCTCN2018073172-appb-000009
对中心化后的数据求协方差矩阵：
Figure PCTCN2018073172-appb-000010
将协方差矩阵R的特征值λ i按照从大到小的顺序排列，特征值λ i越大代表其表征图像信息的能力越强，因此，把前m个特征值相应的特征向量u i构成主成分矩阵U：
U＝[u 1,u 2,···,u m]
这m个分量就提取了代表图像主要信息的特征。将其转秩后作为特定测量矩阵17。由于该特定测量矩阵17包含特定目标物体图像16信息，从而能够实现在复杂的场景中只对特定目标成像，而忽略与目标图像无关的背景图像。
根据压缩感知原理，经样本库训练得到的特定测量矩阵17表示为：
Φ＝U T＝[φ 1,φ 2,···,φ m] T (φ i＝u i)
步骤S4：对生成的实验图像进行压缩采样，并根据该数据重构出物体图像。
单光子探测器142的高灵敏度的光电二极管的输出电压表示为：
Figure PCTCN2018073172-appb-000011
其中，x∈R p×q，x表示物体图像，
Figure PCTCN2018073172-appb-000012
n∈{1,2,···,m}，
Figure PCTCN2018073172-appb-000013
表示特定测量矩阵的第n维，它实现对物体图像x的第n次测量；
重复这个过程m次，可以得到的测量值Y为：
Figure PCTCN2018073172-appb-000014
其中，Φ∈R m×(p×q)是训练样本库获得的特定测量矩阵，Y∈R m×1是测量值；
利用时域筛选和大小筛选得到准确的实验数据。进一步利用最小全变分优化算法重构出物体图像。
图3是一组训练样本库的部分图像。该训练样本库由同一物体在不同角度下拍摄得到，共360张，尺寸大小均为800×600像素。具体建立过程为：将训练物体放置于黑色背景下的圆形旋转台上，通过AVT相机在固定位置拍摄图像。从0度至359度，每旋转1度就在灰白模式拍摄一次，一共得到360张不同角度的训练样本图像。图3中给出的图像依次为0度、20度、40度......180度角度下拍摄获得的图像。
图4是训练样本库获得的特定测量矩阵的图像；图中分别为取特定测量矩阵的第1、101、201、301行数据生成的图像，由于采用主成分分析方法，故行数排在前面的图像较后面的信息丰富且像素值更高。
图5是基于压缩感知的对象成像方法的仿真实验结果图。通过AVT相机拍摄的带有背景干扰物的的图像分别如图5(a)和5(c)所示，大小均为800×600像素。利用训练得到的特定测量矩阵对其进行压缩采样，并利用最小全变分优化算法重构，图5(b)表示只使用0.075％的测量数据对图5(a)的重构结果，图5(d)表示只使用0.075％的测量数据对图5(c)的重构结果。实验仿真结果证明，该方法能较好的去除图像中的干扰背景，只对特定对象成像，验证了提出的基于压缩感知的对象成像系统是切实可行的。
相比于现有技术，本发明结合新兴的压缩感知理论和模式识别技术，在全光域环境中同时实现压缩采样和图像识别，大大降低了存储和传输的数据量，极大地提高了目标识别的效率，实现了压缩对象成像的要求。
上述实施例为本发明较佳的实施方式，但本发明的实施方式并不受上述实施例的限制，其他的任何未背离本发明的精神实质与原理下所作的改变、修饰、替代、组合、简化，均应为等效的置换方式，都包含在本发明的保护范围之内。

Claims (11)

  1. 一种基于压缩感知的对象成像系统，其特征在于，包括依次连接的光源生成单元、滤波单元、图像生成单元、图像采集单元和图像重构单元；
    光源生成单元生成实验激光；
    滤波单元控制激光光强衰减、滤除高频散射光、形成零级衍射光斑，然后控制衍射光斑形成平行光；
    图像生成单元生成物体图像与特定测量矩阵叠加的实验图像；
    图像采集单元对实验图像进行压缩采样；
    图像重构单元对采样数据进行重构恢复出物体图像。
  2. 根据权利要求1所述的基于压缩感知的对象成像系统，其特征在于，光源生成单元包括激光器和可以反射改变其传播方向的反射镜。
  3. 根据权利要求1所述的基于压缩感知的对象成像系统，其特征在于，滤波单元包括与光路平行、依次布置的圆形可调衰减器、针孔滤波器和傅里叶透镜。
  4. 根据权利要求1所述的基于压缩感知的对象成像系统，其特征在于，图像生成单元包括与光路平行、依次布置的第一偏振片、加载特定测量矩阵的空间光调制器和第二偏振片。
  5. 根据权利要求1所述的基于压缩感知的对象成像系统，其特征在于，图像采集单元包括与光路平行、依次布置的会聚透镜和单光子探测器。
  6. 根据权利要求1所述的基于压缩感知的对象成像系统，其特征在于，图像重构单元包括计算机。
  7. 一种基于权利要求1所述的系统的成像方法，其特征在于，包括以下步骤：
    步骤S1：利用激光光源发出实验激光；
    步骤S2：将实验激光进行衰减处理，滤除高频散射光后获得零级衍射光斑，并调整衍射光斑为平行光；
    步骤S3：将平行光照射在物体图像上，透过加载特定测量矩阵的空间光调制器生成物体图像和特定测量矩阵叠加的实验图像；
    步骤S4：对实验图像进行压缩采样，图像重构单元根据采样数据重构出物体图像。
  8. 根据权利要求7所述的成像方法，其特征在于，步骤S3中特定测量矩阵的计算方法是：建立包含特定目标物体图像的样本库，利用主成分分析法训练样本图像得到特定测量矩阵。
  9. 根据权利要求8所述的成像方法，其特征在于，根据主成分分析方法，设有N张尺寸均为p×q像素的图像训练样本集X，每个样本由其像素灰度值组成一个向量X i，由向量构成的训练样本集X＝{X 1,X 2,···,X i,···,X N}，首先计算训练样本集均值：
    Figure PCTCN2018073172-appb-100001
    进一步对数据进行中心化：
    Figure PCTCN2018073172-appb-100002
    对中心化后的数据求协方差矩阵：
    Figure PCTCN2018073172-appb-100003
    将协方差矩阵R的特征值λ i按照从大到小的顺序排列，把前m个特征值相对应的特征向量u i构成主成分矩阵U：
    U＝[u 1,u 2,···,u m]
    这m个分量就提取了代表图像主要信息的特征；
    将其转秩后作为特定测量矩阵，根据压缩感知原理，经样本库训练得到的特定测量矩阵表示为：
    Φ＝U T＝[φ 1,φ 2,···,φ m] T(φ i＝u i)。
  10. 根据权利要求7所述的成像方法，其特征在于，步骤S4中采集实验测量数据：单点采集光信号并转换为电信号，输出的电压值表示为：
    Figure PCTCN2018073172-appb-100004
    其中，x∈R p×q，x表示物体图像，
    Figure PCTCN2018073172-appb-100005
    n∈{1,2,···,m}，
    Figure PCTCN2018073172-appb-100006
    表示特定测量矩阵的第n维，它实现对物体图像x的第n次测量；
    重复这个过程m次，可以得到的测量值Y为：
    Figure PCTCN2018073172-appb-100007
    其中，Φ∈R m×(p×q)是特定测量矩阵，Y∈R m×1是测量值。
  11. 根据权利要求7所述的成像方法，其特征在于，步骤S4中，图像重构单元包括数据筛选处理模块和图像重构处理模块，数据筛选处理模块用于将图像采集单元采集的数据进行时域筛选和大小筛选得到准确的实验数据；图像重构处理模块利用最小全变分优化算法重构出物体图像。
PCT/CN2018/073172 2017-06-01 2018-01-18 一种基于压缩感知的对象成像系统及其成像方法 WO2018218974A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/618,376 US11368608B2 (en) 2017-06-01 2018-01-18 Compressed sensing based object imaging system and imaging method therefor

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710402262.4 2017-06-01
CN201710402262.4A CN107121709B (zh) 2017-06-01 2017-06-01 一种基于压缩感知的对象成像系统及其成像方法

Publications (1)

Publication Number Publication Date
WO2018218974A1 true WO2018218974A1 (zh) 2018-12-06

Family

ID=59729166

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/073172 WO2018218974A1 (zh) 2017-06-01 2018-01-18 一种基于压缩感知的对象成像系统及其成像方法

Country Status (3)

Country Link
US (1) US11368608B2 (zh)
CN (1) CN107121709B (zh)
WO (1) WO2018218974A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111337973A (zh) * 2018-12-19 2020-06-26 中国石油天然气集团有限公司 地震数据重建方法、系统
CN111640069A (zh) * 2020-04-17 2020-09-08 上海交通大学 基于光感知网络和相位补偿的压缩成像方法、系统和装置
CN115442505A (zh) * 2022-08-30 2022-12-06 山西大学 一种单光子压缩感知成像系统及其方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107121709B (zh) * 2017-06-01 2023-07-25 华南师范大学 一种基于压缩感知的对象成像系统及其成像方法
CN110646810B (zh) * 2019-09-27 2021-08-06 北京理工大学 一种散斑优化压缩感知鬼成像方法及系统
CN111028302B (zh) * 2019-11-27 2023-07-25 华南师范大学 一种基于深度学习的压缩对象成像方法及系统
CN113240610B (zh) * 2021-05-27 2023-05-12 清华大学深圳国际研究生院 一种基于仿人眼机制的双通道鬼成像重建方法及系统
CN113890997B (zh) * 2021-10-19 2023-06-13 中国科学院国家空间科学中心 基于随机抖动的高动态范围压缩感知成像系统及方法
CN116708086A (zh) * 2022-02-25 2023-09-05 维沃移动通信有限公司 感知方法、装置及通信设备

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7616320B2 (en) * 2006-03-15 2009-11-10 Bahram Javidi Method and apparatus for recognition of microorganisms using holographic microscopy
CN102389321A (zh) * 2011-06-23 2012-03-28 深圳市开立科技有限公司 一种快速光声三维成像装置
CN102393911A (zh) * 2011-07-21 2012-03-28 西安电子科技大学 基于压缩感知的背景杂波量化方法
CN104021522A (zh) * 2014-04-28 2014-09-03 中国科学院上海光学精密机械研究所 基于强度关联成像的目标图像分离装置及分离方法
CN204360096U (zh) * 2014-12-10 2015-05-27 华南师范大学 基于压缩传感理论的数字全息成像装置
CN105451024A (zh) * 2015-12-31 2016-03-30 北京大学 一种采用压缩感知的数字全息图编码传输方法
CN107121709A (zh) * 2017-06-01 2017-09-01 华南师范大学 一种基于压缩感知的对象成像系统及其成像方法
CN206930789U (zh) * 2017-06-01 2018-01-26 华南师范大学 一种基于压缩感知的对象成像系统

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2318875B1 (en) * 2008-07-25 2019-11-06 Sloan Kettering Institute For Cancer Research Rapid confocal microscopy to support surgical procedures
US8471750B2 (en) * 2010-09-29 2013-06-25 The Johns Hopkins University System and method for compressive sensing
KR101729976B1 (ko) * 2011-06-09 2017-04-25 한국전자통신연구원 영상 인식 장치 및 그것의 영상 인식 방법
CN104054266B (zh) * 2011-10-25 2016-11-23 中国科学院空间科学与应用研究中心 一种时间分辨单光子或极弱光多维成像光谱系统及方法
CN102722866B (zh) * 2012-05-22 2014-07-23 西安电子科技大学 基于主成份分析的压缩感知方法
EP3210372B1 (en) * 2014-10-21 2023-07-19 University College Cork-National University of Ireland, Cork Smart photonic imaging method and apparatus
US11300449B2 (en) * 2015-03-24 2022-04-12 University Of Utah Research Foundation Imaging device with image dispersing to create a spatially coded image
US20160313548A1 (en) * 2015-04-21 2016-10-27 Olympus Corporation Method for capturing image of three-dimensional structure of specimen and microscopic device
CN106331442B (zh) * 2015-07-02 2021-01-15 松下知识产权经营株式会社 摄像装置
JP6814983B2 (ja) * 2016-03-31 2021-01-20 パナソニックＩｐマネジメント株式会社 撮像装置
CN106203374B (zh) * 2016-07-18 2018-08-24 清华大学深圳研究生院 一种基于压缩感知的特征识别方法及其系统

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7616320B2 (en) * 2006-03-15 2009-11-10 Bahram Javidi Method and apparatus for recognition of microorganisms using holographic microscopy
CN102389321A (zh) * 2011-06-23 2012-03-28 深圳市开立科技有限公司 一种快速光声三维成像装置
CN102393911A (zh) * 2011-07-21 2012-03-28 西安电子科技大学 基于压缩感知的背景杂波量化方法
CN104021522A (zh) * 2014-04-28 2014-09-03 中国科学院上海光学精密机械研究所 基于强度关联成像的目标图像分离装置及分离方法
CN204360096U (zh) * 2014-12-10 2015-05-27 华南师范大学 基于压缩传感理论的数字全息成像装置
CN105451024A (zh) * 2015-12-31 2016-03-30 北京大学 一种采用压缩感知的数字全息图编码传输方法
CN107121709A (zh) * 2017-06-01 2017-09-01 华南师范大学 一种基于压缩感知的对象成像系统及其成像方法
CN206930789U (zh) * 2017-06-01 2018-01-26 华南师范大学 一种基于压缩感知的对象成像系统

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111337973A (zh) * 2018-12-19 2020-06-26 中国石油天然气集团有限公司 地震数据重建方法、系统
CN111337973B (zh) * 2018-12-19 2022-08-05 中国石油天然气集团有限公司 地震数据重建方法、系统
CN111640069A (zh) * 2020-04-17 2020-09-08 上海交通大学 基于光感知网络和相位补偿的压缩成像方法、系统和装置
CN111640069B (zh) * 2020-04-17 2022-11-11 上海交通大学 基于光感知网络和相位补偿的压缩成像方法、系统和装置
CN115442505A (zh) * 2022-08-30 2022-12-06 山西大学 一种单光子压缩感知成像系统及其方法
CN115442505B (zh) * 2022-08-30 2023-07-21 山西大学 一种单光子压缩感知成像系统及其方法

Also Published As

Publication number Publication date
US20210144278A1 (en) 2021-05-13
CN107121709B (zh) 2023-07-25
US11368608B2 (en) 2022-06-21
CN107121709A (zh) 2017-09-01

Similar Documents

Publication Publication Date Title
WO2018218974A1 (zh) 一种基于压缩感知的对象成像系统及其成像方法
CN100402996C (zh) 确定波场相位、物体成象、相位振幅成象的方法和装置
CN110455834B (zh) 基于光强传输方程的x射线单次曝光成像装置及方法
CN105467806B (zh) 单像素全息相机
CN110927115B (zh) 基于深度学习的无透镜双型融合目标检测装置及方法
CN108508588B (zh) 一种多约束信息的无透镜全息显微相位恢复方法及其装置
CN108469685A (zh) 一种超分辨率关联成像系统及成像方法
Li et al. Quantitative phase imaging (QPI) through random diffusers using a diffractive optical network
Wang et al. Zero-order term suppression in off-axis holography based on deep learning method
CN113298700A (zh) 一种在散射场景中的高分辨图像重构方法
Fructuoso et al. Photoelastic analysis of partially occluded objects with an integral-imaging polariscope
CN206930789U (zh) 一种基于压缩感知的对象成像系统
Madsen et al. On-axis digital holographic microscopy: Current trends and algorithms
Zhang et al. Deep-learning-based halo-free white-light diffraction phase imaging
WO2022132496A1 (en) Totagraphy: coherent diffractive/digital information reconstruction by iterative phase recovery using special masks
Li et al. Fourier ptychography reconstruction based on reweighted amplitude flow with regularization by denoising and deep decoder
WO2020186394A1 (zh) 成像方法及装置
Zhu et al. Digital holography with polarization multiplexing for underwater imaging and descattering
TWI804349B (zh) 刀具影像檢測方法
Proppe et al. 3d-2d neural nets for phase retrieval in noisy interferometric imaging
Hu et al. Hybrid method for accurate phase retrieval based on higher order transport of intensity equation and multiplane iteration
CN111612884B (zh) 一种透射式无透镜三维显微重构方法及系统
US20230410256A1 (en) System and method for improving image resolution of 3-d refractive index microscope based on ai technology
CN111861907B (zh) 一种高动态范围激光焦斑图像的去噪方法
Henrot et al. Fast deconvolution of large fluorescence hyperspectral images

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18808782

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18808782

Country of ref document: EP

Kind code of ref document: A1