CN110136187B - A method for reducing the computational cost of correlated imaging based on compressive sensing observation matrix segmentation - Google Patents
A method for reducing the computational cost of correlated imaging based on compressive sensing observation matrix segmentation Download PDFInfo
- Publication number
- CN110136187B CN110136187B CN201910414311.5A CN201910414311A CN110136187B CN 110136187 B CN110136187 B CN 110136187B CN 201910414311 A CN201910414311 A CN 201910414311A CN 110136187 B CN110136187 B CN 110136187B
- Authority
- CN
- China
- Prior art keywords
- matrix
- light field
- reading
- field intensity
- file
- 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.)
- Expired - Fee Related
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
- G06T7/55—Depth or shape recovery from multiple images
- G06T7/557—Depth or shape recovery from multiple images from light fields, e.g. from plenoptic cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10052—Images from lightfield camera
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
本发明公开了一种基于压缩感知观测矩阵分割减小关联成像计算开销的方法，包括：将实验获得的关联成像测量数据利用压缩感知算法进行图像重构；针对重构成像需要大量的数据，导致进行压缩感知运算时出现的内存溢出问题，尝试将压缩感知进行优化并解决问题。实验结果表明：该方法可以有效的减少压缩感知重构算法中的内存溢出问题，并且能够成功的重构出所需的图像。
The invention discloses a method for reducing the computational cost of correlation imaging based on compressive sensing observation matrix segmentation. The memory overflow problem that occurs when the compressed sensing operation is performed, try to optimize the compressed sensing and solve the problem. The experimental results show that the method can effectively reduce the memory overflow problem in the compressive sensing reconstruction algorithm, and can successfully reconstruct the desired image.
Description
Technical Field
The invention belongs to the field of correlation imaging based on compressed sensing, and particularly relates to a method for optimizing a memory by compressed sensing.
Background
The correlation imaging is a novel imaging mode which is started in recent years, and the correlation reconstruction of the target to be measured is realized by performing coincidence measurement through two or more detectors based on second-order or even higher-order correlation information of a light field. The correlation imaging technique has attracted the attention of a large number of researchers due to its special non-localized imaging properties and its excellent anti-noise performance. The related imaging device has a simple structure and certain super-resolution capability, so the related imaging device has a very wide application prospect.
However, in the related imaging, in order to obtain a reconstructed image with higher quality, a large amount of experimental data must be acquired, and the time is long. This results in a large amount of post-processing computation, which is a great challenge for the development of the related imaging.
Compressed sensing, which comes into the line of sight of researchers, can alleviate this problem to some extent. The compressed sensing makes up for the defect that the original associated imaging needs a large amount of data acquisition in the signal acquisition and processing technology. Compressed sensing can restore the original signal very quickly and accurately only by the sampling data amount which is less than the conventional Nyquist sampling processing data amount, and the information processing speed is improved. Nevertheless, the application of compressed sensing also faces a problem of large memory requirement when processing a large amount of associated imaging data, which is also a problem to be solved.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide an optimization method for the problem that compressed sensing needs a large amount of memory during correlated imaging reconstruction.
The technical scheme provided by the invention is as follows:
an optimization method for compressed sensing application in correlated imaging comprises the following steps:
performing longitudinal column selection on the compressed sensing matrix A according to requirements, and performing compressed sensing image reconstruction;
and performing horizontal row selection on the compressed sensing matrix A according to requirements, and performing compressed sensing image reconstruction.
Performing a correlation imaging experiment to obtain each frame of reference arm light field required by image reconstruction and obtain corresponding bucket data;
reading a reference arm light field of the first frame to obtain an M x M light field intensity matrix;
rotating the light field intensity matrix by 90 degrees clockwise to obtain a rotated light field intensity matrix;
preferably, the results obtained will be column-selected in the longitudinal direction of the a-matrix. The method specifically comprises the following steps:
changing the rotated light field intensity matrix into a one-dimensional array containing M × N elements through reshape operation;
intercepting continuous P x N elements (continuous P rows of the rotated light intensity matrix) in the one-dimensional array according to requirements;
carrying out reshape operation on the intercepted one-dimensional array, and recombining the one-dimensional array into a light field intensity matrix of P x N;
traversing the obtained light field intensity matrix and writing the light field intensity matrix into a new file;
the same operation is carried out on each frame of reference arm light field matrix and the light field matrix is written into a corresponding new file according to the sequence number;
preferably, after column selection in the compressed sensing a matrix is performed longitudinally, compressed sensing image reconstruction is performed, and how to construct the a matrix specifically includes:
reading the processed light field intensity matrix file frame by frame;
reading the first frame of light field to obtain a light field matrix of P x N, and performing reshape operation to change the light field matrix into a one-dimensional array containing P x N elements;
constructing an empty A matrix, and adding the one-dimensional array after reshape into the A matrix as a first row;
and continuously reading the light field intensity matrix file of the second frame, and taking the file as a second row of the A matrix after reshape is carried out. The above operation is performed until all the light field intensity matrix files are read and processed.
Preferably, the compressed sensing image reconstruction further comprises:
reading the bucket data measured by the experiment, carrying out normalization processing on the bucket data, and carrying out normalization processing on each row of the matrix A;
and performing compressed sensing image reconstruction on the A matrix and the bucket data after the normalization processing to obtain an expected reconstructed image.
Preferably, the compressed sensing a matrix is selected horizontally, and before performing horizontal row selection, the method further includes:
reading a bucket data file, and performing Gaussian fitting on the bucket data file to obtain a mean E and a variance sigma of the bucket data file;
and subtracting n times of sigma from the mean value E obtained by Gaussian fitting to obtain an upper limit a, and adding n times of sigma to the mean value E to obtain a lower limit b.
Preferably, the a matrix construction and the image reconstruction are characterized in that the processing of the light field intensity matrix file comprises:
reading a first frame light field intensity matrix file, and judging whether a corresponding bucket numerical value is smaller than a or larger than b;
if the corresponding bucket data does not meet the requirements, discarding and reading the next frame of file;
if the corresponding bucket data meet the requirements, clockwise rotating the read M x N light field matrix by 90 degrees, and forming reshape of the light field matrix into a one-dimensional array containing M x N elements;
taking the one-dimensional array as the first row of the A matrix, and taking the corresponding bucket value as the first element of the one-dimensional array bucket 0;
reading a second frame of light field intensity matrix file, carrying out the same judgment, abandoning if the requirement is not met, taking reshape of the second frame as a second row of the A matrix if the requirement is met, and taking a corresponding bucket numerical value as a second element of the bucket 0;
the above operations are repeated until all the light field intensity matrix files are read and processed.
Reading the reconstructed one-dimensional array bucket0, performing normalization processing on the reconstructed one-dimensional array bucket0, and performing normalization processing on each row of the matrix A;
and performing compressed sensing image reconstruction on the A matrix and the bucket0 data after the normalization processing to obtain an expected reconstructed image.
The invention at least comprises the following beneficial effects: after the associated imaging experiment data is acquired, an attempt is made to reconstruct the image using a compressed sensing method. When the method is realized by using the longitudinal column selection method of the matrix A, a part of an ideal image can be successfully reconstructed according to the selected size, and the data volume read in by a computer can be effectively reduced, so that the requirement of compressed sensing on the memory of the computer is reduced. When the transverse row selection method of the matrix A is used for realizing the transverse row selection, a partial light field intensity matrix which has the largest contribution to image information can be obtained, and an image is successfully reconstructed, so that the requirement of compressed sensing on a computer memory is reduced.
Drawings
FIG. 1 is a schematic flow chart of an optimization method for compressive sensing application in correlation imaging according to the present invention;
FIG. 2 is a schematic diagram of a vertical column selection process of a compressed sensing A matrix according to the present invention;
FIG. 3 illustrates the experimental data image reconstruction process after vertical column selection;
FIG. 4 illustrates the preprocessing step of the packet data before horizontal line selection;
FIG. 5 illustrates a horizontal row selection and image reconstruction process;
FIG. 6 is a reconstructed graph of experiments with different values of n, wherein n is 1.28, 1.07, 0.71, 0.5, 0.35, 0.21 and 0.1;
FIG. 7. memory consumption required for row selection of matrix A;
FIG. 8 shows image reconstruction results for different selected columns;
FIG. 9. memory consumption required for A matrix column selection;
FIG. 10: reconstructing the signal-to-noise ratio of the corresponding selected line image under different values of n;
FIG. 11: image reconstruction signal-to-noise ratio for different column selection operations
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the advantages of the technical solutions of the present invention clearer, the present invention is described in detail below with reference to the accompanying drawings and examples.
As shown in fig. 1, the method for optimizing a compressive sensing application in correlated imaging according to an embodiment of the present invention includes the following steps:
s1, taking the associated imaging experiment data as input, performing longitudinal column selection of the matrix A, and writing a new file;
as shown in fig. 2, the specific process of step S1 is as follows:
s11, firstly, reading a frame of light field of a correlation imaging experiment reference arm to obtain an M x M light field intensity matrix;
s12, clockwise rotating the light field intensity matrix by 90 degrees to obtain a rotated light field intensity matrix;
s13, changing the rotated light field intensity matrix into a one-dimensional array containing M × N elements through reshape operation;
s14, intercepting continuous P x N elements (continuous P rows of the rotated light intensity matrix) in the one-dimensional array according to requirements;
s15, carrying out reshape operation on the intercepted one-dimensional array, and recombining the one-dimensional array into a light field intensity matrix of P × N;
s16, traversing the obtained light field intensity matrix and writing the light field intensity matrix into a new file;
s17, performing the same operation on each frame of reference arm light field matrix and writing the same operation into a corresponding new file according to the sequence number;
s2, reading the new file to perform compressed sensing image reconstruction;
as shown in fig. 3, the specific process of step S2 is as follows:
s21, constructing an empty A matrix;
s22, reading the processed light field intensity matrix file;
s23, reading the first frame of light field to obtain a light field matrix of P x N, and performing reshape operation to change the light field matrix into a one-dimensional array containing P x N elements;
s24, adding the one-dimensional array after reshape into the matrix A as a current tail row;
s25, judging whether the reading of the file is finished or not, if not, reading the file of the next frame and repeating the above operations, and if so, performing the next step;
s26, reading the bucket data and carrying out normalization processing on the bucket data;
s27, normalizing each line of the A matrix;
and S28, carrying out compressed sensing image reconstruction on the A matrix and the bucket data after the normalization processing, and obtaining an expected reconstructed image.
S3, performing bucket data preprocessing operation before the matrix A is transversely selected as follows;
as shown in fig. 5, the specific method of step S3 includes:
s31, reading the bucket data measured by the experiment;
s32, performing Gaussian fitting on the bucket data to obtain a mean value E and a variance sigma of the bucket data;
and S33, subtracting n times of sigma from the mean value E obtained by Gaussian fitting to obtain an upper limit a, and adding n times of sigma to the mean value E to obtain a lower limit b.
And S4, constructing an A matrix and reconstructing an image, wherein the processing of the light field intensity matrix file comprises the following steps.
As shown in fig. 6, the step S4 of determining the distortion target includes the following steps:
s41, reading a frame of light field intensity matrix file, and judging whether a bucket value corresponding to the file is smaller than a or larger than b;
s42, if the corresponding bucket data do not meet the requirements, discarding and reading the next frame of file;
s43, if the corresponding bucket data meet the requirements, clockwise rotating the read M x N light field matrix by 90 degrees, and reshape the light field matrix into a one-dimensional array containing M x N elements;
s44, taking the one-dimensional array as the first row of the matrix A, and taking the corresponding bucket numerical value as the first element of the one-dimensional array bucket 0;
and S45, reading the light field intensity matrix file of the second frame, carrying out the same judgment, abandoning if the light field intensity matrix file does not meet the requirement, taking reshape of the second frame as the current tail row of the A matrix if the light field intensity matrix file meets the requirement, and taking the corresponding bucket numerical value as the second element of the bucket 0.
And S46, repeating the operation until all the light field intensity matrix files are read and processed.
S47, reading the reconstructed one-dimensional array bucket0, carrying out normalization processing on the reconstructed one-dimensional array bucket, and carrying out normalization processing on each row of the matrix A;
and S48, carrying out compressed sensing image reconstruction on the A matrix and the bucket0 data after normalization processing, and obtaining an expected reconstructed image.
In order to verify the wide applicability of the embodiment of the present invention, two sets of experiments are prepared in the embodiment of the present invention, which respectively correspond to the horizontal row selection image reconstruction of the a matrix and the vertical column selection image reconstruction of the a matrix.
The horizontal row-selection image reconstruction of the a-matrix is based on experimental data provided by a correlation imaging experiment using the capital GI as the target object. And performing horizontal row selection operation on the experimental data of the A matrix, and then performing image reconstruction, wherein the image reconstructed by the experiment is a complete GI letter, and the image reconstructed by the experiment is a first group of experiments. The longitudinally-selected column image reconstruction of the A matrix is to perform the longitudinally-selected column operation of the A matrix according to the same data provided by the above experiment and then perform image reconstruction, and the experimentally reconstructed image should be a part of GI letters.
Therefore, the specific contents of the embodiment of the invention in the actual experiment are as follows:
(1) a matrix horizontal row selection image reconstruction experiment
And inputting 5000 frames of light field intensity files of associated imaging experimental data, and calculating the mean E and the variance sigma of the bucket. And then, calculating the selected upper limit and the selected lower limit of the bucket data. Let the upper limit a equal E-n σ and the lower limit b equal E + n σ. And transforming the value of n to select the rows of the matrix A. In this experiment, n is equal to 1.28, 1.07, 0.71, 0.5, 0.35, 0.21 and 0.1 respectively, seven groups of experiments are carried out, the number of frames actually used for reconstruction is 617, 998, 1951, 2905, 3410, 4072 and 4503 respectively, and the experimental results are shown in the following fig. 6. The computer memory and time required to perform the calculations were examined during the experiment as shown in fig. 7 below. The experimental result graph was subjected to SNR analysis with reference to the graph, and the result is shown in fig. 10.
(2) A matrix longitudinal column selection image reconstruction experiment
Firstly, 5000 frames of light field intensity files of related imaging experimental data are input, column selection operation of the matrix A is carried out, and the front 1/2, 1/3, 1/4, 1/5/, 1/6 groups of data of the matrix A are selected respectively. And after the column selection operation is finished, respectively storing the data into new folders. Then, the compressed sensing image reconstruction experiments are performed, and the experimental results are shown in fig. 8 below. The computer memory required for the calculation was examined during the experiment as shown in fig. 9 below. The experimental results were concatenated and then analyzed for SNR with the reference map, the results are shown in fig. 11.
According to the experimental results, the compressed sensing matrix A column selection operation provided by the embodiment of the invention can well reduce the consumed computer memory and successfully reconstruct the expected image when a large amount of associated imaging data is utilized to reconstruct the image. The image can be reconstructed by utilizing data which contributes more to the image well when the compressed sensing A matrix column selection operation is carried out. The experimental data and the operation time required by operation can be effectively reduced.
While embodiments of the invention have been described above, it is not limited to the applications set forth in the description and the embodiments, which are fully applicable in various fields of endeavor to which the invention pertains, and further modifications may readily be made by those skilled in the art, it being understood that the invention is not limited to the details shown and described herein without departing from the general concept defined by the appended claims and their equivalents.
Claims (3)
1. A method for reducing associated imaging computation overhead based on compressed sensing observation matrix segmentation is characterized by comprising the following steps:
s1, taking the associated imaging experiment data as input, performing compressed sensing A matrix longitudinal column selection, and writing in a new file;
s11, firstly, reading a frame of light field of a correlation imaging experiment reference arm to obtain an M x M light field intensity matrix;
s12, clockwise rotating the light field intensity matrix by 90 degrees to obtain a rotated light field intensity matrix;
s13, changing the rotated light field intensity matrix into a one-dimensional array containing M × N elements through reshape operation;
s14, intercepting continuous P x N elements in the one-dimensional array according to requirements, wherein P is continuous P rows of the rotated light intensity matrix;
s15, carrying out reshape operation on the intercepted one-dimensional array, and recombining the one-dimensional array into a light field intensity matrix of P × N;
s16, traversing the obtained light field intensity matrix and writing the light field intensity matrix into a new file;
s17, performing the same operation on each frame of reference arm light field matrix and writing the same operation into a corresponding new file according to the sequence number;
wherein, before the column selection of the compressed sensing A matrix in the longitudinal direction, the method further comprises the following steps:
performing a correlation imaging experiment to obtain each frame of reference arm light field required by image reconstruction and obtain corresponding bucket data;
reading a reference arm light field of the first frame to obtain an M x N light field intensity matrix;
rotating the light field intensity matrix by 90 degrees clockwise to obtain a rotated light field intensity matrix;
s2, reading the new file to perform compressed sensing image reconstruction;
s21, constructing an empty A matrix;
s22, reading the processed light field intensity matrix file;
s23, reading the first frame of light field to obtain a light field matrix of P x N, and performing reshape operation to change the light field matrix into a one-dimensional array containing P x N elements;
s24, adding the one-dimensional array after reshape into the matrix A as a current tail row;
s25, judging whether the reading of the file is finished or not, if not, reading the file of the next frame and repeating the reading operation, and if so, performing the next step;
s26, reading the bucket data and carrying out normalization processing on the bucket data;
s27, normalizing each line of the A matrix;
and S28, carrying out compressed sensing image reconstruction on the A matrix and the bucket data after the normalization processing, and obtaining an expected reconstructed image.
2. The method for reducing associated imaging computational overhead based on compressed sensing observation matrix segmentation as claimed in claim 1, wherein reading the bucket data and performing a normalization process on the bucket data before further comprising preprocessing the bucket data, and the specific operations are as follows:
reading a bucket data file, and performing Gaussian fitting on the bucket data file to obtain a mean E and a variance sigma of the bucket data file;
and subtracting n times of sigma from the mean value E obtained by Gaussian fitting to obtain an upper limit a, and adding n times of sigma to the mean value E to obtain a lower limit b.
3. The method for reducing associated imaging computational overhead based on compressed sensing observation matrix segmentation as claimed in claim 1, wherein the step 2 further comprises the following steps:
constructing an empty A matrix; reading a first frame light field intensity matrix file, and judging whether a corresponding bucket numerical value is smaller than a or larger than b; if the corresponding bucket numerical value does not meet the requirement of being smaller than a or larger than b, the corresponding bucket data does not meet the requirement, and the next frame of file is abandoned and read;
if the corresponding bucket numerical value is smaller than a or larger than b, the corresponding bucket data meets the requirement, the read P x N light field matrix is clockwise rotated by 90 degrees, and reshape of the light field matrix is formed into a one-dimensional array containing P x N elements;
taking the one-dimensional array as the first row of the A matrix, and taking the corresponding bucket value as the first element of the one-dimensional array bucket 0;
reading a second frame of light field intensity matrix file, carrying out the same judgment, abandoning if the requirement is not met, taking reshape of the second frame as a second row of the A matrix if the requirement is met, and taking a corresponding bucket numerical value as a second element of the bucket 0;
repeating the above operations until all the light field intensity matrix files are read and processed;
reading the reconstructed one-dimensional array bucket0, performing normalization processing on the reconstructed one-dimensional array bucket0, and performing normalization processing on each row of the matrix A;
and performing compressed sensing operation on the A matrix and the bucket0 data after the normalization processing to obtain an expected reconstructed image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910414311.5A CN110136187B (en) | 2019-05-17 | 2019-05-17 | A method for reducing the computational cost of correlated imaging based on compressive sensing observation matrix segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910414311.5A CN110136187B (en) | 2019-05-17 | 2019-05-17 | A method for reducing the computational cost of correlated imaging based on compressive sensing observation matrix segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110136187A CN110136187A (en) | 2019-08-16 |
| CN110136187B true CN110136187B (en) | 2021-09-24 |
Family
ID=67575093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910414311.5A Expired - Fee Related CN110136187B (en) | 2019-05-17 | 2019-05-17 | A method for reducing the computational cost of correlated imaging based on compressive sensing observation matrix segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110136187B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111193925B (en) * | 2019-12-25 | 2021-12-07 | 杭州中威电子股份有限公司 | Image compressed sensing coding and normalization method based on block vector inner product |
| CN111551955B (en) * | 2020-06-22 | 2022-07-08 | 北京理工大学 | A bionic block ghost imaging method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1484905A3 (en) * | 2003-06-06 | 2009-08-12 | GE Medical Systems Global Technology Company LLC | Image processing method and apparatus |
| CN104159048A (en) * | 2014-07-24 | 2014-11-19 | 南京邮电大学 | Compressive sensing uniform weighting relevance imaging method for non-uniform light field |
| CN106203374A (en) * | 2016-07-18 | 2016-12-07 | 清华大学深圳研究生院 | A kind of characteristic recognition method based on compressed sensing and system thereof |
| CN107622515A (en) * | 2017-09-06 | 2018-01-23 | 郑州大学 | Physical reconstruction method of compressed light field |
| CN109035282A (en) * | 2018-08-02 | 2018-12-18 | 吉林工程技术师范学院 | The thresholding method of Hadamard coded modulation relevance imaging |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10546433B2 (en) * | 2017-08-03 | 2020-01-28 | The University Of North Carolina At Chapel Hill | Methods, systems, and computer readable media for modeling garments using single view images |
-
2019
- 2019-05-17 CN CN201910414311.5A patent/CN110136187B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1484905A3 (en) * | 2003-06-06 | 2009-08-12 | GE Medical Systems Global Technology Company LLC | Image processing method and apparatus |
| CN104159048A (en) * | 2014-07-24 | 2014-11-19 | 南京邮电大学 | Compressive sensing uniform weighting relevance imaging method for non-uniform light field |
| CN106203374A (en) * | 2016-07-18 | 2016-12-07 | 清华大学深圳研究生院 | A kind of characteristic recognition method based on compressed sensing and system thereof |
| CN107622515A (en) * | 2017-09-06 | 2018-01-23 | 郑州大学 | Physical reconstruction method of compressed light field |
| CN109035282A (en) * | 2018-08-02 | 2018-12-18 | 吉林工程技术师范学院 | The thresholding method of Hadamard coded modulation relevance imaging |
Non-Patent Citations (3)
| Title |
|---|
| "Orthonormalization method in ghost imaging";Luo Bin.et al;《optics express》;20180903;第26卷(第18期);全文 * |
| "压缩感知及其图像处理应用研究进展与展望";任越美等;《自动化学报》;20140831;第40卷(第8期);全文 * |
| "压缩感知重构算法在"鬼"成像中的应用研究";董小亮灯;《信号处理》;20130630;第29卷(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110136187A (en) | 2019-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111260586B (en) | Correction method and device for distorted document image | |
| CN109784372B (en) | Target classification method based on convolutional neural network | |
| CN110222784B (en) | Solar cell defect detection method integrating short-term and long-term depth features | |
| CN110136187B (en) | A method for reducing the computational cost of correlated imaging based on compressive sensing observation matrix segmentation | |
| Su et al. | High-similarity-pass attention for single image super-resolution | |
| CN113689517A (en) | Image texture synthesis method and system of multi-scale channel attention network | |
| CN114092467B (en) | A scratch detection method and system based on lightweight convolutional neural network | |
| CN115115921A (en) | Wafer defect mode identification method based on residual error network | |
| CN113112447A (en) | Tunnel surrounding rock grade intelligent determination method based on VGG convolutional neural network | |
| Song et al. | ESRGAN-DP: Enhanced super-resolution generative adversarial network with adaptive dual perceptual loss | |
| CN112232102B (en) | Building target recognition method and system based on deep neural network and multi-task learning | |
| CN104992407A (en) | Image super-resolution method | |
| CN114639102A (en) | Cell segmentation method and device based on key point and size regression | |
| CN104732491B (en) | A kind of single frames Remote Sensing Image Super Resolution processing method of edge prior guiding | |
| CN114049939A (en) | Pneumonia CT image generation method based on UNet-GAN network | |
| CN118711207A (en) | A texture segmentation and enhancement method and system for seismic profile images | |
| CN116977473B (en) | A sparse angle CT reconstruction method and device based on projection domain and image domain | |
| Li et al. | Exploring frequency-inspired optimization in transformer for efficient single image super-resolution | |
| CN111833264A (en) | A method and system for fast image denoising based on CUDA | |
| Luo et al. | Learning re-sampling methods with parameter attribution for image super-resolution | |
| CN119251842B (en) | Improved semantic segmentation network construction method and system for identifying bridge road cracks | |
| Liu | Lightweight single image super-resolution by channel split residual convolution | |
| He et al. | Canny edge detection on a virtual hexagonal image structure | |
| Meng et al. | FPPN: fast pixel purification network for single-image super-resolution | |
| CN118570735B (en) | Crowd counting method based on semantic information optimization |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210924 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |