CN112348912A - Image reconstruction and foreign matter detection method based on RPCA and PCA - Google Patents
Image reconstruction and foreign matter detection method based on RPCA and PCA Download PDFInfo
- Publication number
- CN112348912A CN112348912A CN202011178617.4A CN202011178617A CN112348912A CN 112348912 A CN112348912 A CN 112348912A CN 202011178617 A CN202011178617 A CN 202011178617A CN 112348912 A CN112348912 A CN 112348912A
- Authority
- CN
- China
- Prior art keywords
- image
- data
- rpca
- pca
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 32
- 239000013598 vector Substances 0.000 claims abstract description 57
- 239000011159 matrix material Substances 0.000 claims abstract description 46
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 11
- 238000006243 chemical reaction Methods 0.000 claims abstract description 9
- 238000012216 screening Methods 0.000 claims abstract description 7
- 230000008030 elimination Effects 0.000 claims abstract description 5
- 238000003379 elimination reaction Methods 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 2
- 238000013450 outlier detection Methods 0.000 claims 7
- 230000007547 defect Effects 0.000 abstract description 11
- 230000000694 effects Effects 0.000 abstract description 6
- 230000017105 transposition Effects 0.000 abstract 1
- 238000000513 principal component analysis Methods 0.000 description 4
- 230000006835 compression Effects 0.000 description 3
- 238000007906 compression Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/001—Texturing; Colouring; Generation of texture or colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Image Processing (AREA)
Abstract
The invention relates to the technical field of defect detection, in particular to an image reconstruction and foreign object detection method based on RPCA and PCA, which comprises the following steps: acquiring a base vector: acquiring a sample image set; stretching and merging each sample image of the image set according to an RPCA algorithm; iterating to obtain a large low-rank matrix graph; carrying out standardization, down-sampling and PCA decomposition to obtain a basis vector and a transposition thereof required by image data conversion; image data reconstruction: matrix multiplication is carried out on the image to be detected and the base vector to obtain temporary data converted into a new space; performing matrix multiplication on the temporary data and the base vector transpose to obtain data which is converted back to the image space and subjected to noise reduction and foreign matter elimination; detecting foreign matters: differentiating the reconstructed image and the source image to be detected to obtain a set of noise data and foreign matter data; and screening out noise to obtain a foreign matter data set. The invention combines the RPCA algorithm and the PCA algorithm, and can also take effect in the scene requiring real-time performance.
Description
Technical Field
The invention relates to the technical field of defect detection, in particular to an image reconstruction and foreign matter detection method based on RPCA and PCA.
Background
The current MINILED detection environment has the pixel quality low, and MINILED original paper body objective reasons such as too little can't use traditional algorithm to carry out the detection and the location of defect, and the detection environment of industry needs high precision and detection efficiency again simultaneously.
The conventional algorithms currently include PCA and RPCA.
PCA refers to a principal component analysis algorithm; the principal component analysis is based on finding an orthogonal basis to construct a new matrix numerical space, and an original numerical matrix is converted into the new matrix numerical space to reduce the data volume, so that the compression processing of data is realized, and special noise similar to Gaussian noise categories can be eliminated.
With respect to PCA to find orthogonal basis space, there are several important steps, which have specific roles respectively, according to the data dimension reduction principle of PCA:
(1) covariance: typically to characterize the correlation between two variables. If the covariance is positive, the correlation is positive; if negative, negative correlation is obtained; if 0, the two variables are irrelevant;
(2) characteristic value: after EVD and SVD decomposition, two feature matrixes and one feature value matrix can be obtained, wherein each feature value represents the proportion of the corresponding feature, namely the more important feature in the score is, the larger the feature value is. In the PCA logic, the noise is reduced by screening out the features with small proportion and reserving the features with large proportion;
(3) feature centralization: corresponding to each pixel point of each sample in the image. Feature centralization, i.e., feature normalization, is to reduce the direct difference of features and to keep the sample features on a uniform scale.
The actual implementation flow of PCA is as follows:
1. sorting original matrix Xm×n
2. Solving an original matrix Xm×nOf the covariance matrix Sm×n＝Cov(X)
3. Eigenvalue and eigenvector of solving covariance matrix
4. Selecting the eigenvectors corresponding to the K eigenvalues with the largest eigenvalues to be combined into an orthogonal basis combination matrix Wn×k
5. Performing matrix space conversion, and calculating matrix Z in new spacem×k：
Zm×k＝Xm×nWn×k
Wherein, because K is specified artificially and the range is K ∈ [0, n ], when K is smaller than n, PCA conversion can be carried out to realize compression of matrix data.
The disadvantages of PCA are:
1. the noise reduction effect of the PCA algorithm is only additive, and only specific noise such as Gaussian noise can be processed, but discrete noise cannot be processed;
2. the PCA algorithm needs to carry out singular value decomposition for image compression every time, and the calculation amount is large, so that the PCA algorithm cannot be applied to a real-time scene.
RPCA refers to a robust principal component analysis algorithm; the robust principal component analysis is to apply the concept of machine learning, find a common low-rank matrix of a sample set, and then differentiate the obtained low-rank matrix and an original image to obtain a corresponding noise image. And returning the noise image to the original image for noise reduction.
The operation of RPCA requires a large amount of operation time, so based on accelerated logic, the conditions for solving RPCA by means of ALM (augmented lagrange multiplier) are divided into two types: IALM (non-precision ALM) and EALM (precision ALM).
The principle of using ALM to resolve RPCA is to fix the remaining parameters while solving for one parameter. Let D be a + E (D is the matrix to be decomposed, a is the low rank matrix, E is the noise matrix). It expands into lagrange formula:
for the EALM, two loops are constructed in the implementation, which are assumed to be an inner loop and an outer loop. The inner loop is a fixed penalty factor Y and a coefficient mu, and the matrixes A and E are converged until the matrixes A and E are not converged, and the matrixes Y and mu are updated again. It can be seen intuitively that this is extremely computationally intensive, but logically it is the result that is most needed to accurately fit RPCA.
The IALM differs from the EALM in that it only builds a loop, i.e., A, E, Y, μ are updated in the same loop, each loop being an approximate fit in the case of control Y and the coefficient μ. It is logically known that such a fitting result is less accurate than the earm, but has the advantage of reducing the amount and time of operation.
Correspondingly, RPCA has the disadvantages that:
1. a certain sufficient number of sample sets are required, and the final effect is proportional to the number of sample sets;
2. because of the partial process with machine learning, extra time is needed to train and learn, and the time cost cannot adapt to the real-time scene.
As can be seen, neither the PCA nor the RPCA algorithms are suitable for real-time scene detection.
Disclosure of Invention
The invention mainly aims to solve the problem of detection of unknown defects in the MINILED industrial detection environment, and can detect and position most of foreign matters on the premise of unknown defect details. Therefore, the invention provides an image reconstruction and foreign object detection method based on RPCA and PCA.
The specific scheme of the invention is as follows.
The image reconstruction and foreign body detection method based on RPCA and PCA comprises the following steps:
acquiring a base vector: acquiring a certain number of sample image sets; stretching and merging each sample image of the image set according to an RPCA algorithm; iterating according to an RPCA algorithm to obtain a low-rank matrix big image; normalizing each independent sample of the low-rank matrix large graph; down-sampling the standardized large low-rank matrix image to form a small low-rank image; carrying out PCA decomposition on the low-rank small image to obtain a basis vector and a transpose thereof required by image data conversion;
image data reconstruction: performing matrix multiplication operation on the image to be detected and the base vector to obtain temporary data converted into a new space; performing matrix multiplication operation on the temporary data converted into the new space and the base vector transpose to obtain data converted back into the image space after noise reduction and foreign matter elimination;
detecting foreign matters: differentiating the reconstructed image and the source image to be detected to obtain a set of noise data and foreign matter data; and screening out noise to obtain a foreign matter data set.
Further, the method also comprises the following steps of: clustering the foreign matter data set, analyzing and acquiring the center information corresponding to each category of foreign matter data, and corresponding to the image space coordinates to obtain the positioning information of the foreign matter.
Further, according to the RPCA algorithm, stretch merging is performed on each sample image of the image set, specifically: and stretching each sample image of the image set from a matrix data structure of m x n into a vector structure of m x n x 1, and splicing vectors of all the sample images into a large graph structure of m x n x z, wherein m and n are the sizes of single samples, and z is the number of samples.
Preferably, the method further comprises preprocessing the sample images before stretching each sample image from the matrix data structure of m x n to a vector structure of m x n x 1.
Further, the normalization process for each individual sample of the low rank matrix big map is:
wherein i, j are position coordinates of corresponding pixels, S' is a normalized single pixel value, S is a sample value after low-rank processing,
is the global mean of that pixel location.
Further, down-sampling the normalized low-rank matrix large graph to form a low-rank small graph, specifically: and randomly reserving g sample vectors in the z sample vectors, and recombining the g sample vectors into a low-rank small graph.
Preferably, in PCA decomposition of low rank minimaps, all eigenvalues are retained as basis vectors.
Further, for the multi-dimensional image, in the step of obtaining the basis vectors, each dimension is decomposed separately to obtain the basis vectors under each dimension, and finally, a plurality of groups of basis vectors and transposes thereof are obtained.
Further, for the multi-dimensional image, in the image data reconstruction step, the multi-dimensional image data is decomposed into each dimension data, and then data reconstruction is performed.
Further, for the multi-dimensional image, in the foreign object detection step, the reconstructed image and the source image to be detected are differentiated in each dimension respectively, and the intersection of the difference results of each dimension is obtained as the set of the noise data and the foreign object data.
The invention has the beneficial effects that: the scheme of the invention combines the RPCA algorithm and the PCA algorithm, so that the method can also take effect in a scene requiring real-time performance, can be quickly adapted according to a specific application scene, and simultaneously reserves the effects and advantages of the RPCA algorithm and the PCA algorithm; the scheme of the invention consumes little time in the detection process, and the actual time complexity mainly consists of the operation of the reconstructed image; the temporal complexity of the reconstructed image is determined only by the basis vectors and the image size.
Drawings
The invention is further illustrated by means of the attached drawings, but the embodiments in the drawings do not constitute any limitation to the invention, and for a person skilled in the art, other drawings can be derived on the basis of the following drawings without inventive effort.
Fig. 1 is a schematic flow chart of single-channel basis vector acquisition according to an embodiment of the present invention.
Fig. 2 is a schematic flowchart of a multi-channel basis vector acquisition process according to an embodiment of the present invention.
Fig. 3 is a schematic flowchart of single-dimensional data reconstruction according to an embodiment of the present invention.
Fig. 4 is a schematic flowchart of multi-dimensional data reconstruction according to an embodiment of the present invention.
Fig. 5 is a schematic flow chart of single-channel foreign object detection according to an embodiment of the present invention.
Fig. 6 is a flowchart illustrating a multi-channel foreign object detection method according to an embodiment of the invention.
Detailed Description
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
The invention comprehensively utilizes the characteristics of the RPCA algorithm and the PCA algorithm, readjusts and optimizes the algorithm, designs the scheme of the invention for image reconstruction and foreign object detection of an MINILED application scene, and can meet the requirements of high precision and high efficiency.
As shown in fig. 1, the present embodiment provides a single-channel basis vector acquisition process:
(1) single-channel sample set: a number of sample image sets are acquired. When acquiring a sample image set, it is necessary to ensure the uniformity of the field of view and the stability of the imaging environment when acquiring a sample as much as possible.
(2) Pretreatment: performing corresponding image preprocessing according to the environment difference so as to obtain a sample with better quality when performing RPCA processing subsequently; for example, an image captured in an environment with poor lighting conditions may be subjected to processing such as contrast enhancement or brightness enhancement before the following normal processing is performed.
(3) Stretching and combining: according to the algorithm principle of RPCA, all samples are collected, each sample image is stretched from a matrix data structure of m x n into a vector structure of (m x n) x 1, and then all sample vectors are spliced into a large graph structure of (m x n) x z, wherein m and n are the sizes of single samples, and z is the number of samples.
(4) RPCA: and (3) iterating a matrix large graph meeting the low-rank requirement by using an RPCA algorithm, wherein the noise reduction processing is actually carried out on the sample image set according to the principle of the low-rank matrix.
(5) Sample centralization: because a bias factor inevitably exists in each sample acquisition, before the basis vector is obtained, normalization needs to be performed on each independent sample in the current low-rank matrix large graph (i.e. the sample of the subsequent PCA decomposition), so as to reduce the error between samples as much as possible. The actual treatment is as follows:
wherein i, j is the position coordinate of the corresponding pixel, S' is the normalized single pixel value, and S is the sample after low rank processingThe value of the signal is that the signal is converted into a signal,
is the global mean of that pixel location.
(6) Down-sampling: in order to reduce the operating pressure of subsequent PCA decomposition for obtaining base vectors, g sample vectors in z sample vectors are randomly reserved and recombined into a low-rank minimap with smaller dimension and data size.
(7) PCA decomposition: and carrying out PCA decomposition on the low-rank small image to obtain a basis vector and a transpose thereof required by conversion, wherein the basis vector and the transpose are respectively used for converting to a new data space and converting back to an original image space. It should be noted that after the basis vectors are obtained, feature value screening is not required according to the conventional PCA principle logic, and it can be ensured that the data information quantity of the reconstructed image is sufficient by completely reserving all feature values as the basis vectors.
As shown in fig. 2, the present embodiment provides a multi-channel basis vector obtaining process;
the overall process is the same as the single dimension in the multi-dimensional environment, and it should be noted that in the multi-dimensional environment, each dimension needs to be separately decomposed to sequentially and independently perform the calculation of the basis vector in the single dimension, for example, if the data to be processed is n dimensions, n groups of basis vectors and transposes thereof are finally obtained.
Fig. 3 shows a single-dimensional data reconstruction process according to this embodiment:
(1) multiplication by the basis vector ontology: and performing matrix multiplication operation on the image to be detected and the basis vector to obtain temporary data converted into a new space. It should be noted that before the operation is performed, the image to be measured still needs to be subjected to sample centering, and at this time, the mean sample required for the sample centering is a mean image obtained when the basis vector is obtained.
(2) Multiplication by base vector transpose: and performing matrix multiplication operation on the processed data (namely the temporary data converted into the new space) of the image to be detected and the base vector transpose to obtain the data converted back into the image space after noise reduction and foreign matter elimination. Correspondingly, after the conversion back to the image space, the operation of de-centering the sample is required to eliminate the offset effect caused by the centering of the sample.
Fig. 4 shows a multi-dimensional data reconstruction process according to the present embodiment;
the overall principle of data reconstruction in the multi-dimensional data environment is the same as that in the single-dimensional environment; when reconstructing multidimensional data, the multidimensional data is decomposed into independent single-dimensional data for reconstruction according to the logic of solving for the basis vector, so that the independence of the dimensional data and the consistency of the logic of solving for the basis vector are guaranteed.
In addition, after a single-channel (namely single-dimensional) image with noise reduction and foreign matter elimination is obtained, the image can be recombined according to the original color space rule; for example, the source image data is in a three-channel RPG format, and the source image data is split and recombined according to a B-G-R channel sequencing rule before and after the splitting processing; and finally, obtaining a reconstructed image.
Fig. 5 is a flow chart of single-channel foreign object detection according to this embodiment.
When a reconstructed image from which foreign matter and noise are removed can be actually acquired, the information at this time is as follows:
M+V+F＝Morg
wherein M is pure image data without noise and foreign objects; v is the noise information amount, and F is the foreign matter information amount; morgIs the original data.
Therefore, the single-channel foreign object detection flow provided by this embodiment is as follows:
(1) and differentiating the reconstructed image and the source image to be detected: from the information relational expression, it is understood that the difference data between the reconstructed image and the original image is a set of noise data and foreign object data.
(2) Screening and removing noise: the foreign-matter data is separated from the noise data using a corresponding denoising algorithm according to the characteristics of the noise itself. The purpose of this step is to eliminate the possible non-target foreign matter data, for example, the small-particle discrete point noise can be directly screened out by the size difference with the foreign matter data. The selection of the noise filtering algorithm may be determined according to a specific use scenario, and is not described herein again.
(3) Clustering and analyzing foreign body data: in the previous step, relatively pure foreign object data is obtained, and since the foreign object data is a set, clustering processing, such as contour classification, connected domain processing, K nearest neighbor and the like, is firstly required. Then, the clustered foreign matter data can be further analyzed and processed; for example, the center information corresponding to each category of foreign matter data is obtained, and the image space coordinates are corresponding to the center information, namely the positioning information of the foreign matter; or, the clustered data is identified and subjected to semantic analysis and the like.
Fig. 6 is a multi-channel foreign object detection process according to this embodiment;
the detection process under the multi-dimension is similar to that of the single dimension, and the difference lies in that dimension division processing is needed in the logic of difference processing, finally the intersection of all dimensions is taken as a result, subsequent clustering and analysis processing are continued, the central information corresponding to the foreign matter data of each category is obtained, the image space coordinate is corresponded, and the positioning information of the foreign matter is finally obtained.
For the scheme of the invention, the inventor applies the specific implementation thereof to the actual scene, specifically as follows:
1. at least 300 sample images are acquired under the same FOV (field of view). The sample image needs to satisfy the condition of uniform angle and view field position.
2. And training and modeling are carried out by using the sample set, and a conversion basis matrix is obtained.
3. When in use, the uniform view field and angle are ensured. And after the corresponding samples are acquired, the image to be detected is transformed by using the basis matrix, and a reconstructed image is obtained. And acquiring sample sets with different numbers (original images, 20 samples, 50 samples, 100 samples and 200 samples) respectively to acquire image examples reconstructed after the conversion matrixes are acquired. Since the images under different numbers of samples involve color variations, the black and white drawings cannot show the variations, and therefore the present embodiment is not illustrated.
4. And after the difference is carried out on the reconstructed image and the source image, a foreign body image is obtained through screening treatment. Images screened by a simple differential threshold under the above sample number can be obtained. Similarly, since images at different thresholds involve color changes, black and white figures cannot show the changes, and thus the embodiment is not illustrated.
5. And then, subsequent corresponding operations such as positioning, classifying, labeling and the like can be carried out on the foreign object image according to the actual application requirement.
The embodiment of the invention provides a defect and foreign matter detection scheme on the premise of unknown foreign matter and defect specific information, and the scheme can be quickly adapted according to a specific application scene.
Specifically, the scheme of the invention combines the RPCA algorithm and the PCA algorithm, so that the algorithm can be effective in a scene requiring real-time performance, and meanwhile, the utility and the advantages of the RPCA algorithm and the PCA algorithm are kept:
1. the RPCA and the PCA have no real-time performance, but the two algorithms are adjusted and combined to be well suitable for a real-time scene;
2. the method can also carry out detection with certain robustness on the premise that specific information of foreign matters or defects is unknown, namely, the method has certain universality.
The scheme of the invention consumes little time in the detection process, and the actual time complexity is mainly formed by the operation of the reconstructed image. The temporal complexity of the reconstructed image is determined only by the basis vectors and the image size. And does not include a loop operation, which is a single linear operation.
In the field of industrial vision, a solution for detecting and positioning the defect of the foreign matter on the premise of not needing the foreign matter and the specific information of the defect does not exist at present; therefore, the scheme of the invention is a new breakthrough and an innovative solution in the field of industrial visual foreign body defect detection.
The above description is only a preferred embodiment of the present invention, and all equivalent changes or modifications of the structure, characteristics and principles described in the present patent application are included in the protection scope of the present patent application.
Claims (10)
1. The image reconstruction and foreign object detection method based on RPCA and PCA is characterized by comprising the following steps:
acquiring a base vector: acquiring a certain number of sample image sets; stretching and merging each sample image of the image set according to an RPCA algorithm; iterating according to an RPCA algorithm to obtain a low-rank matrix big image; normalizing each independent sample of the low-rank matrix large graph; down-sampling the standardized large low-rank matrix image to form a small low-rank image; carrying out PCA decomposition on the low-rank small image to obtain a basis vector and a transpose thereof required by image data conversion;
image data reconstruction: performing matrix multiplication operation on the image to be detected and the base vector to obtain temporary data converted into a new space; performing matrix multiplication operation on the temporary data converted into the new space and the base vector transpose to obtain data converted back into the image space after noise reduction and foreign matter elimination;
detecting foreign matters: differentiating the reconstructed image and the source image to be detected to obtain a set of noise data and foreign matter data; and screening out noise to obtain a foreign matter data set.
2. The RPCA and PCA-based image reconstruction and anomaly detection method according to claim 1, further comprising the step of: clustering the foreign matter data set, analyzing and acquiring the center information corresponding to each category of foreign matter data, and corresponding to the image space coordinates to obtain the positioning information of the foreign matter.
3. The RPCA and PCA-based image reconstruction and outlier detection method of claim 1, wherein each sample image of the image set is stretched and merged according to an RPCA algorithm, specifically: and stretching each sample image of the image set from a matrix data structure of m x n into a vector structure of m x n x 1, and splicing vectors of all the sample images into a large graph structure of m x n x z, wherein m and n are the sizes of single samples, and z is the number of samples.
4. The RPCA and PCA based image reconstruction and anomaly detection method of claim 3 further comprising preprocessing the sample images prior to stretching each sample image from a matrix data structure of m x n to a vector structure of m x n 1.
5. The RPCA and PCA-based image reconstruction and outlier detection method of claim 1 wherein the normalization of each individual sample of the large low rank matrix is:
wherein i, j are position coordinates of corresponding pixels, S' is a normalized single pixel value, S is a sample value after low-rank processing,
is the global mean of that pixel location.
6. The image reconstruction and outlier detection method based on RPCA and PCA as claimed in claim 3, wherein the normalized large low rank matrix map is down-sampled to form a small low rank map, specifically: and randomly reserving g sample vectors in the z sample vectors, and recombining the g sample vectors into a low-rank small graph.
7. The RPCA and PCA-based image reconstruction and outlier detection method of claim 1 wherein all eigenvalues are retained as basis vectors in PCA decomposition of low rank mini-maps.
8. The image reconstruction and foreign object detection method based on RPCA and PCA as claimed in claim 1, wherein for multi-dimensional images, in the step of obtaining basis vectors, each dimension is decomposed separately to obtain basis vectors under each dimension, and finally, a plurality of sets of basis vectors and their transposes are obtained.
9. The image reconstruction and outlier detection method based on RPCA and PCA as claimed in claim 1, wherein for multi-dimensional images, in the image data reconstruction step, the multi-dimensional image data is decomposed into each dimension data before data reconstruction.
10. The image reconstruction and outlier detection method based on RPCA and PCA as claimed in claim 1, wherein for multi-dimensional images, in the outlier detection step, the reconstructed image is differentiated from the source image to be detected in each dimension, and the intersection of the difference results in each dimension is obtained as the set of noise data and outlier data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011178617.4A CN112348912A (en) | 2020-10-29 | 2020-10-29 | Image reconstruction and foreign matter detection method based on RPCA and PCA |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011178617.4A CN112348912A (en) | 2020-10-29 | 2020-10-29 | Image reconstruction and foreign matter detection method based on RPCA and PCA |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112348912A true CN112348912A (en) | 2021-02-09 |
Family
ID=74357004
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011178617.4A Pending CN112348912A (en) | 2020-10-29 | 2020-10-29 | Image reconstruction and foreign matter detection method based on RPCA and PCA |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112348912A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114792295A (en) * | 2022-06-23 | 2022-07-26 | 深圳憨厚科技有限公司 | Method, device, equipment and medium for correcting blocked object based on intelligent photo frame |
| US11982569B2 (en) | 2021-08-20 | 2024-05-14 | Eagle Technology, Llc | Spectrographic system that compresses fourier transform spectral data and associated methods |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105427300A (en) * | 2015-12-21 | 2016-03-23 | 复旦大学 | Low-rank expression and learning dictionary-based hyperspectral image abnormity detection algorithm |
| CN108510013A (en) * | 2018-07-02 | 2018-09-07 | 电子科技大学 | The steady tensor principal component analytical method of improvement based on low-rank kernel matrix |
| CN108577840A (en) * | 2018-02-11 | 2018-09-28 | 南京邮电大学 | A kind of steady PCA imaging methods of dynamic magnetic resonance |
| CN109063616A (en) * | 2018-07-23 | 2018-12-21 | 杭州安恒信息技术股份有限公司 | A kind of time series Singularity detection method restored based on matrix low-rank |
| US20190050372A1 (en) * | 2017-08-14 | 2019-02-14 | City University Of Hong Kong | Systems and methods for robust low-rank matrix approximation |
-
2020
- 2020-10-29 CN CN202011178617.4A patent/CN112348912A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105427300A (en) * | 2015-12-21 | 2016-03-23 | 复旦大学 | Low-rank expression and learning dictionary-based hyperspectral image abnormity detection algorithm |
| US20190050372A1 (en) * | 2017-08-14 | 2019-02-14 | City University Of Hong Kong | Systems and methods for robust low-rank matrix approximation |
| CN108577840A (en) * | 2018-02-11 | 2018-09-28 | 南京邮电大学 | A kind of steady PCA imaging methods of dynamic magnetic resonance |
| CN108510013A (en) * | 2018-07-02 | 2018-09-07 | 电子科技大学 | The steady tensor principal component analytical method of improvement based on low-rank kernel matrix |
| CN109063616A (en) * | 2018-07-23 | 2018-12-21 | 杭州安恒信息技术股份有限公司 | A kind of time series Singularity detection method restored based on matrix low-rank |
Non-Patent Citations (2)
| Title |
|---|
| CHEN, YY等: "ROBUST PRINCIPAL COMPONENT ANALYSIS WITH MATRIX FACTORIZATION", 《2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)》, 25 October 2018 (2018-10-25), pages 2411 - 2415 * |
| 王文韬: "矩阵重建求解算法研究及应用", 《中国优秀硕士学位论文全文数据库》, no. 11, 30 November 2014 (2014-11-30), pages 1 - 62 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11982569B2 (en) | 2021-08-20 | 2024-05-14 | Eagle Technology, Llc | Spectrographic system that compresses fourier transform spectral data and associated methods |
| CN114792295A (en) * | 2022-06-23 | 2022-07-26 | 深圳憨厚科技有限公司 | Method, device, equipment and medium for correcting blocked object based on intelligent photo frame |
| CN114792295B (en) * | 2022-06-23 | 2022-11-04 | 深圳憨厚科技有限公司 | Method, device, equipment and medium for correcting blocked object based on intelligent photo frame |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106650770B (en) | Mura defect detection method based on sample learning and human eye visual characteristics | |
| Raytchev et al. | Head pose estimation by nonlinear manifold learning | |
| WO2016091017A1 (en) | Extraction method for spectral feature cross-correlation vector in hyperspectral image classification | |
| WO2017080196A1 (en) | Video classification method and device based on human face image | |
| CN112348912A (en) | Image reconstruction and foreign matter detection method based on RPCA and PCA | |
| CN105893916A (en) | New method for detection of face pretreatment, feature extraction and dimensionality reduction description | |
| CN114863189B (en) | Intelligent image identification method based on big data | |
| Paul et al. | PCA based geometric modeling for automatic face detection | |
| WO2017092272A1 (en) | Face identification method and device | |
| CN108932468B (en) | Face recognition method suitable for psychology | |
| CN113963193A (en) | Method and device for generating vehicle body color classification model and storage medium | |
| Setiawan et al. | Mixture gaussian V2 based microscopic movement detection of human spermatozoa | |
| Dave | A lip localization based visual feature extraction method | |
| Akhmetshina et al. | Improvement of Grayscale Images in Orthogonal Basis of the Type-2 Membership Function. | |
| CN111325720B (en) | Multi-visual feature integrated reference-free super-resolution image quality evaluation method | |
| CN111951298B (en) | Target tracking method integrating time sequence information | |
| CN113887652A (en) | Remote sensing image dim target detection method based on form and multi-example learning | |
| CN107122795B (en) | Pedestrian re-identification method based on coring characteristics and random subspace integration | |
| CN111126485A (en) | Lie-KFDA scene classification method and system based on Lie group machine learning kernel function | |
| Huang et al. | Color Classification of Vehicles Based on Two-Layer Salincy, Illumination-Invariant Transformation, and Adaptive KNN. | |
| CN109684955B (en) | Intelligent environment recognition method based on deep learning | |
| CN110991313B (en) | Moving small target detection method and system based on background classification | |
| CN113435301B (en) | Animal fur microscopic image classification method, device, electronic equipment and storage medium | |
| Zhan et al. | Robust local tangent space alignment | |
| CN116735610A (en) | Steel pipe surface defect detection method based on machine vision |
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 |