CN104966278A - Hyperspectral image noise removing method and apparatus - Google Patents

Hyperspectral image noise removing method and apparatus Download PDF

Info

Publication number
CN104966278A
CN104966278A CN201510423725.6A CN201510423725A CN104966278A CN 104966278 A CN104966278 A CN 104966278A CN 201510423725 A CN201510423725 A CN 201510423725A CN 104966278 A CN104966278 A CN 104966278A
Authority
CN
China
Prior art keywords
energy
noise
tensor
spectrum image
high spectrum
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201510423725.6A
Other languages
Chinese (zh)
Other versions
CN104966278B (en
Inventor
孟书书
王文钦
邵怀宗
陈慧
潘晔
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201510423725.6A priority Critical patent/CN104966278B/en
Publication of CN104966278A publication Critical patent/CN104966278A/en
Application granted granted Critical
Publication of CN104966278B publication Critical patent/CN104966278B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Image Processing (AREA)

Abstract

The present invention discloses a hyperspectral image noise removing method and apparatus. The method comprises: performing Tucker decomposition on a hyperspectral image to obtain a core tensor; calculating an energy information vector of the core tensor in three dimensions; calculating an energy-to-noise ratio of the hyperspectral image in the three dimensions; calculating a tensor rank according to the energy information vector and the energy-to-noise ratio; and decomposing the hyperspectral image according to the tensor rank to remove noise.

Description

A kind of method and device removing the noise of high spectrum image
Technical field
The present invention relates to technical field of image processing, particularly relate to a kind of method and the device of removing the noise of high spectrum image.
Background technology
The spectrum picture of spectral resolution within the scope of the 10l order of magnitude is called high spectrum image.Usually, the equipment gathering high spectrum image includes multiple sensor that can gather the lightwave data of different wave length, and each sensor can generate a width two dimensional gray figure on the wavelength at its place.Like this, these two gray-scale maps being combined, is exactly the high spectrum image of a width three-dimensional.
Removing the noise of high spectrum image is that application high spectrum image carries out the crucial pre-treatment step of of the subsequent treatment such as terrain classification and target identification.First 3 d image data can be transformed into two-dimensional image data by the method for the noise of more existing removal high spectrum images, then does denoising by the method for two-dimentional denoising to high spectrum image.Converted by 3 d image data in the process of journey two-dimensional image data, the spatial structural form in 3 d image data can be destroyed, and the denoising effect causing it final is not ideal enough.
Summary of the invention
The object of the present invention is to provide a kind of method and the device of removing the noise of high spectrum image, with the problem that the effect solving prior art removal high spectrum image noise is undesirable.
An embodiment provides a kind of method removing the noise of high spectrum image, comprising: Tucker decomposition is done to high spectrum image and obtains core tensor; Calculate the energy information vector of core tensor in three dimensions; Calculate the energy noise ratio of high spectrum image in three dimensions; According to energy information vector energy noise than compute tensor order; And decompose to remove noise to high spectrum image according to tensor order.
An alternative embodiment of the invention provides a kind of device removing the noise of high spectrum image, comprising: core tensor resolution module, obtains core tensor for doing Tucker decomposition to high spectrum image; Energy information vectors calculation module, for calculating the energy information vector of core tensor in three dimensions; Energy noise than computing module, for calculating the energy noise ratio of high spectrum image in three dimensions; Tensor order computing module, for according to energy information vector energy noise than compute tensor order; And picture breakdown module, for decomposing to remove noise to high spectrum image according to tensor order.
Accompanying drawing explanation
By reading hereafter detailed description of the preferred embodiment, various other advantage and benefit will become cheer and bright for those of ordinary skill in the art.Accompanying drawing only for illustrating the object of preferred implementation, and does not think limitation of the present invention.And in whole accompanying drawing, represent identical parts by identical reference symbol.Wherein in the accompanying drawings, the multiple identical parts of alphabetic flag instruction after reference number, when making a general reference these parts, by its last alphabetic flag of omission.In the accompanying drawings:
Fig. 1 is the process flow diagram of an embodiment of the method for the noise of removal high spectrum image of the present invention;
The process flow diagram of the embodiment that Fig. 2 is the step 102 in method shown in Fig. 1;
The process flow diagram of the embodiment that Fig. 3 is the step 103 in method shown in Fig. 1;
The process flow diagram of the embodiment that Fig. 4 is the step 104 in method shown in Fig. 1;
Fig. 5 is the schematic block diagram of a device embodiment of the noise of removal high spectrum image of the present invention.
In the accompanying drawings, use same or similar label to refer to same or similar element.
Embodiment
Illustrative embodiments of the present invention is described in detail referring now to accompanying drawing.The embodiment that to should be appreciated that shown in accompanying drawing and describe is only exemplary, is intended to explain principle of the present invention and spirit, and not limits the scope of the invention.
It is the process flow diagram of an embodiment 100 of the method for the noise of removal high spectrum image of the present invention with reference to figure 1, Fig. 1.Embodiment 100 shown in Fig. 1 can comprise following steps 101 to 105.
In a step 101, Tucker decomposition is done to high spectrum image and obtain core tensor.
In one embodiment of the invention, can do high spectrum image and decompose without the Tucker that blocks.That is: if containing noisy high spectrum image then by decomposing without the Tucker blocked the core tensor obtained wherein, represent real number space dimension, I 1, I 2, I 3represent respectively and tie up size in three dimensions containing noisy high spectrum image at space level dimension, spatial vertical peacekeeping spectrum.
In a step 102, the energy information vector of core tensor in three dimensions is calculated.
In one embodiment of the invention, with reference to figure 2, step 102 can comprise following sub-step 201 to 202.
In sub-step 201, tensor is done in three dimensions to core tensor and launches to obtain launching matrix.
In one embodiment of the invention, matrix is launched launch matrix launch matrix
In sub-step 202, by calculating the row norm of expansion matrix to obtain energy information vector.
In one embodiment of the invention, Matrix C is launched (1)row norm c 1namely be the energy information vector in the first dimension.Similarly, Matrix C is launched (2)row norm c 2namely be the energy information vector in the second dimension; Launch Matrix C (3)row norm c 3namely be the energy information vector in third dimension.
In step 103, the energy noise ratio of high spectrum image in three dimensions is calculated.
In one embodiment of the invention, with reference to figure 3, step 103 can comprise following sub-step 301 to 303.
In sub-step 301, estimate the regression coefficient vector in three dimensions by least square method.
In one embodiment of the invention, the regression coefficient vector in three dimensions can be estimated by following formula (1):
α ^ i ( n ) = ( W i ( n ) T W i ( n ) ) - 1 W i ( n ) T w i ( n ) - - - ( 1 )
Wherein represent the regression coefficient vector in the n-th dimension, represent high spectrum image n dimension launch in transpose of a matrix matrix, the matrix that the later remaining columns of removing i-th row forms; represent transposed matrix; represent high spectrum image n dimension launch in transpose of a matrix matrix the i-th column vector.
In sub-step 302, estimate the noise in three dimensions according to regression coefficient vector high spectrum image.
In one embodiment of the invention, the noise in three dimensions can be estimated by following formula (2):
ϵ ^ i ( n ) = w i ( n ) - W i ( n ) α ^ i ( n ) - - - ( 2 )
Wherein, represent the noise in the n-th dimension,
In sub-step 303, according to the normalized energy noise ratio under energy balane three dimensions of the noise on the energy of high spectrum image and three dimensions.
In one embodiment of the invention, the normalized energy noise ratio in three dimensions can be calculated by following formula (3):
Wherein, ω nrepresent the normalized energy noise ratio in the n-th dimension, || || frepresent and ask F-norm.
At step 104, according to energy information vector energy noise than compute tensor order.
In one embodiment of the invention, the tensor order r in the n-th dimension ncan be the value of satisfied following formula (4):
m i n r n | | ω n - Σ i = r n + 1 I n c n ( i ) Σ j = 1 I n c n ( j ) | | F 2 - - - ( 4 )
That is, when r time minimum n, just can think the tensor order in the n-th dimension.
Above-mentioned r is obtained in order to search for n, in one embodiment of the invention, with reference to figure 4, step 104 can comprise following sub-step 401 to 404.
In sub-step 401, search for multiple alternative tensor order.
Specifically, one by one r can be made nequal 1 to I nbetween each number, formed multiple alternative tensor order.
In sub-step 402, calculate the ratio of noise energy in energy information vector and gross energy according to alternative tensor order.
Specifically, can calculate according to each alternative tensor order to obtain the ratio of noise energy in multiple energy information vector and gross energy.Molecule in this formula is noise energy, and denominator is gross energy.
In sub-step 403, calculate energy noise than the difference with ratio.
Specifically, can pass through calculate the ratio of noise energy in the energy information vector obtained by each alternative tensor order and gross energy ω is compared with energy noise ndifference.
In sub-step 404, elect alternative tensor order minimum for difference as tensor order.
In one embodiment of the invention, the difference obtained due to sub-step 403 is a vector, therefore, can be judged the size of this difference by the method for the F-norm calculating this vector.R corresponding to the difference of F-Norm minimum n, just can think the tensor order in the n-th dimension.
In step 105, decompose to remove noise to high spectrum image according to tensor order.
Tensor order r=(r in three dimensions that can calculate by step 104 1, r 2, r 3) to containing noisy high spectrum image decompose, obtain removing the later high spectrum image of noise.
Specifically, tensor order r=(r can be used 1, r 2, r 3) to containing noisy high spectrum image carry out Tucker decomposition, by giving up last I n-r n(n=1,2,3) individual composition removes noise.
To the method for the noise of the removal high spectrum image that described herein according to the embodiment of the present invention.
The method of the noise of the removal high spectrum image that the present invention proposes, integrally can process high spectrum image, and not need the view data of three-dimensional to be first transformed into 2-D data.Therefore, relative to prior art, the inventive method can protect the more structural information of view data inside.In addition, owing to having taken into full account the noise of space dimension and spectrum dimension, therefore, relative to prior art, the inventive method can utilize the information distribution situation of each dimension more fully.Based on this, the inventive method can remove the noise of high spectrum image better.
Similar with the method, present invention also offers the corresponding device removing the noise of high spectrum image.
Figure 5 shows that the schematic block diagram of an embodiment 500 of the device of the noise of removal high spectrum image of the present invention.
As shown in the figure, device 500 can comprise: core tensor resolution module 501, obtains core tensor for doing Tucker decomposition to high spectrum image; Energy information vectors calculation module 502, for calculating the energy information vector of core tensor in three dimensions; Energy noise than computing module 503, for calculating the energy noise ratio of high spectrum image in three dimensions; Tensor order computing module 504, for according to energy information vector energy noise than compute tensor order; And picture breakdown module 505, for decomposing to remove noise to high spectrum image according to tensor order.
In one embodiment of the invention, core tensor resolution module 501 may further include: without blocking Tucker decomposing module, decomposes without the Tucker blocked for doing high spectrum image.
In one embodiment of the invention, energy information vectors calculation module 502 may further include: core tensor launches module, launches to obtain launching matrix for doing tensor in three dimensions to core tensor; And row norm calculation module, for the row norm by calculating expansion matrix to obtain energy information vector.
In one embodiment of the invention, energy noise may further include than computing module 503: regression coefficient vector estimation block, for estimating the regression coefficient vector in three dimensions by least square method; Noise estimation module, for estimating the noise in three dimensions according to regression coefficient vector high spectrum image; And normalized energy noise ratio computing module, for according to the normalized energy noise ratio in energy balane three dimensions of the noise on the energy of high spectrum image and three dimensions.
In one embodiment of the invention, tensor order computing module 504 may further include: alternative tensor order search module, for searching for multiple alternative tensor order; Ratio calculation module, for the ratio according to the noise energy in alternative tensor order calculating energy information vector and gross energy; Difference calculating module, for calculating energy noise than the difference with ratio; And choose module, for electing alternative tensor order minimum for difference as tensor order.
To the device of the noise of the removal high spectrum image that described herein according to the embodiment of the present invention.
The device of the noise of the removal high spectrum image that the present invention proposes, integrally can process high spectrum image, and not need the view data of three-dimensional to be first transformed into 2-D data.Therefore, relative to prior art, apparatus of the present invention can protect the more structural information of view data inside.In addition, owing to having taken into full account the noise of space dimension and spectrum dimension, therefore, relative to prior art, apparatus of the present invention can utilize the information distribution situation of each dimension more fully.Based on this, apparatus of the present invention can remove the noise of high spectrum image better.

Claims (10)

1. remove a method for the noise of high spectrum image, it is characterized in that, comprising:
Tucker decomposition is done to described high spectrum image and obtains core tensor;
Calculate the energy information vector of described core tensor in three dimensions;
Calculate the energy noise ratio of described high spectrum image in three dimensions;
According to described energy information vector, energy noise is than compute tensor order; And
Decompose to remove noise to described high spectrum image according to described tensor order.
2. method according to claim 1, is characterized in that, is describedly Tucker to described high spectrum image and decomposes the step obtaining core tensor and comprise further:
Described high spectrum image is done and decomposes without the Tucker blocked.
3. method according to claim 1, is characterized in that, the step of the energy information vector of the described core tensor of described calculating comprises further:
In three dimensions, do tensor to described core tensor to launch to obtain launching matrix; And
By calculating the row norm of described expansion matrix to obtain described energy information vector.
4. method according to claim 1, is characterized in that, the step of the energy noise ratio of the described high spectrum image of described calculating in three dimensions comprises further:
The regression coefficient vector in three dimensions is estimated by least square method;
According to described regression coefficient vector, high spectrum image estimates the noise in three dimensions; And
According to the normalized energy noise ratio in energy balane three dimensions of the noise on the energy of described high spectrum image and three dimensions.
5. method according to claim 1, is characterized in that, described according to described energy information vector energy noise comprise further than the step of compute tensor order:
Search for multiple alternative tensor order;
The ratio of noise energy in described energy information vector and gross energy is calculated according to described alternative tensor order;
Calculate described energy noise than the difference with described ratio; And
Elect alternative tensor order minimum for described difference as tensor order.
6. remove a device for the noise of high spectrum image, it is characterized in that, comprising:
Core tensor resolution module, obtains core tensor for doing Tucker decomposition to described high spectrum image;
Energy information vectors calculation module, for calculating the energy information vector of described core tensor in three dimensions;
Energy noise than computing module, for calculating the energy noise ratio of described high spectrum image in three dimensions;
Tensor order computing module, for energy noise according to described energy information vector than compute tensor order; And
Picture breakdown module, for decomposing to remove noise to described high spectrum image according to described tensor order.
7. device according to claim 6, is characterized in that, described core tensor resolution module comprises further:
Without blocking Tucker decomposing module, decompose without the Tucker blocked for doing described high spectrum image.
8. device according to claim 6, is characterized in that, described energy information vectors calculation module comprises further:
Core tensor launches module, launches to obtain launching matrix for doing tensor in three dimensions to described core tensor; And
Row norm calculation module, for the row norm by calculating described expansion matrix to obtain described energy information vector.
9. method according to claim 1, is characterized in that, described energy noise comprises further than computing module:
Regression coefficient vector estimation block, for estimating the regression coefficient vector in three dimensions by least square method;
Noise estimation module, estimates the noise in three dimensions for high spectrum image according to described regression coefficient vector; And
Normalized energy noise ratio computing module, for according to the normalized energy noise ratio in energy balane three dimensions of the noise on the energy of described high spectrum image and three dimensions.
10. method according to claim 1, is characterized in that, described tensor order computing module comprises further:
Alternative tensor order search module, for searching for multiple alternative tensor order;
Ratio calculation module, for calculating the ratio of noise energy in described energy information vector and gross energy according to described alternative tensor order;
Difference calculating module, for calculating described energy noise than the difference with described ratio; And
Choose module, for electing alternative tensor order minimum for described difference as tensor order.
CN201510423725.6A 2015-07-17 2015-07-17 A kind of method and device for the noise for removing high spectrum image Expired - Fee Related CN104966278B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510423725.6A CN104966278B (en) 2015-07-17 2015-07-17 A kind of method and device for the noise for removing high spectrum image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510423725.6A CN104966278B (en) 2015-07-17 2015-07-17 A kind of method and device for the noise for removing high spectrum image

Publications (2)

Publication Number Publication Date
CN104966278A true CN104966278A (en) 2015-10-07
CN104966278B CN104966278B (en) 2018-01-19

Family

ID=54220309

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510423725.6A Expired - Fee Related CN104966278B (en) 2015-07-17 2015-07-17 A kind of method and device for the noise for removing high spectrum image

Country Status (1)

Country Link
CN (1) CN104966278B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106157340A (en) * 2016-07-07 2016-11-23 天津工业大学 Interference high spectrum image decomposition method based on the full variational algorithm of self-adapting regular parameter
CN106651793A (en) * 2016-11-29 2017-05-10 北京空间机电研究所 PST stray light test data processing method
CN114036984A (en) * 2021-11-08 2022-02-11 山东大学 High-dimensional spectrum data noise reduction method based on TUCKALS3 algorithm

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120038798A1 (en) * 2010-08-11 2012-02-16 Woods Gary L Techniques for Removing Noise in a Compressive Imaging Device
CN102436586A (en) * 2011-10-28 2012-05-02 哈尔滨工业大学 Hyper spectral image classification method based on wavelet threshold denoising and empirical mode decomposition
CN104063852A (en) * 2014-07-07 2014-09-24 温州大学 Tensor recovery method based on indexed nuclear norm and mixed singular value truncation
CN104182941A (en) * 2014-08-26 2014-12-03 中国石油大学(华东) Hyperspectral image band noise removing method
CN104268838A (en) * 2014-09-28 2015-01-07 武汉大学 Wavelet denoising algorithm oriented to hyperspectral databases

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120038798A1 (en) * 2010-08-11 2012-02-16 Woods Gary L Techniques for Removing Noise in a Compressive Imaging Device
CN102436586A (en) * 2011-10-28 2012-05-02 哈尔滨工业大学 Hyper spectral image classification method based on wavelet threshold denoising and empirical mode decomposition
CN104063852A (en) * 2014-07-07 2014-09-24 温州大学 Tensor recovery method based on indexed nuclear norm and mixed singular value truncation
CN104182941A (en) * 2014-08-26 2014-12-03 中国石油大学(华东) Hyperspectral image band noise removing method
CN104268838A (en) * 2014-09-28 2015-01-07 武汉大学 Wavelet denoising algorithm oriented to hyperspectral databases

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JOSÉ M. BIOUCAS-DIAS,JOSÉ M. P. NASCIMENTO: "Hyperspectral Subspace Identification", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 *
YI PENG ET AL: "Decomposable Nonlocal Tensor Dictionary Learning for Multispectral Image Denoising", 《2014 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 *
郭贤等: "采用张量子空间的高光谱影像多维滤波算法", 《测绘学报》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106157340A (en) * 2016-07-07 2016-11-23 天津工业大学 Interference high spectrum image decomposition method based on the full variational algorithm of self-adapting regular parameter
CN106651793A (en) * 2016-11-29 2017-05-10 北京空间机电研究所 PST stray light test data processing method
CN106651793B (en) * 2016-11-29 2019-05-24 北京空间机电研究所 A kind of PST Stray Light Test data processing method
CN114036984A (en) * 2021-11-08 2022-02-11 山东大学 High-dimensional spectrum data noise reduction method based on TUCKALS3 algorithm
CN114036984B (en) * 2021-11-08 2024-04-16 山东大学 High-dimensional spectrum data noise reduction method based on TUCKALS algorithm

Also Published As

Publication number Publication date
CN104966278B (en) 2018-01-19

Similar Documents

Publication Publication Date Title
US10748029B2 (en) Spatial transformer modules
CN109643383B (en) Domain split neural network
CN111369487B (en) Hyperspectral and multispectral image fusion method, system and medium
Nir et al. Optimal and efficient streak detection in astronomical images
US9501852B2 (en) Method and apparatus for image fusion
CN104732535B (en) It is a kind of to constrain sparse non-negative matrix factorization method
CN110139046B (en) Tensor-based video frame synthesis method
WO2017214968A1 (en) Method and apparatus for convolutional neural networks
CN108154133B (en) Face portrait-photo recognition method based on asymmetric joint learning
Kashyap et al. Synthesis and estimation of random fields using long-correlation models
CN111488901B (en) Method and device for extracting features from input images in multiple modules in CNN
CN103810755A (en) Method for reconstructing compressively sensed spectral image based on structural clustering sparse representation
CN103971346A (en) SAR (Synthetic Aperture Radar) image spot-inhibiting method based on spare domain noise distribution constraint
CN104966278A (en) Hyperspectral image noise removing method and apparatus
CN106157254A (en) Rarefaction representation remote sensing images denoising method based on non local self-similarity
CN104680487A (en) Non-local image inpainting method based on low-rank matrix recovery
Vos et al. A generative machine learning approach to RFI mitigation for radio astronomy
EP3771999B1 (en) Method and apparatus for extracting image data in parallel from multiple convolution windows, device, and computer-readable storage medium
Kazaryan et al. Research of generalized wavelet transformations of Haar correctness in remote sensing of the Earth
CN109558819B (en) Depth network lightweight method for remote sensing image target detection
CN113496228B (en) Human body semantic segmentation method based on Res2Net, transUNet and cooperative attention
CN115565034A (en) Infrared small target detection method based on double-current enhanced network
Moon et al. Super resolution based on deep learning technique for constructing digital elevation model
Lopez-Martinez et al. Image restoration with a microscanning imaging system
CN114494021A (en) Image reconstruction method and device, electronic equipment and storage medium

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into 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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20180119

Termination date: 20180717