CN104574268A - Cloud and fog removing method based on non-subsample contourlet transform and non-negative matrix factorization - Google Patents

Cloud and fog removing method based on non-subsample contourlet transform and non-negative matrix factorization Download PDF

Info

Publication number
CN104574268A
CN104574268A CN201410843092.XA CN201410843092A CN104574268A CN 104574268 A CN104574268 A CN 104574268A CN 201410843092 A CN201410843092 A CN 201410843092A CN 104574268 A CN104574268 A CN 104574268A
Authority
CN
China
Prior art keywords
cloud
remote sensing
low
frequency
images
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
CN201410843092.XA
Other languages
Chinese (zh)
Other versions
CN104574268B (en
Inventor
李�杰
王春哲
李学军
孙向阳
李明晶
郭盼
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.)
Changchun University
Original Assignee
Changchun University
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 Changchun University filed Critical Changchun University
Priority to CN201410843092.XA priority Critical patent/CN104574268B/en
Publication of CN104574268A publication Critical patent/CN104574268A/en
Application granted granted Critical
Publication of CN104574268B publication Critical patent/CN104574268B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/73—Deblurring; Sharpening
    • 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
    • G06T2207/10032—Satellite or aerial image; Remote sensing

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a cloud and fog removing method based on non-subsample contourlet transform and non-negative matrix factorization, and relates to the technical fields of image sensing and remote sensing and telemetry. The method solves the problem that cloud layer underlying surface information is lost during the process of cloud and fog removing by the conventional cloud and fog removing method, and comprises the steps of performing NSCT transformation on two remote sensing images containing cloud and fog, so as to obtain respective lower-frequency coefficients and sub-band coefficients in respective directions; performing threshold processing on the lower-frequency coefficients of the two images and estimating the ground information of a cloud layer region, arranging two lower-frequency coefficient image matrixes according to priority, and forming a new matrix and performing non-negative matrix factorization, so as to obtain a lower-frequency coefficient containing the common information of the two images; performing non-negative matrix factorization and fusion processing on the sub-band coefficients in respective directions of the two images, so as to obtain a new sub-band in respective directions; performing NSCT inverse transformation on the new low-frequency coefficient and the sub-band coefficient in respective directions, so as to obtain a cloudless remote sensing image. The method greatly improves and accelerates the cloud and fog removing quality and the algorithm speed.

Description

Defogging method based on non-subsampled contourlet transformation and non-negative matrix decomposition
Technical Field
The invention relates to the technical field of image processing and remote sensing telemetry, in particular to a defogging method based on Non-Subsampled Contourlet Transform (NSCT for short) and Non-negative Matrix Factorization (NMF for short).
Background
The remote sensing image has important significance for national defense and economic construction of a country, and has wide application in aspects such as mastering vegetation distribution, volcanic activity, land disasters and weather, analyzing atmospheric components, planet detection and the like. However, the coverage of the cloud layer in a large number of remote sensing images makes the regions concerned by people blurred, so that the remote sensing platform with lower time resolution cannot obtain the ground information shielded by the cloud layer, which brings much inconvenience to the subsequent processing of the images, such as the inability to perform image recognition in the subsequent processing and the difficulty in ensuring the accuracy in the classification of the images. Therefore, the influence of cloud layers is effectively reduced or removed, and the increase of the utilization rate of remote sensing data is particularly important in the preprocessing of remote sensing images, so that the finding of a method for effectively removing cloud layer shielding has great significance for the processing of the remote sensing images. The cloud and fog removal of the current remote sensing image is mainly based on an image processing technology, and the technology is mature. Image Processing (Image Processing) generally refers to a technique for analyzing an Image with a computer to achieve a desired result.
The cloud layer information contained in the remote sensing image is removed and the information of the underlying surface of the cloud layer is recovered, so that the utilization rate of the remote sensing image is improved.
NSCT is first proposed by cunha a.l. equal to 2006, and shows many unique advantages in multi-resolution analysis of images. The NSCT decomposes the image into different scales through a Non-subsampled Pyramid (NSP for short), and local features of the image can be obtained under different scales; the image is decomposed into sub-bands of each direction by a Non-subsampled directed Filter bank (NSDFB), and the edge information of the image in each direction can be captured. The NSCT can decompose the cloud layer information of the remote sensing image into more fine layer numbers through the two transformations, and the cloud layer is easy to process.
Non-Negative Matrix Factorization (NMF) is a Matrix Factorization algorithm proposed by Lee and Seung and 1999, such that elements of a Matrix after being decomposed do not contain negative values. The characteristics of the multiple images can be extracted by carrying out NMF decomposition on the multiple image matrixes, so that the NMF is widely applied to the fusion field of the multispectral images and the hyperspectral images, and the fused image information is richer.
Disclosure of Invention
The invention provides a cloud and fog removing method based on non-subsampled contourlet transform and non-negative matrix decomposition, aiming at solving the problem that the information of the underlying surface of a cloud layer is lost in the cloud and fog removing process by adopting the conventional cloud and fog removing method.
The cloud haze removing method based on non-subsampled contourlet transform and non-negative matrix decomposition is realized by the following steps:
performing NSCT transformation on two remote sensing images with different cloud layer area distributions to respectively obtain low-frequency coefficients of the two remote sensing images and sub-band coefficients of each image in each direction;
step two, carrying out threshold processing on the low-frequency coefficients of the two remote sensing images obtained in the step one, and respectively obtaining new low-frequency coefficients of the two remote sensing images;
estimating new low-frequency coefficients of the two remote sensing images in the step two to obtain formation information of a cloud layer area in the new low-frequency coefficients; arranging the new low-frequency coefficients of the two remote sensing images according to a row priority criterion to form a matrix;
performing NMF decomposition on the matrix in the step three to obtain a characteristic matrix containing the characteristics of the two remote sensing images, and taking the characteristic matrix as a low-frequency coefficient finally containing the characteristics of the two images;
step five, carrying out NMF decomposition on the sub-band coefficient of each direction of each image in the step one to obtain a new direction coefficient containing the characteristics of two images;
and step six, carrying out NSCT inverse transformation on the low-frequency coefficient obtained in the step four and the sub-band coefficient in each direction obtained in the step five to obtain a clear defogging image.
The invention has the beneficial effects that: according to the method, NSCT transformation is carried out on two remote sensing images containing cloud and mist to obtain respective low-frequency coefficients and sub-band coefficients of the low-frequency coefficients in each direction; respectively carrying out threshold processing on the low-frequency coefficients of the two images and estimating surface feature information of the cloud layer area by using an estimation method, arranging the two estimated low-frequency coefficient image matrixes according to the row priority order to form a new matrix, and carrying out non-negative matrix decomposition on the matrix to obtain a low-frequency coefficient containing common information of the two images; carrying out nonnegative matrix decomposition and fusion processing on the coefficients of the two sub-bands in each direction to obtain new sub-bands in each direction; and (4) applying the new low-frequency and all-direction sub-band coefficients to carry out NSCT inverse transformation to obtain a cloud-free remote sensing image. The cloud defogging method for sampling the non-subsampled contourlet transform and the non-negative matrix decomposition effectively improves the quality of cloud defogging and the speed of an algorithm, solves the problem that a cloud-free image is difficult to obtain in the current multivariate data fusion cloud defogging method, can realize the cloud defogging of the remote sensing images as long as the cloud layer areas of the two remote sensing images are distributed differently, and provides a new idea for the field of cloud defogging of the remote sensing images.
Drawings
FIG. 1 is a flow chart of the cloud removal method based on non-subsampled contourlet transform and non-negative matrix factorization according to the present invention;
FIG. 2 is a corresponding low-frequency and high-frequency image obtained by NSP decomposition of a layer number of a test image Zone Plate based on non-subsampled contourlet transform and a non-negative matrix decomposition method according to the present invention; fig. 2a, fig. 2b and fig. 2c are schematic diagrams of a Zone Plate original image, a layer of low-frequency subbands decomposed by NSP, and a layer of high-frequency subbands decomposed by NSP, respectively;
fig. 3 is a diagram illustrating the effect of the cloud defogging method based on the non-subsampled contourlet transform and the non-negative matrix factorization method according to the present invention, where fig. 3a and 3b are two original cloud images A, B, fig. 3c and 3d are low frequency coefficients of fig. 3a and 3b, respectively, fig. 3e and 3f are low frequency coefficient estimates of fig. 3c and 3d, and fig. 3g is an image after cloud removal.
Detailed Description
The present embodiment is described with reference to fig. 1 to 3, and is based on a non-downsampling contourlet transform and non-negative matrix decomposition defogging method, which is implemented by the following steps:
firstly, performing NSCT transformation on two remote sensing cloud pictures A and B with different cloud layer area distributions to obtain a low-frequency coefficient of the cloud picture A and sub-band coefficients of the cloud picture A in all directions, and a low-frequency coefficient of the cloud picture B and sub-band coefficients of the cloud picture B in all directions; combining fig. 3c and fig. 3d in fig. 3;
step two, carrying out threshold processing on the low-frequency coefficients of the images obtained in the step one and estimating the formation information of the cloud layer region in the low-frequency coefficients by using the existing estimation method (as shown in formula 1), as shown in fig. 3e and fig. 3 f;
thirdly, arranging the low-frequency coefficients of the cloud picture A and the cloud picture B in the second step according to a row priority criterion to form a new matrix;
performing NMF decomposition on the matrix in the third step to obtain a characteristic matrix containing the characteristics of the cloud picture A and the cloud picture B, and taking the characteristic matrix as a final low-frequency coefficient;
step five, performing non-negative matrix decomposition and fusion on the respective direction coefficients in the cloud picture A and the cloud picture B by adopting an NMF decomposition method to obtain new direction coefficients;
and step six, performing NSCT inverse transformation on the low-frequency coefficients and the coefficients in each direction in the step four and the step five to obtain a clear cloud and fog removing image, as shown in fig. 3 g.
The estimation method employed in the present embodiment is expressed by equation (1):
S ^ ( i , j ) = S ( i , j ) - ( mean haze - mean clear ) - - - ( 1 )
wherein,the low-frequency coefficient after estimation by equation (1) is used, S (i, j) is the low-frequency coefficient before estimation, meanhazeIs NSCT scoreMean of the cloud regions of the solutionclearIs the mean value of the ground feature information of the non-cloud region decomposed by NSCT.
The NSCT conversion according to the present embodiment is as follows:
assuming that NSCT transformation is carried out on an image matrix with M rows and N columns, firstly, NSP is used for carrying out multi-scale decomposition to obtain low-frequency coefficients and high-frequency coefficients under different scales, and then NSDFB is used for carrying out directional decomposition on the high-frequency coefficients to obtain sub-band coefficients in all directions. Fig. 2a in fig. 2 is a Zone Plate original image, fig. 2b and fig. 2c are corresponding low and high frequency images obtained by performing NSP decomposition on a Zone Plate with one layer, wherein the NMF decomposition is as follows:
assuming a low-frequency coefficient V of a two-image matrix of M rows and N columns1And V2Is arranged according to the rule of line priority to form a matrix V 'of M multiplied by N lines and 1 column'1And V'2V 'is'1And V'2Constitute a new matrix V ═ V'1,V′2]Performing NMF decomposition on the matrix V to obtain V ≈ WMN×1H1×1Will matrix WMN×1Is changed into a matrix W of M rows and N columns'M×NW 'mentioned'M×NI.e. a low frequency coefficient containing the common features of the two images.
And finally, carrying out NSCT inverse transformation on the obtained new low-frequency coefficient and the fused sub-band coefficients in all directions to obtain a clear cloud-removed image.

Claims (2)

1. The cloud haze removing method based on non-subsampled contourlet transform and non-negative matrix decomposition is characterized by comprising the following steps of:
performing NSCT transformation on two remote sensing images with different cloud layer area distributions to respectively obtain low-frequency coefficients of the two remote sensing images and sub-band coefficients of each image in each direction;
step two, carrying out threshold processing on the low-frequency coefficients of the two remote sensing images obtained in the step one, and respectively obtaining new low-frequency coefficients of the two remote sensing images;
estimating new low-frequency coefficients of the two remote sensing images in the step two to obtain formation information of a cloud layer area in the new low-frequency coefficients; arranging the new low-frequency coefficients of the two remote sensing images according to a row priority criterion to form a matrix;
performing NMF decomposition on the matrix in the step three to obtain a characteristic matrix containing the characteristics of the two remote sensing images, and taking the characteristic matrix as a low-frequency coefficient finally containing the characteristics of the two images;
step five, carrying out NMF decomposition on the sub-band coefficient of each direction of each image in the step one to obtain a new direction coefficient containing the characteristics of two images;
and step six, carrying out NSCT inverse transformation on the low-frequency coefficient obtained in the step four and the sub-band coefficient in each direction obtained in the step five to obtain a clear defogging image.
2. The method for defogging based on non-subsampled contourlet transform and non-negative matrix factorization of claim 1, wherein the estimation method adopted in the third step is to obtain a new low frequency coefficient, and the formula is as follows:
S ^ ( i , j ) = S ( i , j ) - ( mean haze - mean clear )
in the formula,for the new low-frequency coefficient estimated by the formula, S (i, j) is the low-frequency coefficient of the two remote sensing images，meanhazeIs the mean of the cloud region of the NSCT decompositionclearIs the mean value of the ground feature information of the non-cloud region decomposed by NSCT.
CN201410843092.XA 2014-12-30 2014-12-30 Cloud and mist method is gone based on non-down sampling contourlet transform and Non-negative Matrix Factorization Expired - Fee Related CN104574268B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410843092.XA CN104574268B (en) 2014-12-30 2014-12-30 Cloud and mist method is gone based on non-down sampling contourlet transform and Non-negative Matrix Factorization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410843092.XA CN104574268B (en) 2014-12-30 2014-12-30 Cloud and mist method is gone based on non-down sampling contourlet transform and Non-negative Matrix Factorization

Publications (2)

Publication Number Publication Date
CN104574268A true CN104574268A (en) 2015-04-29
CN104574268B CN104574268B (en) 2017-06-16

Family

ID=53090255

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410843092.XA Expired - Fee Related CN104574268B (en) 2014-12-30 2014-12-30 Cloud and mist method is gone based on non-down sampling contourlet transform and Non-negative Matrix Factorization

Country Status (1)

Country Link
CN (1) CN104574268B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416739A (en) * 2018-01-16 2018-08-17 辽宁师范大学 Traffic image defogging method based on non-down sampling contourlet and Markov random field

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101303764A (en) * 2008-05-16 2008-11-12 西安电子科技大学 Method for self-adaption amalgamation of multi-sensor image based on non-lower sampling profile wave
US20120224783A1 (en) * 2011-03-01 2012-09-06 Lawrence Shao-Hsien Chen Methods and Systems for Document-Image Correction
CN103632354A (en) * 2012-08-24 2014-03-12 西安元朔科技有限公司 Multi focus image fusion method based on NSCT scale product

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101303764A (en) * 2008-05-16 2008-11-12 西安电子科技大学 Method for self-adaption amalgamation of multi-sensor image based on non-lower sampling profile wave
US20120224783A1 (en) * 2011-03-01 2012-09-06 Lawrence Shao-Hsien Chen Methods and Systems for Document-Image Correction
CN103632354A (en) * 2012-08-24 2014-03-12 西安元朔科技有限公司 Multi focus image fusion method based on NSCT scale product

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
冯鑫等: "基于改进非下采样轮廓波的图像融合算法", 《农业机械学报》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416739A (en) * 2018-01-16 2018-08-17 辽宁师范大学 Traffic image defogging method based on non-down sampling contourlet and Markov random field
CN108416739B (en) * 2018-01-16 2021-09-24 辽宁师范大学 Traffic image defogging method based on contour wave and Markov random field

Also Published As

Publication number Publication date
CN104574268B (en) 2017-06-16

Similar Documents

Publication Publication Date Title
CN108573276B (en) Change detection method based on high-resolution remote sensing image
CN109272455B (en) Image defogging method based on weak supervision generation countermeasure network
Yang et al. Image denoising using nonsubsampled shearlet transform and twin support vector machines
CN105809198B (en) SAR image target recognition method based on depth confidence network
CN106408530A (en) Sparse and low-rank matrix approximation-based hyperspectral image restoration method
CN109658351B (en) Hyperspectral image denoising method combining L0 gradient constraint and local low-rank matrix recovery
CN107977661B (en) Region-of-interest detection method based on FCN and low-rank sparse decomposition
CN108875645B (en) Face recognition method under complex illumination condition of underground coal mine
CN102867187B (en) NSST (NonsubsampledShearlet Transform) domain MRF (Markov Random Field) and adaptive threshold fused remote sensing image change detection method
CN110503613A (en) Based on the empty convolutional neural networks of cascade towards removing rain based on single image method
CN103077508A (en) Transform domain non local and minimum mean square error-based SAR (Synthetic Aperture Radar) image denoising method
CN105096272A (en) De-hazing method based on dual-tree complex wavelet
US20240135704A1 (en) Computer Vision Systems and Methods for Detecting and Aligning Land Property Boundaries on Aerial Imagery
CN111126185B (en) Deep learning vehicle target recognition method for road gate scene
Jang et al. Object classification using CNN for video traffic detection system
CN113421198B (en) Hyperspectral image denoising method based on subspace non-local low-rank tensor decomposition
Yang et al. Detail-aware near infrared and visible fusion with multi-order hyper-Laplacian priors
CN104574268B (en) Cloud and mist method is gone based on non-down sampling contourlet transform and Non-negative Matrix Factorization
CN116958547A (en) Remote sensing image cloud removal method based on generation countermeasure network
Biswas et al. Remote sensing image fusion using PCNN model parameter estimation by gamma distribution in shearlet domain
CN104809733A (en) Ancient building wall polluted inscription character image edge extraction method
CN104751470A (en) Image quick-matching method
Deshmukh et al. Image enhancement techniques for fingerprint identification
CN103345739B (en) A kind of high-resolution remote sensing image building area index calculation method based on texture
CN115330747A (en) DPS-Net deep learning-based rice plant counting, positioning and size estimation method

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
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: 20170616

Termination date: 20181230