CN106023112A - Image rain removing method and system based on wavelet analysis - Google Patents
Image rain removing method and system based on wavelet analysis Download PDFInfo
- Publication number
- CN106023112A CN106023112A CN201610347657.4A CN201610347657A CN106023112A CN 106023112 A CN106023112 A CN 106023112A CN 201610347657 A CN201610347657 A CN 201610347657A CN 106023112 A CN106023112 A CN 106023112A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- rain
- wavelet
- fusion
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000004927 fusion Effects 0.000 claims abstract description 71
- 239000011159 matrix material Substances 0.000 claims abstract description 64
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 40
- 238000001914 filtration Methods 0.000 claims abstract description 22
- 238000004422 calculation algorithm Methods 0.000 claims description 47
- 238000004040 coloring Methods 0.000 claims description 20
- 230000002146 bilateral effect Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 6
- 230000000694 effects Effects 0.000 description 16
- 230000006870 function Effects 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 8
- 238000001514 detection method Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 208000031481 Pathologic Constriction Diseases 0.000 description 2
- 108010076504 Protein Sorting Signals Proteins 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 210000001215 vagina Anatomy 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000000686 essence Substances 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000000155 melt Substances 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- G06T5/70—
-
- G06T5/77—
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20028—Bilateral filtering
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention relates to an image rain removing method and system based on wavelet analysis, and belongs to the technical field of image processing. The method comprises the steps that a) according to wavelet analysis, layer decomposition is carried out on a video frame image, image information of decomposed layers is analyzed, and a layer including background and color information, a layer including raindrop noise and a layer including image texture and object edge information are obtained; b) double-side filtering side-maintaining denoising is carried out on the layer including the raindrop noise; and c) a fusion coefficient matrix is calculated, wavelet fusion is carried out on the layer including background and color information, the layer including raindrop noise and the layer including image texture and object edge information according to the fusion coefficient matrix, and according to a fusion result, the image is reconstructed to obtain a rain-removed image. According to the method and system, interference of dynamic features is avoided, and the raindrops can be removed more accurately and more effectively.
Description
Technical field
The invention belongs to technical field of image processing, particularly relate to a kind of image rain removing method based on wavelet analysis and be
System.
Background technology
Image imaging is had a great impact by rain, and image image blur and information can be caused to cover, and its direct result is to regard
Frequently the definition of image declines, and the digitized processing of video image also can suffer from this and hydraulic performance decline.To polluted by raindrop
Video image carries out repair process and is conducive to the further process of image, including target detection based on image, identify, follow the trail of,
The performance of the technology such as segmentation and monitoring improves.And video image based on wavelet analysis goes rain technology in modern military, traffic
And the field such as security monitoring all has wide practical use.
About the research of raindrop characteristic in video image by the extensive concern of international academic community, go the research of rain algorithm
Also from (Starik S, Werman M.Simulation of rainin videos [C] Proceeding such as Starik in 2003
Of Texture Workshop, ICCV.Nice, France:2003,2:406-409) median method that proposes starts to have obtained fast
The development of speed, the method for process has been no longer limited to the simplest median calculation, degree of bias calculating, K mean cluster, Kalman
A lot of method such as filtering, dictionary learning and sparse coding, guiding filtering, interframe luminance difference, HSV space, optical flow method, motion segmentation etc.
The most gradually starting to apply in video image in the algorithm of raindrop detection and removal, the effect that raindrop are removed the most gradually is enhanced.
The interframe luminance difference that Garg etc. propose to utilize raindrop to bring at first carries out raindrop initial survey, then utilizes rectilinearity and the direction of raindrop
Consistent feature is screened further, and the pixel intensity finally according to front and back's frame removes raindrop impact, can preferably meet raindrop
Do not cover the raindrop detection in the case of sequential frame image and removal;The influence of color that raindrop bring to pixel is considered by Zhang etc.
Including, thus improve the accuracy of raindrop detection, improve and based on brightness flop go the application on coloured image of the rain algorithm
Effect;Brightness impact and the influence of color of raindrop are applied in the algorithm by Liu etc. simultaneously, detect raindrop with two frames and remove;
Tripathi etc. first study the probabilistic statistical characteristics of raindrop pixel intensity change, then utilize the symmetry that raindrop pixel intensity changes
Property realize raindrop detection, be based only upon time domain and additionally consider that effect when affecting of locus is incomplete same;Kang etc. are first
Utilize bilateral filtering that rain figure is divided into HFS and low frequency part, and process further to HFS obtain non-rain composition,
Rain figure is obtained in conjunction with low frequency part;Huang etc. carry out image segmentation first with context constraint, and utilize context-aware
Carry out single width image based on wavelet analysis and remove rain, and propose innovatory algorithm on this basis, first literary composition has been used super complete
HFS is processed by standby dictionary.
Particularly recent years, video image goes rain technology to become new study hotspot.How to ensure high robust
On the premise of improve and go accuracy rate and the real-time of rain, be the current video image focus of going to rain field.The algorithm that presently, there are
In, being applied to the detection of static scene video raindrop has more ripe achievement in research with the algorithm removed, but applies dynamically
Time on video in scene, algorithm it is considered that video occurs the interference that moving object is brought, for raindrop characteristic district
Do not spend the highest moving object and be unable to reach preferable Detection results.Automatically lead additionally, process in real time what multinomial technology was applied
The occasions such as boat system, safety monitoring system there is the biggest application demand.These application scenarios generally require timely everywhere
Reason result, feeds back to user, and the delayed user of being likely to result in of Video processing does the judgement made mistake.Therefore raindrop inspection in video
Survey and not only need to improve precision with removing, it is similarly desirable to increase processing speed, and need to find optimal balance point therebetween.But
It is current algorithm processing speed and the precision that also cannot take into account various scene, it is achieved the real-time removing rain algorithm is current research face
To an important topic.
In sum, existing image go rain technology to suffer a disadvantage in that existing image goes rain algorithm to be substantially base
Carry out rain in pixel intensity and raindrop morphological characteristic, go rain effect less desirable;Meanwhile, existing image goes rain technology pair
Going rain effect less desirable in dynamic scene, algorithm complex and algorithm real-time can not be taken into account well.
Summary of the invention
The invention provides a kind of image rain removing method based on wavelet analysis and system, it is intended to solve existing image and go
Rain technology goes rain effect less desirable for dynamic scene, and the technology that algorithm complex and algorithm real-time can not be taken into account
Problem.
The present invention is achieved in that a kind of image rain removing method based on wavelet analysis, including:
Step a: according to wavelet analysis, video frame images carried out block layer decomposition, and analyze the image letter of described exploded view layer
Breath, respectively obtains and comprises background and the figure layer of colouring information, the figure layer comprising rain cell noise and comprise image texture and object
The figure layer of marginal information;
Step b: the described figure layer comprising rain cell noise is carried out bilateral filtering and protects limit denoising;
Step c: calculate fusion coefficients matrix, comprises background and colouring information according to described fusion coefficients matrix to described
Figure layer, the figure layer comprising rain cell noise and the figure layer comprising image texture and object edge information carry out Wavelet Fusion respectively,
And carry out image reconstruction according to fusion results and obtain rain image.
The technical scheme that the embodiment of the present invention is taked also includes: in described step a, described according to wavelet analysis to video
Two field picture carries out block layer decomposition particularly as follows: described video frame images is decomposed into ten layers by Malla algorithm based on wavelet analysis,
The decomposition formula of Malla algorithm is:
Ci=HcHrCi-1
In above-mentioned formula, H and G is the coefficient matrix of scaling function φ (x) and wavelet function ψ (x) respectively, Ci、WithThe low frequency part of correspondence image Ci-1, the HFS of vertical direction, the HFS of vertical direction respectively
And the HFS of diagonal.
The technical scheme that the embodiment of the present invention is taked also includes: described step a also includes: detects and comprises rain cell noise
Figure layer;The described figure layer comprising rain cell noise is the second to the 4th high frequency coefficient figure layer.
The technical scheme that the embodiment of the present invention is taked also includes: in described step c, described calculating fusion coefficients matrix
Calculation is: defines raindrop pollution level coefficient according to the light characteristic of raindrop, melts according to raindrop pollution level coefficient calculations
Syzygy matrix number;Described calculating fusion coefficients matrix is particularly as follows: make raindrop pollution level coefficient S=G × E, and wherein, G is local
Gradient, E is local energy, and two parameters of partial gradient and local energy are multiplied and obtain new variable S, the S the biggest pollution of value
The most serious;Being normalized s-matrix and obtain S ', coefficient matrix and S ' matrix are for being weighted image reconstruction algorithm
Optimize and obtain fusion coefficients matrix.
The technical scheme that the embodiment of the present invention is taked also includes: described partial gradient G is defined as:
In above-mentioned formula, (i, j) (i, (i, j) both horizontally and vertically, M and N divides Δ xf j) to be respectively point with Δ yf
Wei the length of side in region;
The local energy E of described pixel is expressed as:
Another technical scheme that the embodiment of the present invention is taked is: a kind of image based on wavelet analysis goes rain system, including
Wavelet decomposition module, image denoising module and wavelet fusion module;Described wavelet decomposition module is used for according to wavelet analysis regarding
Frequently two field picture carries out block layer decomposition, analyzes the image information of described exploded view layer, respectively obtains and comprises background and colouring information
Figure layer, the figure layer comprising rain cell noise and comprise the figure layer of image texture and object edge information;Described image denoising module
Limit denoising is protected for the described figure layer comprising rain cell noise being carried out bilateral filtering;Described wavelet fusion module is used for calculating
Fusion coefficients matrix, the figure layer comprising background and colouring information to described according to described fusion coefficients matrix, comprises rain cell noise
Figure layer and the figure layer that comprises image texture and object edge information carry out Wavelet Fusion respectively, and carry out according to fusion results
Image reconstruction obtains rain image.
The technical scheme that the embodiment of the present invention is taked also includes: described wavelet decomposition module according to wavelet analysis to frame of video
Image carries out block layer decomposition particularly as follows: described video frame images is decomposed into ten layers by Malla algorithm based on wavelet analysis,
The decomposition formula of Malla algorithm is:
Ci=HcHrCi-1
In above-mentioned formula, H and G is the coefficient matrix of scaling function φ (x) and wavelet function ψ (x) respectively, Ci、WithCorrespondence image C respectivelyi-1Low frequency part, the HFS of vertical direction, vertical direction HFS with
And the HFS of diagonal.
The technical scheme that the embodiment of the present invention is taked also includes: described image denoising module is additionally operable to detect and comprises raindrop
The figure layer of noise;The described figure layer comprising rain cell noise is the second to the 4th high frequency coefficient figure layer.
The technical scheme that the embodiment of the present invention is taked also includes: described wavelet fusion module calculates the meter of fusion coefficients matrix
Calculation mode is: defines raindrop pollution level coefficient according to the light characteristic of raindrop, merges according to raindrop pollution level coefficient calculations
Coefficient matrix;Described calculating fusion coefficients matrix is particularly as follows: make raindrop pollution level coefficient S=G × E, and wherein, G is local ladder
Degree, E is local energy, and two parameters of partial gradient and local energy are multiplied and obtain new variable S, the S the biggest pollution of value more
Seriously;Being normalized s-matrix and obtain S ', coefficient matrix and S ' matrix are excellent for being weighted image reconstruction algorithm
Change and obtain fusion coefficients matrix.
The technical scheme that the embodiment of the present invention is taked also includes: described partial gradient G is defined as:
In above-mentioned formula, (i, j) (i, (i, j) both horizontally and vertically, M and N divides Δ xf j) to be respectively point with Δ yf
Wei the length of side in region;
The local energy E of described pixel is expressed as:
Relative to prior art, what the present invention produced has the beneficial effects that: the embodiment of the present invention based on wavelet analysis
Image rain removing method and system use the method for small echo multi-level decomposition and Wavelet Fusion to differentiate the aspect at rain cell noise place, to containing
The figure layer having rain cell noise carries out bilateral filtering and protects limit denoising, and according to the rule of raindrop effect definition Wavelet Fusion
Then, specific aspect is carried out Wavelet Fusion to reach the purpose that raindrop are removed;The enforcement of the present invention can be avoided by dynamic
The interference of step response, removes raindrop more accurately and effectively, improves the range of rain algorithm, in the situation that the force of rain is the biggest
Rain also can have and good go effect, improve the real-time of rain algorithm.
Accompanying drawing explanation
Fig. 1 is the flow chart of the image rain removing method based on wavelet analysis of the embodiment of the present invention;
Fig. 2 is wavelet decomposition schematic diagram;Wherein, Fig. 2 (a) is original image, and Fig. 2 (b) is the low-frequency information of image, Fig. 2
C () is to the high-frequency structure that Fig. 2 (l) is that the tenth of image arrives ground floor;
Fig. 3 is Wavelet decomposing and recomposing flow chart;
Fig. 4 is the structural representation that the image based on wavelet analysis of the embodiment of the present invention goes rain system.
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, right
The present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, and
It is not used in the restriction present invention.
Refer to Fig. 1, be the flow chart of the image rain removing method based on wavelet analysis of the embodiment of the present invention.The present invention is real
The image rain removing method based on wavelet analysis executing example comprises the following steps:
Step S100: input video two field picture;
Step S200: image carried out small echo multilamellar decomposition according to wavelet analysis, and analyze the image information of each figure layer,
Respectively obtain and comprise background and the figure layer of colouring information, the figure layer comprising rain cell noise and comprise image texture and object edge
The figure layer of information;
In step s 200, wavelet analysis has good temporal frequency locating features, it is possible to signal decomposition is become multiple
There is the frequency sub-band of different frequency sub-band, frequency characteristic and directional characteristic, so wavelet analysis is also referred to as school microscop.Schemed
As decomposing and the inspiration of restructing algorithm, Mallat propose Malla algorithm based on wavelet analysis, the i.e. many resolution decomposition of image and
Reconstruct pyramid algorith.Restructing algorithm is the inverse process of decomposition algorithm, just can recover original signal sequence through liftering.The projection f in space (x, y) can be used to represent two dimensional image signal:
If the coefficient matrix of scaling function φ (x) and wavelet function ψ (x) is that H and G, Malla algorithm must decompose public affairs respectively
Formula is:
In formula (4), Ci、WithCorrespondence image C respectivelyi-1Low frequency part, the HFS of vertical direction,
The HFS of vertical direction and the HFS of diagonal.The restructing algorithm of Malla algorithm can be expressed as:
In formula (5), H*、G*It is respectively the inverse matrix of H, G.
Described figure layer includes comprising background and the figure layer of colouring information, the figure layer comprising rain cell noise and comprising image stricture of vagina
Reason and the figure layer of object edge information.Image can be carried out linearly by wavelet analysis respectively, high pass and low-pass filtering.In original graph
Carry out the low-pass filtering of row and column on picture, the low frequency coefficient approximation component C of ground floor can be obtained1, it include image background with
Colouring information.Carry out the high-pass filtering of row and column on the original image, horizontal high-frequent detail coefficients can be obtainedVertical high
Frequently detail coefficientsWith diagonal high frequency detail coefficientThey include that the image texture of different directions is believed with object edge
Breath.Above-mentioned boundary operation repeats at C1Operate on low frequency component, each frequency component of the corresponding second layer can be obtained
C2、WithIf above-mentioned boundary operation repeats to carry out at m-1 layer, it is possible to obtain Cm、With
The frequency of rain cell noise is the highest, and image texture is taller with the noise ratio raindrop of object edge, image background with
The frequency of colouring information is the lowest.So, by decomposing based on wavelet analysis multilamellar by raindrop pollution image, it appeared that rain cell noise
Sound should be comprised in the high frequency coefficient part of low figure layer, and bigger Decomposition order is usually used to guarantee the thin of image after rain
Joint information.In embodiments of the present invention, the Decomposition order that image carries out small echo multilamellar decomposition is ten layers.Concrete as in figure 2 it is shown,
It it is wavelet decomposition schematic diagram;Wherein, Fig. 2 (a) is original image, and Fig. 2 (b) is the low-frequency information of image, Fig. 2 (c) to Fig. 2 (l)
Be image the tenth to the high-frequency structure of ground floor.From analyzing, it is high that most rain cell noise concentrates on second to the 4th
Frequently, on coefficient figure layer, the 5th to the tenth high frequency coefficient figure layer then comprises overwhelming majority image background and colouring information, ground floor
Comprise image texture and object edge information.
Step S300: detect the figure layer comprising rain cell noise, and the figure layer comprising rain cell noise is carried out bilateral filtering
Protect limit denoising;
In step S300, bilateral filtering is a kind of wave filter that can protect limit denoising, why can reach this denoising
Effect, is because it and is made up of two functions.One function is to be determined filter coefficient by geometric space distance, and another is by picture
Element difference determines filter coefficient.Bilateral filtering considers the difference of spatial domain and codomain simultaneously, therefore is capable of protecting limit and retains
Noise remove, follow-up image co-registration is better achieved.In two-sided filter, the value of output pixel depends on neighborhood picture
The weighted array of the value of element.(i, j, k l) depend on defining taking advantage of of territory core (formula (7)) and codomain core (formula (8)) weight coefficient w
Long-pending:
Step S400: define raindrop pollution level coefficient, according to raindrop pollution level coefficient according to the light characteristic of raindrop
Calculate fusion coefficients matrix, according to the fusion coefficients matrix figure to comprising the background figure layer with colouring information, comprising rain cell noise
Layer and the figure layer comprising image texture and object edge information carry out Wavelet Fusion respectively, and carry out image by fusion results
Reconstruct obtains rain image;
In step S400, the tonal gradation of the pixel owing to being covered by raindrop is bigger than background gray scale, can produce edge effect
Should, then, partial gradient can be used for measuring the change of gray scale, and partial gradient is defined as:
In formula (10), (i, j) (i, (i, j) both horizontally and vertically, M and N is respectively Δ xf j) to be respectively point with Δ yf
The length of side in region.Owing to raindrop brightness is basically unchanged, raindrop pixel has higher and more stable energy, the local of pixel
Energy can be expressed as:
Make two parameters of raindrop pollution level coefficient S=G × E, i.e. partial gradient and local energy be multiplied and obtain one newly
Variable S, the S the biggest pollution of value the most serious.Being normalized s-matrix and obtain S ', coefficient matrix and S ' matrix are for right
Image reconstruction algorithm is weighted optimization and obtains fusion coefficients matrix, believes with color comprising background according to fusion coefficients matrix
Breath, rain cell noise and image texture carry out Wavelet Fusion, finally by three partial fusion knots respectively with the figure layer of object edge information
Fruit carries out image reconstruction and obtains rain image, and the color of image and details so can be made undistorted.In embodiments of the present invention,
In order to remove raindrop, for the figure layer at rain cell noise place, the weights of fusion coefficients matrix should be less than 1;And for not by
The figure layer that raindrop pollute, the weights of fusion coefficients matrix are set as more than 1;Concrete as it is shown on figure 3, be Wavelet decomposing and recomposing flow process
Figure.
After the figure layer carrying out different images information carries out Wavelet Fusion respectively, then to 9 video frame images of continuous print
Carrying out image co-registration, the rain image that goes after fusion replaces middle 5th two field picture, to reach finally to go rain effect.
Step S500: output video frame image.
Refer to Fig. 4, be the image based on wavelet analysis of the embodiment of the present invention structural representation that goes rain system.This
The image based on wavelet analysis of bright embodiment goes rain system to include image input module, wavelet decomposition module, image denoising mould
Block, wavelet fusion module and image output module;Specifically:
Image input module is used for input video two field picture;
Wavelet decomposition module for carrying out small echo multilamellar decomposition according to wavelet analysis to image, and analyzes the figure of each figure layer
As information, respectively obtain comprise background and the figure layer of colouring information, the figure layer comprising rain cell noise and comprise image texture with
The figure layer of object edge information;Wherein, wavelet analysis has good temporal frequency locating features, it is possible to signal decomposition is become many
The individual frequency sub-band with different frequency sub-band, frequency characteristic and directional characteristic, so wavelet analysis is also referred to as school microscop.It is subject to
Picture breakdown and the inspiration of restructing algorithm, Mallat proposes Malla algorithm based on wavelet analysis, the i.e. many resolution decomposition of image
With reconstruct pyramid algorith.Restructing algorithm is the inverse process of decomposition algorithm, just can recover original signal sequence through liftering
Row.The projection f in space (x, y) can be used to represent two dimensional image signal:
If the coefficient matrix of scaling function φ (x) and wavelet function ψ (x) is that H and G, Malla algorithm must decompose public affairs respectively
Formula is:
In formula (4), Ci、WithCorrespondence image C respectivelyi-1Low frequency part, the radio-frequency head of vertical direction
Point, the HFS of vertical direction and the HFS of diagonal.The restructing algorithm of Malla algorithm can be expressed as:
In formula (5), H*、G*It is respectively the inverse matrix of H, G.
Described figure layer includes comprising background and the figure layer of colouring information, the figure layer comprising rain cell noise and comprising image stricture of vagina
Reason and the figure layer of object edge information.Image can be carried out linearly by wavelet analysis respectively, high pass and low-pass filtering.In original graph
Carry out the low-pass filtering of row and column on picture, the low frequency coefficient approximation component C of ground floor can be obtained1, it include image background with
Colouring information.Carry out the high-pass filtering of row and column on the original image, horizontal high-frequent detail coefficients can be obtainedVertical high
Frequently detail coefficientsWith diagonal high frequency detail coefficientThey include that the image texture of different directions is believed with object edge
Breath.Above-mentioned boundary operation repeats at C1Operate on low frequency component, each frequency component of the corresponding second layer can be obtained
C2、WithIf above-mentioned boundary operation repeats to carry out at m-1 layer, it is possible to obtain Cm、With
The frequency of rain cell noise is the highest, and image texture is taller with the noise ratio raindrop of object edge, image background with
The frequency of colouring information is the lowest.So, by decomposing based on wavelet analysis multilamellar by raindrop pollution image, it appeared that rain cell noise
Sound should be comprised in the high frequency coefficient part of low figure layer, and bigger Decomposition order is usually used to guarantee the thin of image after rain
Joint information.In embodiments of the present invention, the Decomposition order that image carries out small echo multilamellar decomposition is ten layers.Concrete as in figure 2 it is shown,
It it is wavelet decomposition schematic diagram;Wherein, Fig. 2 (a) is original image, and Fig. 2 (b) is the low-frequency information of image, Fig. 2 (c) to Fig. 2 (l)
Be image the tenth to the high-frequency structure of ground floor.From analyzing, it is high that most rain cell noise concentrates on second to the 4th
Frequently, on coefficient figure layer, the 5th to the tenth high frequency coefficient figure layer then comprises overwhelming majority image background and colouring information, ground floor
Comprise image texture and object edge information.
Image denoising module is for detecting the figure layer comprising rain cell noise, and carries out the figure layer comprising rain cell noise double
Limit denoising is protected in limit filtering;Wherein, bilateral filtering is a kind of wave filter that can protect limit denoising, why can reach this and go
Make an uproar effect, be because it and be made up of two functions.One function is to be determined filter coefficient by geometric space distance, another by
Pixel value difference determines filter coefficient.Bilateral filtering considers the difference of spatial domain and codomain simultaneously, therefore is capable of Bao Bianbao
The noise remove stayed.In two-sided filter, the value of output pixel depends on the weighted array of the value of neighborhood territory pixel.Weight coefficient w
(i, j, k, l) depend on definition territory core (formula (7)) and the product of codomain core (formula (8)):
Wavelet fusion module defines raindrop pollution level coefficient for the light characteristic according to raindrop, pollutes journey according to raindrop
Degree coefficient calculations fusion coefficients matrix, according to fusion coefficients matrix to comprising the background figure layer with colouring information, comprising rain cell noise
The figure layer of sound and the figure layer comprising image texture and object edge information carry out Wavelet Fusion respectively, and are entered by fusion results
Row image reconstruction obtains rain image;Wherein, the tonal gradation of the pixel owing to being covered by raindrop is bigger than background gray scale, can produce
Edge effect, then, partial gradient can be used for measuring the change of gray scale, and partial gradient is defined as:
In formula (10), (i, j) (i, (i, j) both horizontally and vertically, M and N is respectively Δ xf j) to be respectively point with Δ yf
The length of side in region.Owing to raindrop brightness is basically unchanged, raindrop pixel has higher and more stable energy, the local of pixel
Energy can be expressed as:
Make two parameters of raindrop pollution level coefficient S=G × E, i.e. partial gradient and local energy be multiplied and obtain one newly
Variable S, the S the biggest pollution of value the most serious.Being normalized s-matrix and obtain S ', coefficient matrix and S ' matrix are for right
Image reconstruction algorithm is weighted optimization and obtains fusion coefficients matrix, believes with color comprising background according to fusion coefficients matrix
Breath, rain cell noise and image texture carry out Wavelet Fusion, finally by three partial fusion knots respectively with the figure layer of object edge information
Fruit carries out image reconstruction and obtains rain image, and the color of image and details so can be made undistorted.In embodiments of the present invention,
In order to remove raindrop, for the figure layer at rain cell noise place, the weights of fusion coefficients matrix should be less than 1;And for not by
The figure layer that raindrop pollute, the weights of fusion coefficients matrix are set as more than 1;Concrete as it is shown on figure 3, be Wavelet decomposing and recomposing flow process
Figure.
After the figure layer carrying out different images information carries out Wavelet Fusion respectively, then to 9 video frame images of continuous print
Carrying out image co-registration, the rain image that goes after fusion replaces middle 5th two field picture, to reach finally to go rain effect.
Image output module is used for output video frame image.
The image rain removing method based on wavelet analysis of the embodiment of the present invention and system use small echo multi-level decomposition and small echo
The method merged differentiates the aspect at rain cell noise place, the figure layer containing rain cell noise carries out bilateral filtering and protects at the denoising of limit
Reason, and according to the rule of raindrop effect definition Wavelet Fusion, specific aspect carries out Wavelet Fusion to reach rain
Drip the purpose removed;The enforcement of the present invention can be avoided being disturbed by dynamic characteristic, removes raindrop more accurately and effectively, carries
The high range removing rain algorithm, also can have in the case of the force of rain is very big and good go rain effect, improve rain algorithm
Real-time.
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all essences in the present invention
Any amendment, equivalent and the improvement etc. made within god and principle, should be included within the scope of the present invention.
Claims (10)
1. an image rain removing method based on wavelet analysis, it is characterised in that including:
Step a: video frame images carried out block layer decomposition according to wavelet analysis, and analyze the image information of described exploded view layer,
Respectively obtain and comprise background and the figure layer of colouring information, the figure layer comprising rain cell noise and comprise image texture and object edge
The figure layer of information;
Step b: the described figure layer comprising rain cell noise is carried out bilateral filtering and protects limit denoising;
Step c: calculate fusion coefficients matrix, according to described fusion coefficients matrix to the described figure comprising background and colouring information
Layer, the figure layer comprising rain cell noise and the figure layer comprising image texture and object edge information carry out Wavelet Fusion respectively, and
Carry out image reconstruction according to fusion results and obtain rain image.
Image rain removing method based on wavelet analysis the most according to claim 1, it is characterised in that in described step a,
Described according to wavelet analysis, video frame images is carried out block layer decomposition particularly as follows: Malla algorithm based on wavelet analysis is by described
Video frame images is decomposed into ten layers, and the decomposition formula of Malla algorithm is:
Ci=HcHrCi-1
In above-mentioned formula, H and G is the coefficient matrix of scaling function φ (x) and wavelet function ψ (x) respectively, Ci、WithCorrespondence image C respectivelyi-1Low frequency part, the HFS of vertical direction, the HFS of vertical direction and diagonal side
To HFS.
Image rain removing method based on wavelet analysis the most according to claim 2, it is characterised in that described step a is also wrapped
Include: detect the figure layer comprising rain cell noise;The described figure layer comprising rain cell noise is the second to the 4th high frequency coefficient figure layer.
Image rain removing method based on wavelet analysis the most according to claim 1, it is characterised in that in described step c,
The calculation of described calculating fusion coefficients matrix is: define raindrop pollution level coefficient according to the light characteristic of raindrop, according to
Raindrop pollution level coefficient calculations fusion coefficients matrix;Described calculating fusion coefficients matrix is particularly as follows: make raindrop pollution levels system
Number S=G × E, wherein, G is partial gradient, and E is local energy, and two parameters of partial gradient and local energy are multiplied and obtain one
New variable S, the S the biggest pollution of value is the most serious;Being normalized s-matrix and obtain S ', coefficient matrix and S ' matrix are used for
Image reconstruction algorithm is weighted optimization and obtains fusion coefficients matrix.
Image rain removing method based on wavelet analysis the most according to claim 4, it is characterised in that described partial gradient G
It is defined as:
In above-mentioned formula, (i, j) (i, (i, j) both horizontally and vertically, M and N is respectively Δ xf j) to be respectively point with Δ yf
The length of side in region;
The local energy E of described pixel is expressed as:
6. an image based on wavelet analysis goes rain system, it is characterised in that include wavelet decomposition module, image denoising module
And wavelet fusion module;Described wavelet decomposition module, for video frame images being carried out block layer decomposition according to wavelet analysis, is analyzed
The image information of described exploded view layer, the figure layer respectively obtain the figure layer comprising background and colouring information, comprising rain cell noise with
And comprise the figure layer of image texture and object edge information;Described image denoising module is for the described figure comprising rain cell noise
Layer carries out bilateral filtering and protects limit denoising;Described wavelet fusion module is used for calculating fusion coefficients matrix, according to described fusion
Coefficient matrix comprises background and the figure layer of colouring information, the figure layer comprising rain cell noise to described and comprise image texture and thing
The figure layer of body marginal information carries out Wavelet Fusion respectively, and carries out image reconstruction according to fusion results and obtain rain image.
Image based on wavelet analysis the most according to claim 6 goes rain system, it is characterised in that described wavelet decomposition mould
Tuber, according to wavelet analysis, video frame images is carried out block layer decomposition particularly as follows: Malla algorithm based on wavelet analysis regards described
Frequently two field picture is decomposed into ten layers, and the decomposition formula of Malla algorithm is:
Ci=HcHrCi-1
In above-mentioned formula, H and G is the coefficient matrix of scaling function φ (x) and wavelet function ψ (x) respectively, Ci、WithCorrespondence image C respectivelyi-1Low frequency part, the HFS of vertical direction, the HFS of vertical direction and diagonal side
To HFS.
Image based on wavelet analysis the most according to claim 7 goes rain system, it is characterised in that described image denoising mould
Block is additionally operable to detect the figure layer comprising rain cell noise;The described figure layer comprising rain cell noise is the second to the 4th high frequency coefficient figure
Layer.
Image based on wavelet analysis the most according to claim 6 goes rain system, it is characterised in that described Wavelet Fusion mould
Block calculates the calculation of fusion coefficients matrix: define raindrop pollution level coefficient, according to rain according to the light characteristic of raindrop
Drip pollution level coefficient calculations fusion coefficients matrix;Described calculating fusion coefficients matrix is particularly as follows: make raindrop pollution level coefficient S
=G × E, wherein, G is partial gradient, and E is local energy, and two parameters of partial gradient and local energy are multiplied and obtain one newly
Variable S, the S the biggest pollution of value the most serious;Being normalized s-matrix and obtain S ', coefficient matrix and S ' matrix are for right
Image reconstruction algorithm is weighted optimization and obtains fusion coefficients matrix.
Image based on wavelet analysis the most according to claim 9 goes rain system, it is characterised in that described partial gradient G
It is defined as:
In above-mentioned formula, (i, j) (i, (i, j) both horizontally and vertically, M and N is respectively Δ xf j) to be respectively point with Δ yf
The length of side in region;
The local energy E of described pixel is expressed as:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610347657.4A CN106023112A (en) | 2016-05-24 | 2016-05-24 | Image rain removing method and system based on wavelet analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610347657.4A CN106023112A (en) | 2016-05-24 | 2016-05-24 | Image rain removing method and system based on wavelet analysis |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106023112A true CN106023112A (en) | 2016-10-12 |
Family
ID=57093701
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610347657.4A Pending CN106023112A (en) | 2016-05-24 | 2016-05-24 | Image rain removing method and system based on wavelet analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106023112A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108665419A (en) * | 2017-03-30 | 2018-10-16 | 展讯通信(上海)有限公司 | A kind of method and device of image denoising |
CN109360155A (en) * | 2018-08-17 | 2019-02-19 | 上海交通大学 | Single-frame images rain removing method based on multi-scale feature fusion |
WO2020000253A1 (en) * | 2018-06-27 | 2020-01-02 | 潍坊学院 | Traffic sign recognizing method in rain and snow |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103729828A (en) * | 2013-12-12 | 2014-04-16 | 中国科学院深圳先进技术研究院 | Video rain removing method |
CN104112255A (en) * | 2014-06-19 | 2014-10-22 | 中国科学院深圳先进技术研究院 | Rain removing method and system for single image |
CN104537634A (en) * | 2014-12-31 | 2015-04-22 | 中国科学院深圳先进技术研究院 | Method and system for removing raindrop influences in dynamic image |
-
2016
- 2016-05-24 CN CN201610347657.4A patent/CN106023112A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103729828A (en) * | 2013-12-12 | 2014-04-16 | 中国科学院深圳先进技术研究院 | Video rain removing method |
CN104112255A (en) * | 2014-06-19 | 2014-10-22 | 中国科学院深圳先进技术研究院 | Rain removing method and system for single image |
CN104537634A (en) * | 2014-12-31 | 2015-04-22 | 中国科学院深圳先进技术研究院 | Method and system for removing raindrop influences in dynamic image |
Non-Patent Citations (1)
Title |
---|
CHEN ZHEN等: "A New Algorithm of Rain (Snow) Removal in Video", 《JOURNAL OF MULTIMEDIA》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108665419A (en) * | 2017-03-30 | 2018-10-16 | 展讯通信(上海)有限公司 | A kind of method and device of image denoising |
CN108665419B (en) * | 2017-03-30 | 2020-07-03 | 展讯通信(上海)有限公司 | Image denoising method and device |
WO2020000253A1 (en) * | 2018-06-27 | 2020-01-02 | 潍坊学院 | Traffic sign recognizing method in rain and snow |
CN109360155A (en) * | 2018-08-17 | 2019-02-19 | 上海交通大学 | Single-frame images rain removing method based on multi-scale feature fusion |
CN109360155B (en) * | 2018-08-17 | 2020-10-13 | 上海交通大学 | Single-frame image rain removing method based on multi-scale feature fusion |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Image dehazing using residual-based deep CNN | |
Shahdoosti et al. | Edge-preserving image denoising using a deep convolutional neural network | |
CN105184761A (en) | Image rain removing method based on wavelet analysis and system | |
CN110120020A (en) | A kind of SAR image denoising method based on multiple dimensioned empty residual error attention network | |
CN106296655A (en) | Based on adaptive weight and the SAR image change detection of high frequency threshold value | |
Chen et al. | SAR image despeckling based on combination of fractional-order total variation and nonlocal low rank regularization | |
CN104240192B (en) | A kind of quick single image to the fog method | |
Chen et al. | Remote sensing image quality evaluation based on deep support value learning networks | |
Gu et al. | A single image dehazing method using average saturation prior | |
CN110557521B (en) | Method, device and equipment for removing rain from video and computer readable storage medium | |
CN110717863B (en) | Single image snow removing method based on generation countermeasure network | |
CN105913392A (en) | Degraded image overall quality improving method in complex environment | |
CN106067163A (en) | A kind of image rain removing method based on wavelet analysis and system | |
CN106023112A (en) | Image rain removing method and system based on wavelet analysis | |
Patil et al. | Motion saliency based generative adversarial network for underwater moving object segmentation | |
Nirmalraj et al. | Fusion of visible and infrared image via compressive sensing using convolutional sparse representation | |
Qian et al. | FAOD-Net: a fast AOD-Net for dehazing single image | |
CN103020905A (en) | Sparse-constraint-adaptive NLM (non-local mean) super-resolution reconstruction method aiming at character image | |
CN112505701B (en) | Method for inhibiting speckle of non-local mean value of approximate half-declining integral graph number | |
Li et al. | Research on haze image enhancement based on dark channel prior algorithm in machine vision | |
Wang et al. | An efficient remote sensing image denoising method in extended discrete shearlet domain | |
CN102314675B (en) | Wavelet high-frequency-based Bayesian denoising method | |
CN116596792B (en) | Inland river foggy scene recovery method, system and equipment for intelligent ship | |
Wang et al. | Poissonian blurred hyperspectral imagery denoising based on variable splitting and penalty technique | |
CN105096274B (en) | Infrared image noise-reduction method based on non-down sampling contourlet domain mixing statistical model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161012 |