CN109325920A - Haze image clarification method, system and can storage medium - Google Patents
Haze image clarification method, system and can storage medium Download PDFInfo
- Publication number
- CN109325920A CN109325920A CN201810886690.3A CN201810886690A CN109325920A CN 109325920 A CN109325920 A CN 109325920A CN 201810886690 A CN201810886690 A CN 201810886690A CN 109325920 A CN109325920 A CN 109325920A
- Authority
- CN
- China
- Prior art keywords
- quality
- network parameter
- depth network
- image
- haze
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 79
- 238000005352 clarification Methods 0.000 title claims abstract description 27
- 230000008569 process Effects 0.000 claims abstract description 16
- 241000894006 Bacteria Species 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims abstract description 12
- 238000013441 quality evaluation Methods 0.000 claims description 26
- 238000013135 deep learning Methods 0.000 claims description 15
- 238000003707 image sharpening Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 5
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 230000003399 chemotactic effect Effects 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000003595 mist Substances 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000010419 fine particle Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000031700 light absorption Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 239000013618 particulate matter Substances 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000018612 quorum sensing Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003362 replicative effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Classifications
-
- 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
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to haze image clarification method, system and can storage medium, the haze image clarification method include: S1, random selection the first depth network parameter haze is carried out to haze image with obtain first go haze image and obtain the first picture quality scoring；S2, score the first high-quality depth network parameter of selection according to the first picture quality；S3, several second depth network parameters of quick bacteria group trip optimization operation acquisition are carried out to the first high-quality depth network parameter；S4, haze image is gone by the second depth network parameter acquisition second and obtains the scoring of the second picture quality；S5, it selects the second high-quality depth network parameter and judges the second high-quality depth network parameter whether to be optimal, if so, the second high-quality depth network parameter corresponding second of output goes haze image and terminates this process；If it is not, the second high-quality depth network parameter is defined as the high-quality depth network parameter of new first and executes S3.Implement the present invention, treatment process is fast, as a result accurate objective.
Description
Technical field
The present invention relates to technical field of image processing, more specifically to a kind of haze image clarification method, system
And it can storage medium.
Background technique
In recent years, haze phenomenon is more serious.The image shot in haze weather, due to be mixed in air it is various can
The pollutants such as particulate matter and fine particle are sucked, has serious influence to the absorption of light, refraction and scattering, leads to image mould
Paste, it has not been convenient to human eye viewing.Image visual effect is bad, not only has an impact to image definition, and can give judgement object tape
To bother.
The picture quality shot under haze environment is not good enough, considerably increases pictures subsequent processing difficulty.In addition to this, mist
The presence of haze image can also be monitored to highway communication, Satellite Remote Sensing is brought greater impact.It can be seen that by image
Reason technology reduces the influences of the adverse weathers to imaging effect such as haze, effectively improves picture quality, is with a wide range of applications.
Haze problem is not only related with everyone health, also closely bound up with urban safety.From the angle of image procossing, how is research
Sharpening processing is carried out to haze image, has become urban development a kind of problem in the urgent need to address.
Image goes haze technology to be intended to remove the interference of the unfavorable factors such as mist in image, haze, and image is made to restore effective letter
Breath and feature obtain the image with good visual effect.Currently, haze image sharpening processing technique is broadly divided into two greatly
Class: the image enchancing method based on image procossing, the image recovery method based on physical model.Wherein:
Image enchancing method based on image procossing is to protrude or weaken certain effective letters by enhancing picture contrast
Breath and characteristics of image are to reduce interference of the haze to image, to realize the sharpening of image.This method principle is simple and is easy to
It realizes, but it has that treated image is easy to appear a color distortion and the problems such as image information is lost.In other words, such side
Method is only the reduction of interference of the haze to image, but not essential removal haze.Image enchancing method based on image procossing
Current more commonly used method mainly have based on histogram equalization, warp wavelet, homomorphic filtering and based on Retinex it is theoretical
Method.
Image recovery method based on physical model is to build from the angle of haze Crack cause to atmospheric scattering effect
Mould, then haze is realized by analyzing and handling.This method ratio that although image information and feature save during processing
It is more complete, it is less the missing of image information occur.But its process is complicated, and side larger by haze morphology influence, more commonly used
Method has based on atmospheric physics model, based on image difference, based on the preferential method of dark and based on the method for fusion.
Summary of the invention
The technical problem to be solved in the present invention is that for image during the above-mentioned haze image sharpening of the prior art
The problems such as being easy to appear color distortion and image information loss or haze complex disposal process, unstable result defect provide
A kind of haze image clarification method, system and can storage medium.
The technical solution adopted by the present invention to solve the technical problems is: constructing a kind of haze image clarification method, wraps
Include following steps:
S1, several first depth network parameters of random selection carry out deep learning to a haze image respectively and remove haze to obtain
It takes several first to go haze image, and goes haze image to carry out blind image quality evaluation to described several first to obtain several the
The scoring of one picture quality；
S2, the first high-quality depth is selected from several first depth network parameters according to the first image quality score
Network parameter；
S3, quick bacteria group trip optimization operation is carried out to the described first high-quality depth network parameter, to obtain several second
Depth network parameter；
S4, haze is gone to obtain to haze image progress deep learning by several second depth network parameters
Several second go haze image, and go haze image to carry out blind image quality evaluation to obtain several second to described several second
Picture quality scoring；
S5, the second high-quality depth is selected from several second depth network parameters according to second picture quality scoring
Network parameter, and judge that described second is high-quality according to the corresponding second picture quality scoring of the described second high-quality depth network parameter
Whether depth network parameter is optimal, if so, S6 is thened follow the steps, if it is not, thening follow the steps S6-1；
S6, it then exports the described second high-quality depth network parameter corresponding second and goes haze image and terminate this process；
S6-1, the described second high-quality depth network parameter is defined as the high-quality depth network parameter of new first and executes institute
State step S3.
Preferably, described according to corresponding second image of the described second high-quality depth network parameter in the step S5
Quality score judges whether the described second high-quality depth network parameter is optimal, comprising:
Calculate the corresponding second picture quality scoring of the described second high-quality depth network parameter and the first high-quality depth network
The difference of the corresponding first picture quality scoring of parameter, determines whether the difference meets the first preset condition, if so, determining
The second high-quality depth network parameter is optimal depth network parameter.
Preferably, described according to corresponding second image of the described second high-quality depth network parameter in the step S5
Quality score judges whether the described second high-quality depth network parameter is optimal, comprising:
Calculate whether the corresponding second picture quality scoring of the described second high-quality depth network parameter meets the second default item
Part, if so, determining that the second high-quality depth network parameter is optimal depth network parameter.
Preferably, in the method, the first depth network parameter and the second depth network parameter are respectively to arrange
Vector parameter includes m parameter value in the column vector parameter, and wherein m is more than or equal to 1.
Preferably, in the method, the first depth network parameter and the second depth network parameter are respectively to go
Vector parameter includes n parameter value in the row vector parameter, and wherein n is more than or equal to 1.
Preferably, in the step S1, described to go haze image to carry out blind image quality evaluation to described several first
To obtain several first picture quality scorings, comprising:
DIIVINE algorithm is used to go haze image to carry out blind image quality evaluation to described several first to obtain several the
The scoring of one picture quality；
In the step S4, if described go haze image to carry out blind image quality evaluation to obtain to described several second
Dry second picture quality scoring, comprising:
If the DIIVINE algorithm is used to go haze image to carry out blind image quality evaluation to obtain to described several second
Dry second picture quality scoring.
Preferably, in the step S3, described that quick bacteria group trip is carried out to the described first high-quality depth network parameter
Optimization operation, to obtain several second depth network parameters；It include: that the constraint condition of the quick bacteria group trip optimization operation expires
Foot:
Work as Ji(j+1, k, l) > Jmin(j, k, l),
Wherein:
Jmin(j, k, l) is the corresponding first picture quality scoring of the described first high-quality network parameter,
Variation for the second depth network parameter relative to the described first high-quality network parameter,
θi(j+1, k, l) is the second depth network parameter,
CccFor attracting factor, for indicate the second depth network parameter relative to the described first high-quality network parameter into
Changed factor when row variation,
θb(j, k, l) is the described first high-quality network parameter.
Preferably, the CccIncluding dynamic step length C (k, l), the constraint condition of the dynamic step length are as follows:
C (k, l)=Lred/nk+l-1
Wherein:
LredFor initial chemotactic step-length,
N is step-length downward gradient.
The present invention also constructs a kind of haze image sharpening system, comprising: processor, memory,
The memory, for storing program instruction,
The processor, the program instruction for being stored according to the memory execute any of the above the method
The step of.
The present invention also constructs a kind of computer readable storage medium, is stored thereon with computer program, the computer journey
It realizes when sequence is executed by processor such as the step of any one the method above.
Implement haze image clarification method of the invention, system and can storage medium, having the advantages that can
Realize that, to the adaptive of haze weather, the current haze weather of objective basis is realized to the processing of haze image sharpening.It is processed
Journey is fast, as a result accurate objective.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples, in attached drawing:
Fig. 1 is the program flow diagram of one embodiment of haze image clarification method of the present invention；
Fig. 2 is haze image clarification method Contrast on effect schematic diagram of the present invention.
Specific embodiment
For a clearer understanding of the technical characteristics, objects and effects of the present invention, now control attached drawing is described in detail
A specific embodiment of the invention.
As shown in Figure 1, in haze image clarification method first embodiment of the invention, comprising the following steps:
S1, several first depth network parameters of random selection carry out deep learning to a haze image respectively and remove haze to obtain
It takes several first to go haze image, and goes haze image to carry out blind image quality evaluation to obtain several first figures to several first
As quality score；Specifically, haze image to be inputted to the depth for going haze deep learning network by removing haze deep learning network
Degree study to haze image carry out it is initial remove haze, it is initial here to go during haze, remove haze deep learning network
For initial depth network parameter it can be appreciated that the first depth network parameter is random selection, this can be in a manner of generating at random
One group of first depth network parameter is generated, one group of first depth network parameter may include the first a number of depth net here
Network parameter goes haze deep learning network to be based on each first depth network parameter and carries out depth respectively to haze image
It practises, goes haze image to obtain corresponding with the first depth network parameter first, then pass through blind image quality evaluating method pair
Each first goes haze image to carry out image quality evaluation, obtains each and first the picture quality of haze image is gone to score.
It is to be understood that the first a number of depth network parameter can be incoherent or relatively low degree of correlation depth
Network parameter is spent, the coverage area of depth network parameter can be expanded in this way, reduces the treatment process and processing money of whole flow process
Source.Wherein deep learning network includes convolutional neural networks (Convolutional neural networks), autocoder
(Autoencoder), Recognition with Recurrent Neural Network (Recurrent neural network) etc..Convolutional Neural can be used herein
Network utilizes the local correlations of input data, the number of parameters of fully-connected network is greatly reduced, in image recognition processes
Middle performance is more excellent
S2, joined according to the scoring of the first picture quality from the high-quality depth network of several first depth network parameters selection first
Number；Specifically, several first relatively obtained above go the picture quality of haze image to score, picture quality scoring is selected most
Good first goes the corresponding first depth network parameter of haze image as the first high-quality depth network parameter.
S3, quick bacteria group trip optimization operation is carried out to the first high-quality depth network parameter, to obtain several second depth
Network parameter；Specifically, carrying out quick bacteria group by the first high-quality depth network parameter to acquisition swims optimization operation, pass through
First high-quality depth network parameter is drawn close to intragroup history optimum point, and the others for searching for surrounding are joined than the first depth network
More preferably depth network parameter is counted, this is defined herein as the second depth network parameter.
S4, haze is gone to obtain several second to haze image progress deep learning by several second depth network parameters
Haze image is gone, and goes haze image to carry out blind image quality evaluation to several second and is commented with obtaining several second picture qualities
Point；Specifically, haze deep learning network is gone to distinguish based on several second depth network parameters of above-mentioned selection haze image
Deep learning is carried out, several corresponding second is obtained and goes haze image, then by blind image quality evaluating method to each
Second goes haze image to carry out image quality evaluation, obtains each and second the picture quality of haze image is gone to score.
S5, joined according to the scoring of the second picture quality from the high-quality depth network of several second depth network parameters selection second
Number, and judge that the second high-quality depth network parameter is according to the corresponding second picture quality scoring of the second high-quality depth network parameter
It is no to be optimal, if so, S6 is thened follow the steps, if it is not, thening follow the steps S6-1；Specifically, several relatively obtained above
Two go the picture quality of haze image to score, and select picture quality scoring optimal second to go haze image, and obtain corresponding
Second depth network parameter is as the second high-quality depth network parameter.Here haze is gone by the second scoring optimal second
The corresponding second picture quality scoring of the high-quality depth network parameter of the quality score of image i.e. second, it can be determined that second is high-quality
Whether depth network parameter is optimal, such as when judging that the scoring of the second picture quality meets the condition of optimal scoring, then can be with
Optimal depth network parameter during second high-quality depth network parameter is removed haze as the haze image, then can be with
Carry out the operation of step S6.When the scoring of the second picture quality is unsatisfactory for the condition of optimal scoring, illustrate the second high-quality depth net
Network parameter is not also that the haze image removes optimal depth network parameter during haze.So to carry out the behaviour of step S6-1
Make.
S6, then the second high-quality depth network parameter corresponding second of output goes haze image and terminates this process；Specifically
, according to judging result above, during it can determine that the second high-quality depth network parameter removes haze for the haze image
When optimal depth network parameter, then it is determined that the second high-quality depth network parameter corresponding second goes the haze image to be
The optimal of the haze image goes haze image.The second high-quality depth network parameter corresponding second is thus removed into haze figure
As the final clear image output during the sharpening of this haze image.
S6-1, the second high-quality depth network parameter is defined as the high-quality depth network parameter of new first and executes step
S3.Specifically, according to judging result above, when the second high-quality depth network parameter of judgement is not also that the haze image removes haze
When optimal depth network parameter in the process, using the second high-quality depth network parameter as the first new high-quality depth network
Parameter, is found the high-quality depth network parameter of new second and is determined the step of repeating step S3 and its back, with
To final the second high-quality depth network parameter for going haze image that can be optimal and export the clear of corresponding haze image
Final clear image output during change.
When here, by compared with network parameter is determined in method, first artificial observed image haze degree is manually chosen again
The method of network parameter compares, and is able to carry out real-time haze image sharpening processing.In addition, by using objective blind image
Haze picture quality is gone in quality evaluation algorithm assessment, and obtain it is optimal go haze image, compared with existing method by subjective evaluation
Method, have higher accuracy.Furthermore optimization algorithm is swum using quick bacteria group and intelligently chooses network parameter, can be avoided
Traditional exhaustive method bring low efficiency problem quickly determines optimal depth network parameter.
Further, in step s 5, it is scored, is sentenced according to corresponding second picture quality of the second high-quality depth network parameter
Whether disconnected second high-quality depth network parameter is optimal, comprising: corresponding second image of the high-quality depth network parameter of calculating second
The difference of quality score and the corresponding first picture quality scoring of the first high-quality depth network parameter, determines whether difference meets the
One preset condition, if so, determining that the second high-quality depth network parameter is optimal depth network parameter.Specifically, judging second
Whether high-quality depth network parameter is optimal depth network parameter, can be by the second high-quality depth network parameter corresponding second
Picture quality scoring and corresponding first picture quality of the first high-quality depth network parameter score and are compared, when the two difference very
It is small, in other words when the second picture quality scoring and the first picture quality scoring very close to when, then illustrate haze image pass through
Crossing deep learning goes the optimal effectiveness of haze to tend towards stability, and finds optimal depth network parameter again and removes haze to it
Effect does not influence, then this when can go haze process using the second high-quality depth network parameter as haze image
In optimal depth network parameter, corresponding second to go haze image be the output of final clear image.
Further, in step s 5, it is scored, is sentenced according to corresponding second picture quality of the second high-quality depth network parameter
Whether disconnected second high-quality depth network parameter is optimal, comprising: corresponding second image of the high-quality depth network parameter of calculating second
Whether quality score meets the second preset condition, if so, determining the second high-quality depth network parameter for optimal depth network ginseng
Number.Specifically, the best output knot for setting the haze image also may be implemented in some haze images during removing haze
Fruit, it can the optimal clear image and its corresponding optimal quality score that it can be exported are defined, when second high-quality
The corresponding second picture quality scoring of depth network parameter meets the optimal quality score, then we can confirm that this is defeated
The best of target that second out goes haze image to meet its definition exports as a result, so it is determined that the second high-quality depth net
Road parameter is optimal depth network parameter, and corresponding second to go haze image be the output of final clear image.Here may be used
With understanding, blind image quality evaluating method is different, and the value range of picture quality scoring also will be different, the figure of algorithms of different
Image quality amount score value meets monotonicity principle, in minimum or maximum value, must there is the optimal situation of correspondence image quality.Cause
This only need to determine image sharpening enabling objective figure according to blind image quality evaluation algorithm types during image sharpening
As quality score is minimum or maximum value.
Further, in some embodiments, in haze image clarification method of the invention, the first depth network parameter
It is respectively column vector parameter with the second depth network parameter, includes m parameter value in column vector parameter, wherein m is greater than or waits
In 1.In further embodiments, in haze image clarification method of the invention, the first depth network parameter and the second depth
Network parameter is respectively row vector parameter, includes n parameter value in row vector parameter, wherein n is greater than or equal to 1.Specifically,
Different removes haze deep learning network, since it constitutes difference, can have the parameter of different number and physical significance, such as
It can be column vector parameter, row vector parameter, also can be matrix parameter.Row vector parameter below such as,
Wherein:
bi--- i-th of depth network parameter vector；
--- n-th of parameter value of i-th of depth network parameter vector.
Specifically, in step sl, haze image is gone to carry out blind image quality evaluation to several first to obtain several the
One picture quality scoring, comprising: DIIVINE algorithm is used to go haze image to carry out blind image quality evaluation to obtain to several first
Several first picture qualities are taken to score；In step s 4, haze image is gone to carry out blind image quality evaluation to obtain to several second
Several second picture qualities are taken to score, comprising: to go haze image to carry out blind picture quality to several second using DIIVINE algorithm
Evaluation is to obtain several second picture quality scorings.Specifically, being directed to the blind picture quality of haze image in some embodiments
Evaluation can also be adopted and be gone respectively to first with other methods certainly in other examples using DIIVINE algorithm
Haze image and second goes haze image to carry out blind image quality evaluation, it is to be understood that, it is selected in whole flow process
One blind image quality evaluating method will be selected always to when going the haze image to carry out blind image quality evaluation through a blind figure
Image quality evaluation method completes whole flow process.And it is not suitable for replacing blind image quality evaluation scheme in the process to different phase
Haze image is gone to carry out blind image quality evaluation.
Further, in step s3, quick bacteria group is carried out to the first high-quality depth network parameter and swims optimization operation, with
Obtain several second depth network parameters；Include: the constraint condition satisfaction that quick bacteria group swims optimization operation:
Work as Ji(j+1, k, l) > Jmin(j, k, l),
Wherein:
Jmin(j, k, l) is the corresponding first picture quality scoring of the first high-quality network parameter,
For the second depth network parameter,
θi(j+1, k, l) is the first depth network parameter,
CccFor attracting factor, for indicate the second depth network parameter relative to the described first high-quality network parameter into
Changed factor when row variation,
θb(j, k, l) is the described first high-quality network parameter.
Specifically, the first depth network parameter will be allowed to utilize the high-quality net of surrounding first using above-mentioned new quorum sensing mechanism
The experience of network parameter instructs oneself variation, can substantially shorten search time of the algorithm in solution space.Meanwhile new mechanism
Facilitate the first depth network parameter and jump out locally optimal solution, the first high-quality network parameter can be effectively reduced and escaped from globe optimum
A possibility that ease.
Further, CccIncluding dynamic step length C (k, l), the constraint condition of dynamic step length are as follows:
C (k, l)=Lred/nk+l-1
Wherein:
LredFor initial chemotactic step-length,
N is step-length downward gradient.
Specifically, C (k, l) is on a declining curve with replicating and eliminating-dispersing event times increase.When k+l is smaller, C (k,
L) larger, it can avoid in regional area overspending search time；When k+l is gradually increased, C (k, l) shortens, and it is excellent can to enhance first
Local search ability of the matter network parameter near globe optimum guarantees that algorithm finally tends to globe optimum.
In addition, a kind of haze image sharpening system of the invention, comprising: processor, memory, memory, for storing
The step of program instruction, processor, the program instruction for being stored according to memory executes an any of the above method.Specifically
, above-mentioned haze image clarification method can be carried out by haze image sharpening system, export final sharpening figure
Picture.
In addition, a kind of computer readable storage medium of the invention, is stored thereon with computer program, computer program is located
It manages when device executes and realizes such as the step of any one method above.Specifically, method described above can be used as program storage,
And carry out duplicate copy.Here computer readable storage medium, which can be, but not limited to, can keep and store by instruction execution
The tangible device for the instruction that equipment uses.Computer readable storage medium for example can be but not limited to storage device electric, magnetic is deposited
Store up equipment, light storage device, electric magnetic storage apparatus, semiconductor memory apparatus or above-mentioned any appropriate combination.And packet
Device combination containing above-mentioned storage equipment.
As shown in Fig. 2, for the processing result of haze image clarification method of the invention and the knot of existing processing method
The comparison diagram of fruit, wherein a is classified as original haze image, and as to sharpening processing to volume haze image, b, c, d are existing
The sharpening of haze image is as a result, e is the processing result of haze image clarification method of the invention, it can be seen that its effect is excellent
In the clarification method of existing haze image or suitable with the clarification method result of existing haze image.But
Its process is far superior to the clarification method of existing haze image.
It should be understood that above embodiments only express the preferred embodiment of the present invention, description is more specific and detailed
Carefully, but it cannot be understood as limitations on the scope of the patent of the present invention；It should be pointed out that for the common skill of this field
For art personnel, without departing from the inventive concept of the premise, above-mentioned technical characterstic can be freely combined, can also be done
Several modifications and improvements out, these are all within the scope of protection of the present invention；Therefore, all to be done with scope of the invention as claimed
Equivalents and modification, should belong to the covering scope of the claims in the present invention.
Claims (10)
1. a kind of haze image clarification method, which comprises the following steps:
If S1, several first depth network parameters of random selection carry out deep learning to a haze image respectively and remove haze to obtain
Dry first goes haze image, and goes haze image to carry out blind image quality evaluation to obtain several first figures to described several first
As quality score；
S2, the first high-quality depth network is selected from several first depth network parameters according to the first image quality score
Parameter；
S3, quick bacteria group trip optimization operation is carried out to the described first high-quality depth network parameter, to obtain several second depth
Network parameter；
S4, carrying out deep learning to the haze image by several second depth network parameters, to remove haze several to obtain
Second goes haze image, and goes haze image to carry out blind image quality evaluation to obtain several second images to described several second
Quality score；
S5, the second high-quality depth network is selected from several second depth network parameters according to second picture quality scoring
Parameter, and the described second high-quality depth is judged according to the corresponding second picture quality scoring of the described second high-quality depth network parameter
Whether network parameter is optimal, if so, S6 is thened follow the steps, if it is not, thening follow the steps S6-1；
S6, it then exports the described second high-quality depth network parameter corresponding second and goes haze image and terminate this process；
S6-1, the described second high-quality depth network parameter is defined as the high-quality depth network parameter of new first and executes the step
Rapid S3.
2. haze image clarification method according to claim 1, which is characterized in that in the step S5, described
According to the corresponding second picture quality scoring of the described second high-quality depth network parameter, the described second high-quality depth network parameter is judged
It whether is optimal, comprising:
Calculate the corresponding second picture quality scoring of the described second high-quality depth network parameter and the first high-quality depth network parameter
The difference of corresponding first picture quality scoring, determines whether the difference meets the first preset condition, if so, described in determining
Second high-quality depth network parameter is optimal depth network parameter.
3. haze image clarification method according to claim 1, which is characterized in that in the step S5, described
According to the corresponding second picture quality scoring of the described second high-quality depth network parameter, the described second high-quality depth network parameter is judged
It whether is optimal, comprising:
Calculate whether the corresponding second picture quality scoring of the described second high-quality depth network parameter meets the second preset condition, if
It is then to determine that the described second high-quality depth network parameter is optimal depth network parameter.
4. haze image clarification method according to claim 1, which is characterized in that in the method, described first is deep
It spends network parameter and the second depth network parameter is respectively column vector parameter, include m ginseng in the column vector parameter
Numerical value, wherein m is greater than or equal to 1.
5. haze image clarification method according to claim 1, which is characterized in that in the method, described first is deep
It spends network parameter and the second depth network parameter is respectively row vector parameter, include n ginseng in the row vector parameter
Numerical value, wherein n is greater than or equal to 1.
6. haze image clarification method according to claim 1, which is characterized in that in the step S1, described right
Described several first go haze image to carry out blind image quality evaluation to obtain several first picture quality scorings, comprising:
DIIVINE algorithm is used to go haze image to carry out blind image quality evaluation to obtain several first figures to described several first
As quality score；
It is described to go haze image to carry out blind image quality evaluation to described several second to obtain several the in the step S4
The scoring of two picture qualities, comprising:
The DIIVINE algorithm is used to go haze image to carry out blind image quality evaluation to described several second to obtain several the
The scoring of two picture qualities.
7. haze image clarification method according to claim 1, which is characterized in that in the step S3, described right
The first high-quality depth network parameter carries out quick bacteria group and swims optimization operation, to obtain several second depth network parameters；
Include: the constraint condition satisfaction that the quick bacteria group swims optimization operation:
Work as Ji(j+1, k, l) > Jmin(j, k, l),
Wherein:
Jmin(j, k, l) is the corresponding first picture quality scoring of the described first high-quality network parameter,
Variation for the second depth network parameter relative to the described first high-quality network parameter,
θi(j+1, k, l) is the second depth network parameter,
CccFor attracting factor, for indicating that the second depth network parameter is become relative to the described first high-quality network parameter
Changed factor when change,
θb(j, k, l) is the described first high-quality network parameter.
8. haze image clarification method according to claim 7, which is characterized in that the CccIncluding dynamic step length C (k,
L), the constraint condition of the dynamic step length are as follows:
C (k, l)=Lred/nk+l-1
Wherein:
LredFor initial chemotactic step-length,
N is step-length downward gradient.
9. a kind of haze image sharpening system characterized by comprising processor, memory,
The memory, for storing program instruction,
The processor, the program instruction for being stored according to the memory are executed such as claim 1-8 any one institute
The step of stating method.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
It realizes when being executed by processor such as the step of claim 1-8 any one the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810886690.3A CN109325920B (en) | 2018-08-06 | 2018-08-06 | Haze image sharpening method and system and storable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810886690.3A CN109325920B (en) | 2018-08-06 | 2018-08-06 | Haze image sharpening method and system and storable medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109325920A true CN109325920A (en) | 2019-02-12 |
| CN109325920B CN109325920B (en) | 2022-02-22 |
Family
ID=65263584
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810886690.3A Active CN109325920B (en) | 2018-08-06 | 2018-08-06 | Haze image sharpening method and system and storable medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109325920B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428371A (en) * | 2019-07-03 | 2019-11-08 | 深圳大学 | Image defogging method, system, storage medium and electronic equipment based on super-pixel segmentation |
| CN112241670A (en) * | 2019-07-18 | 2021-01-19 | 杭州海康威视数字技术股份有限公司 | Image processing method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104915933A (en) * | 2015-06-01 | 2015-09-16 | 长安大学 | Foggy day image enhancing method based on APSO-BP coupling algorithm |
| US20160071244A1 (en) * | 2014-09-04 | 2016-03-10 | National Taipei University Of Technology | Method and system for image haze removal based on hybrid dark channel prior |
| CN106910201A (en) * | 2017-02-28 | 2017-06-30 | 江南大学 | A kind of method that image procossing is carried out based on improved fireworks algorithm |
| CN107958465A (en) * | 2017-10-23 | 2018-04-24 | 华南农业大学 | A kind of single image to the fog method based on depth convolutional neural networks |
| US20180122049A1 (en) * | 2016-11-01 | 2018-05-03 | Chun Ming Tsang | Systems and methods for removing haze in digital photos |
| US20180144466A1 (en) * | 2016-11-23 | 2018-05-24 | General Electric Company | Deep learning medical systems and methods for image acquisition |
-
2018
- 2018-08-06 CN CN201810886690.3A patent/CN109325920B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160071244A1 (en) * | 2014-09-04 | 2016-03-10 | National Taipei University Of Technology | Method and system for image haze removal based on hybrid dark channel prior |
| CN104915933A (en) * | 2015-06-01 | 2015-09-16 | 长安大学 | Foggy day image enhancing method based on APSO-BP coupling algorithm |
| US20180122049A1 (en) * | 2016-11-01 | 2018-05-03 | Chun Ming Tsang | Systems and methods for removing haze in digital photos |
| US20180144466A1 (en) * | 2016-11-23 | 2018-05-24 | General Electric Company | Deep learning medical systems and methods for image acquisition |
| CN106910201A (en) * | 2017-02-28 | 2017-06-30 | 江南大学 | A kind of method that image procossing is carried out based on improved fireworks algorithm |
| CN107958465A (en) * | 2017-10-23 | 2018-04-24 | 华南农业大学 | A kind of single image to the fog method based on depth convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| CHENYAN WANG ET AL: "PSO-based Parameters Selection for the Bilateral Filter in Image Denoising", 《GECCO ’17》 * |
| 储颖等: "基于粒子群优化的快速细菌群游算法", 《数据采集与处理》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428371A (en) * | 2019-07-03 | 2019-11-08 | 深圳大学 | Image defogging method, system, storage medium and electronic equipment based on super-pixel segmentation |
| CN112241670A (en) * | 2019-07-18 | 2021-01-19 | 杭州海康威视数字技术股份有限公司 | Image processing method and device |
| CN112241670B (en) * | 2019-07-18 | 2024-03-01 | 杭州海康威视数字技术股份有限公司 | Image processing method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109325920B (en) | 2022-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110929578B (en) | Anti-shielding pedestrian detection method based on attention mechanism | |
| CN111738942A (en) | Generation countermeasure network image defogging method fusing feature pyramid | |
| CN109300090A (en) | A kind of single image to the fog method generating network based on sub-pix and condition confrontation | |
| CN109712165A (en) | A kind of similar foreground picture image set dividing method based on convolutional neural networks | |
| CN107748895A (en) | UAV Landing landforms image classification method based on DCT CNN models | |
| CN110532946B (en) | Method for identifying axle type of green-traffic vehicle based on convolutional neural network | |
| CN109359666A (en) | A kind of model recognizing method and processing terminal based on multiple features fusion neural network | |
| CN110032925A (en) | A kind of images of gestures segmentation and recognition methods based on improvement capsule network and algorithm | |
| CN106934455B (en) | Remote sensing image optics adapter structure choosing method and system based on CNN | |
| CN113160062A (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN109472193A (en) | Method for detecting human face and device | |
| CN108038859A (en) | PCNN figures dividing method and device based on PSO and overall evaluation criterion | |
| CN109086779A (en) | A kind of attention target identification method based on convolutional neural networks | |
| CN112733929A (en) | Improved method for detecting small target and shielded target of Yolo underwater image | |
| CN109360159A (en) | A kind of image completion method based on generation confrontation network model | |
| CN106339984A (en) | Distributed image super-resolution method based on K-means driven convolutional neural network | |
| CN108961358A (en) | A kind of method, apparatus and electronic equipment obtaining samples pictures | |
| CN106296618A (en) | A kind of color image defogging method based on Gaussian function weighted histogram regulation | |
| CN110390673A (en) | Cigarette automatic testing method based on deep learning under a kind of monitoring scene | |
| CN109961407A (en) | Facial image restorative procedure based on face similitude | |
| CN109325920A (en) | Haze image clarification method, system and can storage medium | |
| CN108470196A (en) | A method of handwritten numeral is generated based on depth convolution confrontation network model | |
| CN107291781A (en) | A kind of image management method and device | |
| CN114049478A (en) | Infrared ship image rapid identification method and system based on improved Cascade R-CNN | |
| CN104700416B (en) | The image segmentation threshold that view-based access control model understands determines method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |