CN109785268A - A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm - Google Patents
A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm Download PDFInfo
- Publication number
- CN109785268A CN109785268A CN201910061169.0A CN201910061169A CN109785268A CN 109785268 A CN109785268 A CN 109785268A CN 201910061169 A CN201910061169 A CN 201910061169A CN 109785268 A CN109785268 A CN 109785268A
- Authority
- CN
- China
- Prior art keywords
- image
- double
- spectrum analysis
- ssim
- value
- 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.)
- Withdrawn
Links
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a kind of underwater image restoration methods based on double-spectrum analysis optimization algorithm, include the following steps, A, multiple original series images are acquired by video camera, designs the SSIM high-pass filtering based on image quality evaluation, filter out the lesser testing image sequence of distortion；B, by gaussian pyramid template matching method, coarse positioning module position optimizes the spatial dimension of double-spectrum analysis simultaneously from room and time；C, using the bispectrum of picture signal, the Fourier's amplitude and phase of image are calculated, distortionless image is exported by bispectrum reconstruct.The present invention has the advantage that effectively removing interference of the water to image, reconstruct to obtain underwater clear image by bispectrum.
Description
Technical field
The invention belongs to offshore wind farm underwater picture process fields, and in particular to a kind of based on double-spectrum analysis optimization algorithm
Underwater image restoration method.
Background technique
China's oceanic area is vast, and offshore wind energy resource is abundant, very big using the potential of wind-power electricity generation, with the new skill of wind-powered electricity generation
The independent development application of art, new material and new process, China's offshore wind farm enter the Large scale construction stage, and offshore wind farm is being built
In the process, it needs to carry out the construction location of marine parts calculating measurement, and measurement is dived into the water with one by operator
Determine difficulty, it is therefore desirable to the attention studied increasingly by the personnel of the sector to underwater image restoration.
The influence of water flow fluctuation in water environment be will receive when light transmits in water in image imaging process due to water surface wave
During dynamic, topological transformation has high-speed motion, randomness and very high complexity, therefore restores to distorted image
Very big difficulty is faced, recent domestic researcher always strives to solve this problem.
Carnegie Mellon University, U.S. Alexei A.Efros etc. first proposed in 2005 will solve atmospheric turbulance
The thought of lucky block selection is applied to underwater picture processing, proposes that image caused by water surface ripple is treated from statistical angle to be turned round
Song, but since underwater tumbling frequency is too fast, best lucky block image is difficult to obtain, while the spelling of the lucky block image of non-optimal
It connects and merges the geometric distortion for easily causing image and obscure；Nicolas Paul of group, Electricite De France etc. in 2013 directly into
Row time-domain filtering passes through the Wiener deconvolution based on laplacian distribution, directly removal disturbance and time domain average bring image
Fuzzy method, carries out underwater disturbance image restoration, and Yuandong Tian of Carnegie Mellon University, the U.S. etc. was ground in 2009
Study carefully and spatial distortion model is established according to wave equation, has carried out the restored method of warp image, New South Wales, Australia
Zhiying Wen of university etc. proposed the lucky block fusion method based on double-spectrum analysis technology, double-spectrum analysis at 2010 etc.
Technology is commonly used in caused by astronomical image removal atmospheric turbulance and obscures, and Univ Florida USA Omar Oreifej etc. is 2011
It has studied in year and two-step method image recovery method is imaged through the water surface, propose the method using image registration, to underwater disturbance figure
As carrying out distortion correction, reach image restoration, Li Lei of Northwestern Polytechnical University is equal, Institute of Automation, CAS Wenrui Hu etc.,
Kalyan Kumar Halder of University of New South Wales etc. (, the Andrey of United States Naval Research Laboratory
Kanaev etc. is all based on image registration algorithm and has carried out algorithm improvement research, since the underwater disturbance image based on image registration is multiple
Original method carries out distortion correction by successive ignition to underwater disturbance image and reaches image restoration come the distortion of remedial frames, if
The number of iterations is less, then can distortion correction it is ineffective, while the image restored is there are geometric distortion, fixed for the part in later period
Position measurement brings biggish error.
Summary of the invention
In view of the above-mentioned deficiencies in the prior art, it is an object of the present invention to now provide a kind of water based on double-spectrum analysis optimization algorithm
Lower image recovery method effectively removes interference of the water to image, reconstructs to obtain underwater clear image by bispectrum.
In order to solve the above technical problems, the technical solution adopted by the present invention are as follows: a kind of based on double-spectrum analysis optimization algorithm
Underwater image restoration method includes the following steps,
A, multiple original series images are acquired by video camera, design the SSIM high-pass filtering based on image quality evaluation,
Filter out the lesser testing image sequence of distortion；
B, by gaussian pyramid template matching method, coarse positioning module position optimizes double simultaneously from room and time
The spatial dimension of spectrum analysis；
C, using the bispectrum of picture signal, the Fourier's amplitude and phase of image are calculated, not by bispectrum reconstruct output
The image of distortion.
Further, specific step is as follows by step A:
Quality evaluation function SSIM is found out in the following way for the structural similarity of two images a and b,
Wherein μαIt is the average value of a, μbIt is the average value of b,It is the variance of a,It is the variance of b, σabIt is the association of a and b
Variance；c1=(k1L)2And c2=(k2L)2It is for maintaining stable constant, L is the dynamic range of pixel value, usual k1=
0.01, k2=0.03；
It is -1 to 1, when two image striking resemblances, the value of SSIM by the SSIM value range that above-mentioned formula is calculated
Equal to 1, using time domain mean value image as reference picture, by the SSIM value of sequence of calculation image and time domain mean value image,
The lesser image of SSIM value is that torsional deformation is bigger, needs to reject, and chooses the biggish image of SSIM value as image to be processed
Sequence, the Optimization of Time Domain as double-spectrum analysis.
Further, specific step is as follows by step B:
(a) by image sequence obtained in step A by gauss low frequency filter carry out Gaussian kernel convolution, to image into
The down-sampled image for obtaining size reduction of row；
(b) using gaussian pyramid template matching algorithm to obtained image sequence coarse positioning single part in the picture
Position, and it is partitioned into single part, pyramid template matching algorithm Normalized Cross Correlation Function γ (u, v) is indicated are as follows:
Wherein f (x, y) indicates reference picture, and t is template image,Indicate image mean value, then the value range of γ (u, v)
For [- 1,1], vertex is optimal match point i.e. template image and the highest place of reference picture matching degree, passes through template
Divisible region to be measured out is matched, 1.2-1.5 times that segmentation range is template is chosen.
Further, specific step is as follows by step C:
(a) assumeFor the Fourier transform of 2D signal i (x, y),It can be expressed as its amplitude spectrum
Phase spectrumProduct, formula is
WhereinRepresentation space frequencyPhase angle, target image is obtained by formula (3)；
(b) the bispectrum B (f of two-position signal1, f2) can be obtained by formula (2):
WhereinForComplex conjugate, reconstruct restored image.
Further, the period of SSIM high-pass filtering is 50 frames, and filtering threshold is set as 10 frames, i.e., every 50 frame original image
Sequence filters out 10 high frame images of SSIM value.
Further, the segmentation range that region to be measured is chosen in (b) of step B is 1.5 times of template.
Further, for video camera using SonyTELI30 ten thousand as number CCD camera, camera calibration plate is standard 10mm chess
Disk lattice.
Beneficial effects of the present invention are as follows:
This paper presents a kind of double-spectrum analysis image reconstructing method based on space and time optimization, first design are based on picture quality
The high-pass filtering of evaluation filters out the lesser testing image sequence of distortion, while by gaussian pyramid template matching method, slightly
The position of locating element optimizes the spatial dimension of double-spectrum analysis simultaneously from room and time, then constructs the double of image sequence
Spectrum, effectively removes interference of the water to image, reconstructs to obtain underwater clear image by bispectrum.
Detailed description of the invention:
The following examples can make professional and technical personnel that the present invention be more fully understood, but therefore not send out this
It is bright to be limited among the embodiment described range.
Fig. 1 is the schematic diagram of sequence image SSIM value in step A of the invention.
Fig. 2 is original sequence image schematic diagram in step A of the invention.
Fig. 3 is time domain average image schematic diagram in background technique of the invention.
Fig. 4 is restored image schematic diagram of the invention.
Specific embodiment
Embodiments of the present invention are illustrated by particular specific embodiment below, those skilled in the art can be by this explanation
Content disclosed by book is understood other advantages and efficacy of the present invention easily.
A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm as shown in figure 1, includes the following steps,
A, multiple original series images are acquired by video camera, design the SSIM high-pass filtering based on image quality evaluation,
Filter out the lesser testing image sequence of distortion；
B, by gaussian pyramid template matching method, coarse positioning module position optimizes double simultaneously from room and time
The spatial dimension of spectrum analysis；
C, using the bispectrum of picture signal, the Fourier's amplitude and phase of image are calculated, not by bispectrum reconstruct output
The image of distortion.
Specific step is as follows by step A:
Quality evaluation function SSIM is found out in the following way for the structural similarity of two images a and b,
Wherein μaIt is the average value of a, μbIt is the average value of b,It is the variance of a,It is the variance of b, σabIt is the association of a and b
Variance；c1=(k1L)2And c2=(k2L)2It is for maintaining stable constant, L is the dynamic range of pixel value, usual k1=
0.01, k2=0.03；
It is -1 to 1, when two image striking resemblances, the value of SSIM by the SSIM value range that above-mentioned formula is calculated
Equal to 1, using time domain mean value image as reference picture, by the SSIM value of sequence of calculation image and time domain mean value image,
The lesser image of SSIM value is that torsional deformation is bigger, needs to reject, and chooses the biggish image of SSIM value as image to be processed
Sequence, the Optimization of Time Domain as double-spectrum analysis.
Specific step is as follows by step B:
(a) by image sequence obtained in step A by gauss low frequency filter carry out Gaussian kernel convolution, to image into
The down-sampled image for obtaining size reduction of row；
(b) using gaussian pyramid template matching algorithm to obtained image sequence coarse positioning single part in the picture
Position, and it is partitioned into single part, pyramid template matching algorithm Normalized Cross Correlation Function γ (u, v) is indicated are as follows:
Wherein f (x, y) indicates reference picture, and t is template image,Indicate image mean value, then the value range of γ (u, v)
For[-1,1], vertex is optimal match point i.e. template image and the highest place of reference picture matching degree, passes through template
With divisible region to be measured out, 1.2-1.5 times that segmentation range is template is chosen.
Specific step is as follows by step C:
(a) assumeFor the Fourier transform of 2D signal i (x, y),It can be expressed as its amplitude spectrum
Phase spectrumProduct, formula is
WhereinRepresentation space frequencyPhase angle, target image is obtained by formula (3)；
(b) the bispectrum B (f of two-position signal1, f2) can be obtained by formula (2):
WhereinForComplex conjugate, reconstruct restored image.
For video camera using SonyTELI30 ten thousand as number CCD camera, camera calibration plate is standard 10mm gridiron pattern.Complete
After camera calibration, as shown in Fig. 2, video camera acquires original series image.Consecutive image is carried out in the time domain first equal
Based on the high-pass filtering of structuring SSIM image quality evaluation, the period of SSIM high-pass filtering is 50 frames, and filtering threshold is set as
10 frames, i.e., every 50 frame original sequence filter out 10 high frame images of SSIM value, as shown in figure 1 as can be seen that sequence image
It is fluctuated in SSIM value relatively strong and relatively mixed and disorderly more irregular.Therefore using SSIM high-pass filtering by the way of, can remove distort compared with
Big image, while shortening the space-time expense being further processed.
Image sequence after high-pass filtering screening does time domain mean filter, carries out gaussian pyramid template matching.It marks
Then single part region splits the part zone in image sequence.Point in region to be measured is chosen in (b) of step B
Cut 1.5 times that range is template.By double-spectrum analysis, restored image is reconstructed, Fig. 2 is original series in step A of the invention
Image schematic diagram, water have certain disturbance to image, and Fig. 3 is time domain average image schematic diagram in background technique of the invention,
Fig. 4 is restored image schematic diagram of the invention, it can be seen that restored image shows that more details, the clarity of image obtain
It improves, is conducive to the vision-based detection of next step.
Above-described embodiment is presently preferred embodiments of the present invention, is not a limitation on the technical scheme of the present invention, as long as
Without the technical solution that creative work can be realized on the basis of the above embodiments, it is regarded as falling into the invention patent
Rights protection scope in.
Claims (7)
1. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm, which is characterized in that include the following steps,
A, multiple original series images are acquired by video camera, designs the SSIM high-pass filtering based on image quality evaluation, screening
Lesser testing image sequence is distorted out；
B, by gaussian pyramid template matching method, coarse positioning module position optimizes bispectrum point simultaneously from room and time
The spatial dimension of analysis；
C, using the bispectrum of picture signal, the Fourier's amplitude and phase of image are calculated, it is undistorted by bispectrum reconstruct output
Image.
2. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm according to claim 1, which is characterized in that
Specific step is as follows by the step A:
Quality evaluation function SSIM is found out in the following way for the structural similarity of two images a and b,
Wherein μaIt is the average value of a, μbIt is the average value of b,It is the variance of a,It is the variance of b, σabIt is the covariance of a and b；
c1=(k1L)2And c2=(k2L)2It is for maintaining stable constant, L is the dynamic range of pixel value, usual k1=0.01, k2=
0.03；
It is -1 to 1 by the SSIM value range that above-mentioned formula is calculated, when two image striking resemblances, the value of SSIM is equal to
1, using time domain mean value image as reference picture, pass through the SSIM value of sequence of calculation image and time domain mean value image, SSIM value
Lesser image is that torsional deformation is bigger, needs to reject, and chooses the biggish image of SSIM value as image sequence to be processed, makees
For the Optimization of Time Domain of double-spectrum analysis.
3. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm according to claim 1, which is characterized in that
Specific step is as follows by the step B:
(a) image, drops in the convolution that image sequence obtained in step A is carried out to Gaussian kernel by gauss low frequency filter
Sampling obtains the image of size reduction；
(b) position using gaussian pyramid template matching algorithm to obtained image sequence coarse positioning single part in the picture
It sets, and is partitioned into single part, pyramid template matching algorithm Normalized Cross Correlation Function γ (u, v) is indicated are as follows:
Wherein f (x, y) indicates reference picture, and t is template image,Indicate image mean value, then the value range of γ (u, v) be [-
1,1], vertex is optimal match point i.e. template image and the highest place of reference picture matching degree, passes through template matching
1.2-1.5 times that segmentation range is template is chosen in divisible region to be measured out.
4. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm according to claim 1, which is characterized in that
Specific step is as follows by the step C:
(a) assumeFor the Fourier transform of 2D signal i (x, y),It can be expressed as its amplitude spectrumAnd phase
SpectrumProduct, formula is
WhereinRepresentation space frequencyPhase angle, target image is obtained by formula (3)；
(b) bispectrum of two-position signalIt can be obtained by formula (2):
WhereinForComplex conjugate, reconstruct restored image.
5. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm according to claim 1, which is characterized in that
The period of the SSIM high-pass filtering is 50 frames, and filtering threshold is set as 10 frames, i.e., every 50 frame original sequence filters out
10 high frame images of SSIM value.
6. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm according to claim 3, which is characterized in that
The segmentation range that region to be measured is chosen in (b) of the step B is 1.5 times of template.
7. a kind of underwater image restoration method based on double-spectrum analysis optimization algorithm according to claim 1, it is characterised in that:
For the video camera using SonyTELI30 ten thousand as number CCD camera, camera calibration plate is standard 10mm gridiron pattern.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910061169.0A CN109785268A (en) | 2019-01-23 | 2019-01-23 | A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910061169.0A CN109785268A (en) | 2019-01-23 | 2019-01-23 | A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109785268A true CN109785268A (en) | 2019-05-21 |
Family
ID=66502185
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910061169.0A Withdrawn CN109785268A (en) | 2019-01-23 | 2019-01-23 | A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785268A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111524083A (en) * | 2020-05-07 | 2020-08-11 | 桂林电子科技大学 | Active and passive combined underwater aerial imaging image recovery method based on structured light |
| CN112231869A (en) * | 2020-09-21 | 2021-01-15 | 江苏大学镇江流体工程装备技术研究院 | Method and device for measuring dean vortex motion information |
-
2019
- 2019-01-23 CN CN201910061169.0A patent/CN109785268A/en not_active Withdrawn
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111524083A (en) * | 2020-05-07 | 2020-08-11 | 桂林电子科技大学 | Active and passive combined underwater aerial imaging image recovery method based on structured light |
| CN112231869A (en) * | 2020-09-21 | 2021-01-15 | 江苏大学镇江流体工程装备技术研究院 | Method and device for measuring dean vortex motion information |
| CN112231869B (en) * | 2020-09-21 | 2023-06-16 | 江苏大学镇江流体工程装备技术研究院 | Dien vortex movement information measuring method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111062978B (en) | Texture recognition method for spatio-temporal image flow measurement based on frequency domain filtering technology | |
| CN100433795C (en) | Method for image noise reduction based on transforming domain mathematics morphology | |
| CN100417191C (en) | Method of reducing noise for combined images | |
| CN109785268A (en) | A kind of underwater image restoration method based on double-spectrum analysis optimization algorithm | |
| CN102129694B (en) | Method for detecting salient region of image | |
| CN105335972A (en) | Warp knitting fabric defect detection method based on wavelet contourlet transformation and visual saliency | |
| Li et al. | A research on improved canny edge detection algorithm | |
| CN100433062C (en) | Image noise reducing method for Contourlet transform | |
| CN113139900B (en) | Method for acquiring complete surface image of bar | |
| CN106296763A (en) | A kind of metal material Industry CT Image Quality method for quickly correcting | |
| CN111144318B (en) | Noise reduction method for point cloud data of underwater sonar system | |
| CN117726537A (en) | SAR image denoising network method and system for self-adaptive multi-scale feature fusion AMFFD-Net | |
| CN116012261A (en) | Water surface flare automatic inhibition method based on focal plane polarization camera | |
| CN112419206B (en) | SAR interferogram filtering method based on multi-scale decomposition-reconstruction | |
| CN104657983B (en) | A kind of fabric ball top Density Detection method filtered based on Gabor | |
| CN110032968A (en) | Denoising method based on dual-tree complex wavelet and adaptive semi-soft threshold model | |
| CN106023179A (en) | SAR image coastline extracting method based on geometric active contour model | |
| CN104268831B (en) | Infrared image compensation method under marine sunlight bright band interference | |
| CN103345727A (en) | Method for reconstructing binary optical image spectrum | |
| CN113689449B (en) | Inversion method and system for characteristic parameters of mesoscale vortices | |
| CN114170239A (en) | Large-view-field self-adaptive wall filtering method and system based on hierarchical search | |
| CN114648505A (en) | Method and system for automatically measuring cycle and phase of annular texture image | |
| CN114972075A (en) | Hyperspectral image denoising method based on residual learning and mixed domain attention | |
| CN112927169A (en) | Remote sensing image denoising method based on wavelet transformation and improved weighted nuclear norm minimization | |
| Zhang et al. | Deep learning based methods for water body extraction and flooding evolution analysis based on sentinel-1 images |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190521 |