CN115049562A - Underwater image restoration method based on distortion guidance - Google Patents
Underwater image restoration method based on distortion guidance Download PDFInfo
- Publication number
- CN115049562A CN115049562A CN202210782462.8A CN202210782462A CN115049562A CN 115049562 A CN115049562 A CN 115049562A CN 202210782462 A CN202210782462 A CN 202210782462A CN 115049562 A CN115049562 A CN 115049562A
- Authority
- CN
- China
- Prior art keywords
- image
- distortion
- underwater
- distorted
- network
- 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 36
- 238000001514 detection method Methods 0.000 claims abstract description 32
- 238000011084 recovery Methods 0.000 claims abstract description 32
- 238000005070 sampling Methods 0.000 claims abstract description 15
- 230000002776 aggregation Effects 0.000 claims abstract description 10
- 238000004220 aggregation Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 12
- 230000005540 biological transmission Effects 0.000 claims description 11
- 238000010586 diagram Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 3
- 241000282414 Homo sapiens Species 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- 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]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention relates to an underwater image restoration method based on distortion guidance, which comprises the following steps: s1, constructing an underwater multi-type distortion image database; step S2, training a distortion detection network through I and M in an underwater distortion image database, and predicting a mask image M' of a distortion image; s3: the method comprises the steps of sampling an underwater distorted image, discretizing the underwater distorted image to reduce the dimension space size of an input image, setting a distortion pixel of the underwater distorted image to be 0 according to a mask image M' of the underwater distorted image, inputting the distortion pixel into a transform architecture, and performing Gibbs sampling on an output probability image to obtain a low-resolution prior image of a restored image; s4: an Encode-Decoder structure is adopted as a main framework of a recovery network, and three branched inputs are constructed at a coding end of the recovery network; s5: and constructing a DAM distortion aggregation module, guiding the recovery network by using the intermediate layer information of the distortion detection network, and finally obtaining the recovery image of the underwater distortion image. The method can effectively improve the recovery effect of the underwater distorted image.
Description
Technical Field
The invention relates to the field of image restoration, in particular to an underwater image restoration method based on distortion guidance.
Background
Human beings are crossing the information age, and images play an important role in daily life and work as important carriers of information. The underwater image is used as a carrier of ocean information and is an important way for human beings to explore the ocean information. The underwater image is widely applied to the fields of marine energy exploration, underwater rescue, marine ecological protection, marine organism research and the like. However, compared to a wireless channel, in view of a complex and variable marine environment, an underwater acoustic channel often has severe conditions such as fast speed, narrow bandwidth, fast fading, and the like, so that the distortion degree of an underwater transmission image is far greater than that of an image transmitted in the wireless channel. Therefore, images transmitted through the hydroacoustic channel tend to be severely distorted.
Existing underwater degraded image quality improvement strategies can be divided into enhancement and restoration according to whether the strategies are based on an imaging model or not. Enhancement algorithms selectively highlight features of interest in the image or suppress (mask) some unwanted features in the image by adding some information or transform data to the original image by some means, allowing the image to match the visual response characteristics regardless of the physical imaging process. In contrast, the restoration algorithm performs mathematical modeling on the degradation process of the underwater image, and inverts the degradation process by estimating model parameters, so as to restore the image with reduced quality, thereby obtaining a clear underwater image. However, the existing algorithm cannot locate the distortion region and the distortion type of the underwater transmission distorted image, which has a great influence on the image restoration result.
Disclosure of Invention
In view of the above, an object of the present invention is to provide an underwater image restoration method based on distortion guidance, which can effectively improve the restoration effect of an underwater distorted image.
In order to achieve the purpose, the invention adopts the following technical scheme:
an underwater image restoration method based on distortion guidance comprises the following steps:
step S1, acquiring an underwater original image GT, an underwater distortion image I and a distortion mask image M representing an image distortion area and a distortion type, and constructing an underwater multi-type distortion image database;
step S2, training a distortion detection network through I and M in an underwater distortion image database, and predicting a mask image M' of a distortion image;
s3: the method comprises the steps of sampling an underwater distorted image, discretizing the underwater distorted image to reduce the dimension space size of an input image, setting a distortion pixel of the underwater distorted image to be 0 according to a mask image M' of the underwater distorted image, inputting the distortion pixel into a transform architecture, and performing Gibbs sampling on an output probability image to obtain a low-resolution prior image of a restored image;
s4: an Encode-Decoder structure is adopted as a main framework of a recovery network, and three branched inputs are constructed at a coding end of the recovery network;
s5: and constructing a DAM distortion aggregation module, guiding the recovery network by using the intermediate layer information of the distortion detection network, and finally obtaining the recovery image of the underwater distortion image.
Further, in step S1, the underwater original image GT is photographed through an underwater environment, different distortion types are processed on the underwater original image GT to obtain distortion images and labels of different distortion types, and in the process of processing the original image to obtain different types of distortion images, a distortion area and a distortion type of the image are recorded at the same time, and a distortion detection label mask image M is generated.
Further, the distortion types comprise underwater transmission distortion, ocean snow distortion and motion blur distortion,
for underwater transmission distortion, the underwater transmission distortion is obtained by building a simulated underwater acoustic communication system and transmitting an original image;
for the ocean snow distortion, the ocean snow distortion image layers with different degrees are artificially added to an original image to obtain the ocean snow distortion image layers;
for the target motion blur distortion, a closed area is randomly selected in an original image and motion blur processing is carried out.
Further, the step S2 is specifically:
step S21, based on semantic segmentation algorithm, carrying out pixel-level classification on the distorted image, positioning a distortion region and a distortion type, and adopting U-net as a backbone of the distortion detection network;
in step S22, the distortion detection network is trained by I and M in the underwater multi-type distorted image database, and a mask image M' of a distorted image is predicted.
Further, the step S3 is specifically:
step S31, down-sampling the distorted image of m × 3 to (m/8) × 3, and performing discretization processing on the distorted image;
step S32, according to the distortion mask image M' obtained in the step S2, setting a distortion pixel point corresponding to the discrete image of the down-sampling image to be 0, and inputting the distortion pixel point into a transform after token embedding to obtain a probability image of the low-resolution restoration image;
and step S33, Gibbs sampling is carried out on the probability map, and a low-resolution prior image of the restored image is obtained.
Further, the discretization treatment specifically comprises the following steps: using a clustering algorithm for pixel points of the underwater image to obtain n central sample points which represent pixel values of n three channels which most frequently appear in the underwater image; and for each pixel point of the (m/8) × 3 downsampled image, selecting the central sample point closest to the Euclidean distance of the pixel point for replacement, and reducing the dimensional space of the image from the original (m/8) × (256^3) to (m/8) × (m/8) < n >.
Further, the step S4 is specifically:
step S41, adopting Encoder-Decoder structure as main framework of recovery network;
step S42: the encoding end of the restoration network constructs three branched inputs:
the input of branch 1 is a distortion mask map M' for the distortion detection network location;
the inputs to branch 2 are as follows:
I M =I⊙(1-M')
where I is a distorted image, M' is the predicted distortion mask map, which is a matrix dot product.
The input of branch 3 is the low-resolution prior image of the restored image obtained in step S3.
And step S43, performing concat operation on the three inputs in the channel dimension and inputting the three inputs to the encoding end of the recovery network.
Further, the step S5 is specifically:
step S51: extracting middle layer information of a coding end in the forward propagation process of the distortion detection network, wherein the middle layer information is respectively C, H, W, (2C), H/2, (4C), H/4, W/4) feature graphs with three different scales;
step S52: constructing a DAM distortion aggregation module, extracting three-layer characteristic graphs, performing characteristic fusion of different scales, and performing step-by-step processing by using a convolutional layer and a relu activation function, wherein the output of each stage acts on the input of the next stage as follows:
f i '=DAM(concat[f i ,f i ' -1 ])
f i characteristic diagram of i layer at coding end of L distortion detection network, f i ' is the corresponding output, with dimensions the same as the input dimensions;
step S53: and receiving the output of the DAM distortion aggregation module in three different scales at a decoding end of the recovery network, and performing concat operation on the three layers of feature maps with the same dimension as the decoding end in the channel dimension. The DAM outputs information of different scales from the middle layer of the distortion detection network, and the information can distinguish whether the characteristics are distorted or not, and defines the network restoration direction, so that the restoration network is guided to accurately restore the distorted image.
Further, the following loss functions are set in the process of training the recovery network R:
I R is the restoration result of the restoration network R prediction, GT is the underwater original image, and the two are calculatedLoss of
And D is a discriminator, a game is formed by using the discriminator and a recovery network, the confrontation loss is calculated and generated, Nash balance is achieved through training optimization, and the final total loss function is as follows:
wherein alpha is 1 、α 2 Is a preset value.
Compared with the prior art, the invention has the following beneficial effects:
the method can effectively improve the recovery effect of the underwater distorted image.
Drawings
FIG. 1 is a general framework of the method of the present invention;
FIG. 2 is a distortion detection database in accordance with an embodiment of the present invention;
FIG. 3 is a flow chart of input branch c according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating a distorted image restoration result according to an embodiment of the present invention.
Detailed Description
The invention is further explained by the following embodiments in conjunction with the drawings.
Referring to fig. 1, the present invention provides an underwater image restoration method based on distortion guidance, which includes the following steps:
s1, acquiring an underwater original image GT, an underwater distorted image I and a distortion mask image M representing an image distortion area and a distortion type, and constructing an underwater multi-type distorted image database, wherein the method specifically comprises the following steps:
and constructing an underwater multi-type distortion image database which comprises an underwater original image GT, an underwater distortion image I and a distortion mask image M representing an image distortion area and a distortion type. The underwater original image is shot from a real underwater environment. Processing the underwater original image in different distortion types to obtain different distortion images and labels, as shown in fig. 2; in the process of processing an original image to obtain different types of distorted images, simultaneously recording a distortion area and a distortion type of the image, and generating a distortion detection label mask image M;
preferably, in the embodiment, the distortion type selects underwater transmission distortion, motion blur distortion, ocean snow distortion; for underwater transmission distortion, the underwater transmission distortion is obtained by building a simulated underwater acoustic communication system and transmitting an original image; for target motion blur distortion, a closed area is randomly selected from an original image and motion blur processing is carried out, so that a distortion detection network learns to identify blur characteristics; for the ocean snow distortion, the ocean snow distortion image layers with different degrees are artificially added to an original image to obtain the ocean snow distortion image layers;
step S2, training a distortion detection network through I and M in an underwater distortion image database, and predicting a mask image M' of a distortion image;
preferably, in this embodiment, the step S2 specifically includes the following steps:
s21, applying the idea of semantic segmentation to distortion detection, classifying the distorted images at a pixel level so as to locate distortion areas and distortion types, and adopting U-net as a backbone of a distortion detection network;
s22, training a distortion detection network through I and M in an underwater distortion image database, thereby accurately predicting a mask image M' of a distortion image;
s3: downsampling and discretizing the distorted image to reduce the dimension space size of the input image, setting the distorted pixel of the image to be 0 according to M', inputting the distorted pixel into a transform architecture, and performing Gibbs sampling on the output probability map to obtain a low-resolution prior image of the restored image, wherein the flow of the step S3 is shown in FIG. 3;
preferably, in this embodiment, the step S3 includes the following steps:
s31, down-sampling the 256 × 3 distorted image to 32 × 3, and then discretizing it to further reduce the input dimensional space;
s32, the discretization concrete process is as follows: and (3) using a clustering algorithm for pixel points of a large number of underwater images to obtain 512 central sample points which represent 512 three-channel pixel values which most frequently appear in the underwater images. And selecting the central sample point closest to the Euclidean distance of each pixel point of the 32X 3 down-sampling image for replacement, thereby realizing discretization of the image and reduction of the dimension space. The dimensional space of the image will be 32 x 512 from the original 32 x (256 x 3);
s33, according to the distortion mask image M' obtained in S2, setting a distortion pixel point corresponding to the discrete image of the down-sampled image to be 0, and inputting the distortion pixel point into a transform after token embedding to obtain a probability image of the low-resolution restoration image;
and S34, Gibbs sampling is carried out on the probability map, and a low-resolution prior image of the restored image is obtained.
S4: an Encode-Decoder structure is adopted as a main framework of a recovery network, and three branched inputs are constructed at a coding end of the recovery network;
preferably, in this embodiment, the step S4 includes the following steps:
s41, adopting an Encoder-Decoder structure as a main framework of a recovery network;
s42, as shown in FIG. 1, the coding end of the recovery network constructs the input of three branches, the input of branch (a) is the distortion mask map M' of the distortion detection network positioning; the inputs to branch (b) are as follows:
I M =I⊙(1-M')
i is a distorted image, M' is a predicted distortion mask map, which is a matrix dot product.
The input of branch (c) is the low-resolution prior image of the restored image obtained in step S3.
And S43, performing concat operation on the three inputs in the channel dimension and inputting the concat operation to the encoding end of the recovery network.
S5: and constructing a DAM distortion aggregation module, guiding the recovery network by using the intermediate layer information of the distortion detection network, and finally obtaining the recovery image of the underwater distortion image.
Preferably, in this embodiment, the step S5 includes the following steps:
s51, extracting middle layer information of a coding end in the forward propagation process of the distortion detection network, wherein the middle layer information is respectively feature maps of 64 × 256, 128 × 128 and 256 × 64 with three different scales;
s52, designing and utilizing a DAM distortion aggregation module to perform feature fusion of different scales on the extracted three-layer feature graph, and utilizing a convolution layer and a relu activation function to perform step-by-step processing, wherein the output of each step acts on the input of the next step as follows:
f i '=DAM(concat[f i ,f i ' -1 ])
f i a characteristic diagram representing the i-th layer of the L coding end of the distortion detection network, f i ' is the corresponding output, with dimensions the same as the input dimensions;
s53, receiving the output of the DAM distortion aggregation module in three different scales at the decoding end of the recovery network, and performing concat operation on the three-layer characteristic diagram in the same dimension as the decoding end in the channel dimension. The information of different scales output by the DAM comes from the middle layer of the distortion detection network, and the information can distinguish whether the characteristics are distorted or not, and defines the direction of network restoration, so that the restoration network is guided to accurately restore the distorted image, and the restoration result of the distorted image is as shown in FIG. 4;
s54, setting the following loss function in the process of training the recovery network R:
I R is the restoration result of the restoration network R prediction, GT is the underwater original image, and the two are calculatedLoss of power
Where D is a discriminator, and the countermeasure loss is calculated using the concept of GAN. The final total loss function is as follows:
through experiments, set alpha 1 =1.2,α 2 =0.1。
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (9)
1. An underwater image restoration method based on distortion guidance is characterized by comprising the following steps:
step S1, acquiring an underwater original image GT, an underwater distortion image I and a distortion mask image M representing an image distortion area and a distortion type, and constructing an underwater multi-type distortion image database;
step S2, training a distortion detection network through I and M in an underwater distortion image database, and predicting a mask image M' of a distortion image;
step S3: the method comprises the steps of down-sampling an underwater distorted image, discretizing the underwater distorted image to reduce the dimension space size of an input image, setting a distorted pixel of the underwater distorted image to be 0 according to a mask image M' of the underwater distorted image, inputting the distorted pixel into a Transformer architecture, performing Gibbs sampling on an output probability image, and obtaining a low-resolution prior image of a restored image;
step S4: an Encode-Decoder structure is adopted as a main framework of a recovery network, and three branched inputs are constructed at a coding end of the recovery network;
step S5: and constructing a DAM distortion aggregation module, guiding the recovery network by using the intermediate layer information of the distortion detection network, and finally obtaining the recovery image of the underwater distortion image.
2. The method for restoring an underwater image based on distortion guidance according to claim 1, wherein the step S1 is specifically to capture the underwater original GT through an underwater environment, perform different distortion type processing on the underwater original GT to obtain distorted images and labels of different distortion types, record a distortion area and a distortion type of the image simultaneously during the process of processing the original to obtain different types of distorted images, and generate the distortion detection label mask map M.
3. The distortion-guided underwater image restoration method according to claim 2, wherein the distortion types include underwater transmission distortion, snow distortion, motion blur distortion,
for underwater transmission distortion, the underwater transmission distortion is obtained by building a simulated underwater acoustic communication system and transmitting an original image;
for the ocean snow distortion, the ocean snow distortion image layers with different degrees are artificially added to an original image to obtain the ocean snow distortion image layers;
for the target motion blur distortion, a closed area is randomly selected in an original image and motion blur processing is carried out.
4. The method for restoring an underwater image based on distortion guidance according to claim 1, wherein the step S2 is specifically as follows:
step S21, based on semantic segmentation algorithm, carrying out pixel-level classification on the distorted image, positioning a distortion area and a distortion type, and adopting U-net as a backbone of a distortion detection network;
in step S22, the distortion detection network is trained by I and M in the underwater multi-type distorted image database, and a mask image M' of a distorted image is predicted.
5. The method for restoring an underwater image based on distortion guidance according to claim 1, wherein the step S3 is specifically as follows:
step S31, down-sampling the distorted image of m × 3 to (m/8) × 3, and performing discretization processing on the distorted image;
s32, according to the distortion mask image M' obtained in S2, setting a distortion pixel point corresponding to the discrete image of the down-sampled image to be 0, and inputting the distortion pixel point to a transform after tokeneaddressing to obtain a probability image of the low-resolution restored image;
and step S33, Gibbs sampling is carried out on the probability map, and a low-resolution prior image of the restored image is obtained.
6. A distortion-guided underwater image restoration method according to claim 5, wherein the underwater image restoration method is characterized in that
The discretization treatment process comprises the following specific steps: using a clustering algorithm for pixel points of the underwater image to obtain n central sample points which represent pixel values of n three channels which are most frequently appeared in the underwater image; and for each pixel point of the (m/8) × 3 downsampled image, selecting the central sample point closest to the Euclidean distance of the pixel point for replacement, and reducing the dimensional space of the image from the original (m/8) × (256^3) to (m/8) × (m/8) < n >.
7. The method for restoring an underwater image based on distortion guidance according to claim 1, wherein the step S4 is specifically as follows:
step S41, adopting Encoder-Decoder structure as main framework of recovery network;
step S42: the encoding end of the restoration network constructs three branched inputs:
the input of branch 1 is a distortion mask map M' for the distortion detection network location;
the inputs to branch 2 are as follows:
I M =I⊙(1-M')
where I is a distorted image, M' is a predicted distortion mask map, and an | _ is a matrix dot product.
The input of branch 3 is the low-resolution prior image of the restored image obtained in step S3.
And step S43, performing concat operation on the three inputs in the channel dimension and inputting the three inputs to the encoding end of the recovery network.
8. The method for restoring an underwater image based on distortion guidance according to claim 1, wherein the step S5 is specifically as follows:
step S51: extracting middle layer information of a coding end in the forward propagation process of the distortion detection network, wherein the middle layer information is respectively C, H, W, (2C), H/2, (4C), H/4, W/4) feature graphs with three different scales;
step S52: constructing a DAM distortion aggregation module, performing feature fusion of different scales on the extracted three-layer feature graph, and performing step-by-step processing by using a convolutional layer and a relu activation function, wherein the output of each step acts on the input of the next step as follows:
f i '=DAM(concat[f i ,f i ' -1 ])
f i a characteristic diagram representing the i-th layer of the L coding end of the distortion detection network, f i ' is the corresponding output, with dimensions the same as the input dimensions;
step S53: receiving the output of the DAM distortion aggregation module with three different scales at a decoding end of a recovery network, and performing concat operation on a three-layer characteristic diagram with the same dimension as the decoding end in a channel dimension; the DAM outputs information of different scales from a middle layer of a distortion detection network, and the information can distinguish whether the characteristics are distorted or not, and defines the network restoration direction, so that the restoration network is guided to accurately restore the distorted image.
9. The underwater image restoration method based on the distortion guidance according to claim 1, wherein the following loss functions are set in the process of training the restoration network R:
L l1 =E(||I R -GT|| 1 )
I R is the restoration result predicted by the restoration network R, GT is the underwater original image, and the two calculate L l1 Loss of power
And D is a discriminator, a game is formed by using the discriminator and a recovery network, the confrontation loss is calculated and generated, Nash balance is achieved through training optimization, and the final total loss function is as follows:
wherein alpha is 1 、α 2 Is a preset value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210782462.8A CN115049562B (en) | 2022-07-05 | 2022-07-05 | Underwater image restoration method based on distortion guidance |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210782462.8A CN115049562B (en) | 2022-07-05 | 2022-07-05 | Underwater image restoration method based on distortion guidance |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115049562A true CN115049562A (en) | 2022-09-13 |
CN115049562B CN115049562B (en) | 2024-09-03 |
Family
ID=83165616
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210782462.8A Active CN115049562B (en) | 2022-07-05 | 2022-07-05 | Underwater image restoration method based on distortion guidance |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115049562B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111275647A (en) * | 2020-01-21 | 2020-06-12 | 南京信息工程大学 | Underwater image restoration method based on cyclic generation countermeasure network |
US20210217144A1 (en) * | 2020-01-14 | 2021-07-15 | International Business Machines Corporation | Under water image color correction |
CN113205468A (en) * | 2021-06-01 | 2021-08-03 | 桂林电子科技大学 | Underwater image real-time restoration model based on self-attention mechanism and GAN |
CN113935916A (en) * | 2021-10-12 | 2022-01-14 | 西安电子科技大学 | End-to-end underwater image restoration method based on ambient light perception |
-
2022
- 2022-07-05 CN CN202210782462.8A patent/CN115049562B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210217144A1 (en) * | 2020-01-14 | 2021-07-15 | International Business Machines Corporation | Under water image color correction |
CN111275647A (en) * | 2020-01-21 | 2020-06-12 | 南京信息工程大学 | Underwater image restoration method based on cyclic generation countermeasure network |
CN113205468A (en) * | 2021-06-01 | 2021-08-03 | 桂林电子科技大学 | Underwater image real-time restoration model based on self-attention mechanism and GAN |
CN113935916A (en) * | 2021-10-12 | 2022-01-14 | 西安电子科技大学 | End-to-end underwater image restoration method based on ambient light perception |
Non-Patent Citations (2)
Title |
---|
尹芳;陈田田;吴锐;付自如;于晓洋;: "一种结合暗通道先验和图像融合的水下图像复原算法", 小型微型计算机系统, no. 11, 15 November 2017 (2017-11-15), pages 177 - 182 * |
林祖鑫: "A Distortion-Guided Strategy for Efficient Underwater Image Restoration", OCEANS 2022, HAMPTON ROADS, 20 October 2022 (2022-10-20), pages 7924 - 7937 * |
Also Published As
Publication number | Publication date |
---|---|
CN115049562B (en) | 2024-09-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111639692B (en) | Shadow detection method based on attention mechanism | |
CN113688723B (en) | Infrared image pedestrian target detection method based on improved YOLOv5 | |
CN110210551B (en) | Visual target tracking method based on adaptive subject sensitivity | |
CN111126453B (en) | Fine-grained image classification method and system based on attention mechanism and cut filling | |
CN113936339B (en) | Fighting identification method and device based on double-channel cross attention mechanism | |
CN111783705B (en) | Character recognition method and system based on attention mechanism | |
WO2021218899A1 (en) | Method for training facial recognition model, and method and apparatus for facial recognition | |
CN112150493B (en) | Semantic guidance-based screen area detection method in natural scene | |
CN109117894B (en) | Large-scale remote sensing image building classification method based on full convolution neural network | |
CN111476133B (en) | Unmanned driving-oriented foreground and background codec network target extraction method | |
CN113743269A (en) | Method for identifying video human body posture in light weight mode | |
CN111310766A (en) | License plate identification method based on coding and decoding and two-dimensional attention mechanism | |
WO2020244279A1 (en) | Method and device for identifying video | |
CN112418032A (en) | Human behavior recognition method and device, electronic equipment and storage medium | |
CN116129326A (en) | Video action recognition method based on local and global attention mechanisms | |
CN113192124A (en) | Image target positioning method based on twin network | |
CN114677536B (en) | Pre-training method and device based on Transformer structure | |
CN115471718A (en) | Construction and detection method of lightweight significance target detection model based on multi-scale learning | |
CN115272957A (en) | Crowd counting method based on feature interaction | |
CN114202473A (en) | Image restoration method and device based on multi-scale features and attention mechanism | |
CN113379606A (en) | Face super-resolution method based on pre-training generation model | |
CN117058392A (en) | Multi-scale Transformer image semantic segmentation method based on convolution local enhancement | |
CN115049562B (en) | Underwater image restoration method based on distortion guidance | |
CN116311026A (en) | Classroom scene identity recognition method based on multi-level information fusion Transformer | |
CN115482463A (en) | Method and system for identifying land cover of mine area of generated confrontation network |
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 |