CN111709900A - High dynamic range image reconstruction method based on global feature guidance - Google Patents
High dynamic range image reconstruction method based on global feature guidance Download PDFInfo
- Publication number
- CN111709900A CN111709900A CN202010748229.9A CN202010748229A CN111709900A CN 111709900 A CN111709900 A CN 111709900A CN 202010748229 A CN202010748229 A CN 202010748229A CN 111709900 A CN111709900 A CN 111709900A
- Authority
- CN
- China
- Prior art keywords
- dynamic range
- range image
- high dynamic
- training
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000000605 extraction Methods 0.000 claims abstract description 21
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 14
- 238000013507 mapping Methods 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 19
- 230000003416 augmentation Effects 0.000 claims description 7
- 238000013434 data augmentation Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000013136 deep learning model Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000000844 transformation Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 11
- 238000002474 experimental method Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 description 6
- 230000000007 visual effect Effects 0.000 description 5
- 230000004927 fusion Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 229920006395 saturated elastomer Polymers 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000016776 visual perception 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
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20208—High dynamic range [HDR] image processing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a high dynamic range image reconstruction method based on global feature guidance. Firstly, a high dynamic range image reconstruction model is designed according to a convolutional neural network self-encoder structure, and the high dynamic range image reconstruction model comprises a local feature extraction branch, a global feature extraction branch and a reconstruction branch. Then, the original high dynamic range image is subjected to Reinhard tone mapping algorithm to obtain a low dynamic range image similar to that shot in reality, and a training set is established by using the low dynamic range image and the high dynamic range image. And finally, carrying out constrained reconstruction on the high dynamic range image through the proposed mixed loss function, and training the proposed model by using a random gradient descent algorithm to obtain a model for reconstructing the low dynamic range image to the high dynamic range image. The method disclosed by the invention is used for carrying out experiments on the disclosed data set, and has higher robustness and accuracy.
Description
Technical Field
The invention relates to a high dynamic range image reconstruction method, in particular to a high dynamic range image reconstruction method based on global feature guidance, and belongs to the field of image processing and reconstruction technology utilization.
Background
The dynamic range of an image refers to the ratio of the maximum magnitude value to the minimum luminance value in the image. The pixel value of the High Dynamic Range (HDR) image is matched with the brightness value of the real scene, so that a highlight region and a dark region can be shown better at the same time, and compared with a traditional Low Dynamic Range (LDR) image, the HDR image has richer detail information and wider color gamut, and the quality of the image is greatly improved. The high dynamic range image reconstruction is an image processing technology for reconstructing a high dynamic range image from a low dynamic range image, and has wide application in the fields of medical imaging, movie and television rendering, rocker detection and the like.
The traditional high dynamic range image reconstruction method usually performs simple mapping work on a low dynamic range image through five steps of linearization, dynamic range expansion, saturated region reconstruction, noise removal and color correction. Due to the relatively fixed parameters, excessive loss of information in a saturated region and the like, the traditional high dynamic range image reconstruction method is easy to cause color distortion, artifact and unnatural artificial traces.
In recent years, Convolutional Neural Networks (CNNs) have been widely used in the field of computer vision such as object detection and image segmentation, and have been remarkably developed. The convolutional neural network is very suitable for the problem of complex nonlinear mapping, and can well complement the part with lost information through learning, so that the convolutional neural network is very suitable for a high-dynamic-range image reconstruction task. Endo et al apply a convolutional neural network to high dynamic range image reconstruction for the first time, generate a set of low dynamic range images of different exposure degrees from a single low dynamic range image through a set of convolutional neural networks of a self-encoder structure, and synthesize a high dynamic range image by using the generated images. Marnerides et al propose an end-to-end three-branch convolutional neural network structure that accomplishes the mapping between a single low dynamic range image and a single high dynamic range image by learning the features of different pixel levels and fusing. The existing high dynamic range image reconstruction method based on the convolutional neural network solves the problems of weak robustness, poor generalization capability and the like of the traditional high dynamic range image reconstruction method, but still has the problems of colored distortion, insufficient detail reconstruction, poor visual effect and the like.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a high dynamic range image reconstruction method based on global feature guidance. The method can effectively improve the quality value, the peak signal-to-noise ratio and the structural similarity of the reconstructed high dynamic range image, and has better effect on subjective visual experience.
In order to achieve the purpose, the invention has the following conception:
firstly, a high dynamic range image reconstruction model is designed according to a convolutional neural network self-encoder structure, and the high dynamic range image reconstruction model comprises a local feature extraction branch, a global feature extraction branch and a reconstruction branch. The original high dynamic range image is then passed through Reinhard [ 2 ]1]And obtaining a low dynamic range image similar to the real shooting by using a tone mapping algorithm, and establishing a training set by using the low dynamic range image and the high dynamic range image. And finally, carrying out constrained reconstruction on the high dynamic range image through the proposed mixed loss function, and training the proposed model by using a random gradient descent algorithm to obtain a model for reconstructing the low dynamic range image into the high dynamic range image, namely the high dynamic range image reconstruction method guided by the global characteristics.
According to the conception, the invention adopts the following technical scheme:
a high dynamic range image reconstruction method based on global feature guidance comprises the following steps:
step 3, training set establishment: performing Reinhard tone mapping on the large-scale high dynamic range image sample obtained in the step 2 to obtain a low dynamic range image similar to that shot in reality, respectively cutting the high dynamic range image and the low dynamic range image, and trying the small blocks to construct a data set;
and 5, reconstructing a high dynamic range image: in the model obtained by training in the step 4, low dynamic range images with different sizes are input, namely, the low dynamic range images can be reconstructed and restored into corresponding high dynamic range images.
The method mainly considers the important role of the global features on the high dynamic range image reconstruction, and extracts the global features through the independent global feature extraction branch to guide the local features to carry out the high dynamic range image reconstruction, so that the reconstructed high dynamic range image has no obvious artificial traces on the whole visual perception, and the color tone is more natural. Meanwhile, a new mixed loss function is provided, the reconstruction of the high dynamic range image is restrained from three angles of pixel value, pixel gradient and color similarity, the phenomena of artifact and color aberration existing in the past work are reduced, and the quality of the reconstructed high dynamic range image is improved.
Compared with the prior art, the invention has the following obvious and prominent substantive characteristics and remarkable advantages:
1. the method of the invention fully considers the guiding effect of the image global feature on the local feature, and provides a high dynamic range image reconstruction model based on the global feature guidance.
2. The method provides a brand-new mixed loss function, and the reconstructed high dynamic range image is constrained from three angles of pixel values, pixel gradients and color similarity.
3. The method utilizes the self-encoder structure to construct a high dynamic range image reconstruction model, and utilizes the encoder structure to extract and compress features, thereby effectively reducing the training difficulty of the model and the complexity of the algorithm.
Drawings
FIG. 1 is a network structure diagram of the high dynamic range image reconstruction method based on global feature guidance.
FIG. 2 is a comparison graph of the visualization effect of the "Park" graph reconstructed by the method of the present invention and other methods.
FIG. 3 is a comparison graph of the visualization effect of the "City" graph reconstructed by the method of the present invention and other methods.
FIG. 4 is a graph comparing the visual effect of the reconstruction of the "Courtyard" graph using the mixed loss function constraint and without the mixed loss function constraint according to the present invention.
Detailed Description
The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings:
a block diagram of a network structure of a global feature guided high dynamic range image reconstruction model according to this embodiment is shown in fig. 1. The method is realized in a program simulation under the environment of Ubuntu 16.04 and PyTorch. Firstly, a high dynamic range image reconstruction model is designed according to a convolutional neural network self-encoder structure, and the high dynamic range image reconstruction model comprises a local feature extraction branch, a global feature extraction branch and a reconstruction branch. And then, performing data augmentation on the original high dynamic range image, obtaining a low dynamic range image by a Reinhard tone mapping method, and randomly cutting to generate a corresponding high and low dynamic range image pair to form a data set used for network training. And finally, training the network model by using a random gradient descent algorithm, and constraining the reconstructed high dynamic range image from three angles of pixel values, pixel gradients and color similarities through a mixed loss function to obtain a model for reconstructing the low dynamic range image to the high dynamic range image, namely the high dynamic range image reconstruction model guided by the global characteristics.
The method specifically comprises the following steps:
step 3, training set establishment: performing Reinhard tone mapping on the large-scale high dynamic range image sample obtained in the step 2 to obtain a low dynamic range image similar to that shot in reality, respectively cutting the high dynamic range image and the low dynamic range image, and trying the small blocks to construct a data set;
and 5, reconstructing a high dynamic range image: and (4) inputting the low dynamic range images with different sizes into the model obtained by training in the step (4), and reconstructing and restoring the low dynamic range images into corresponding high dynamic range images.
In step 1, a high dynamic range image reconstruction model based on global feature guidance is provided, which includes a local feature extraction branch, a global feature extraction branch and a reconstruction branch, and a network structure is shown in fig. 1. The local feature extraction branch is of an encoder structure, the local feature is extracted and compressed by using a convolution kernel with the step length of 2 multiplied by 3, the consumption of video memory is greatly reduced, and the calculation formula is as follows:
wherein L isH×W×3Color low dynamic range image with dimension H × W × 3, f representing inputdiRepresenting the ith layer feature compression Module, FlocolFor the final extracted local features, the dimension is
Before entering global feature extraction, an input low dynamic range image is deformed into 256 × 256 × 3 dimensions, a front 4-layer feature compression module of a global feature extraction branch is the same as a local feature extraction branch, so that parameter quantity of a network is greatly reduced, and finally, global features are obtained through 8 groups of convolution kernels of 3 × 3 and 1 convolution kernel of 4 × 4, and are expressed as follows:
wherein L is256×256×3For the input warped low dynamic range image, fgFor parts of the global feature extraction branch different from the local feature extraction branch, FglobolFor extracted global features, the dimension is 1 × 1 × 512.
At a decoding end, the global features and the local features are fused, and the global features are used for guiding the local features to complete reconstruction of the high dynamic range image. The global feature firstly performs padding operation, then is cascaded with the local feature, and completes the guiding effect of the global feature on the local feature through a fusion layer, which is expressed as follows:
in the above formula, CpIndicating a padding operation, fupAn up-sampling module is shown that is,denotes cascade operation, CfDenotes a fusion layer, FfusionRepresenting the fused features.
Finally, a series of upsampling operations are carried out on the fused features, and shallow features are introduced in a jumping connection mode for fusion, so that feature information extracted from different levels in the network is effectively utilized, and finally, a channel is recovered through convolution of 3 x 3, and reconstruction of a high dynamic range image is completed.
In step 2, the training set images 784 used by us are composed of different types of high dynamic range images, such as indoor, outdoor, natural scene, human scene, and the like. The image is enlarged mainly by inversion and mirror image. A training image 6 times of the original atlas is obtained through image augmentation, and the training atlas is effectively expanded.
In the step 3, Reinhard tone mapping is performed on the augmented training atlas obtained in the step 2 to obtain a low dynamic range image similar to that of shooting in real life, in order to improve training efficiency, normalization processing is performed on the corresponding high dynamic range image and low dynamic range image, pixel values of the high dynamic range image and the low dynamic range image are linearly projected between [0 and 1], random clipping is performed on the normalized image pair, and the size of clipping is 256 × 256, so that a training set used by a training network is formed.
In step 4, training a high dynamic range on the training set obtained in step 3The image reconstruction model adopts a Stochastic Gradient Descent (Stochastic Gradient Description) algorithm as an optimization algorithm, the momentum parameter is set to be 0.9, the batch size is set to be 10, and the initial learning rate is 1 × 10-5Finally, the learning rate is adjusted to 1 × 10-7. We constrain the reconstructed high dynamic range image from three angles of pixel value, pixel gradient, color similarity by a mixture loss function, which is expressed as follows:
wherein the content of the first and second substances,in order to be a function of the loss of mixing,in order to be a multi-scale pixel loss function,in order to be a function of the loss of color similarity,for the pixel gradient penalty function, α and β are hyperparameters with values of 5 and 10000.
In the above formula, GT represents the labeled high dynamic range image, GT 'represents the labeled high dynamic range image after down-sampling, H and H' represent the reconstructed high dynamic range image and the reconstructed high dynamic range image after down-sampling, respectively, | | |1Representing the first order norm.
In the above formula, IiDenotes the ith color pixel vector, K denotes a pixel summary, ∈ is 1 × 10-8Is inclined toSetting to prevent the denominator from being 0, | × | | non-conducting phosphor2Representing the second order norm.
In the above formula, dxAnd dyRespectively, the transverse and longitudinal derivatives.
After the training is completed, a model for reconstructing the low dynamic range image to the high dynamic range image can be obtained.
In step 5, the model trained in step 4 reconstructs the low dynamic range image input in any dimension into a high dynamic range image.
The following experiment was performed by randomly extracting 50 high dynamic range images that are not used in training on a public data set to evaluate the generalization ability and the advancement of the global feature guidance-based high dynamic range image reconstruction method proposed by the present invention. The environment of the experiment is an Ubuntu 16.04 operating system, the memory is 16GB, the GPU is GeForce 1070, and the deep learning framework is a PyTorch platform. Using a perceived uniform peak signal-to-noise ratio [ alpha ] [ alpha ]2](Perceptiall Uniform-Peak Signal to noise Ratio, PU-PSNR), Structural Similarity coefficient (SSIM), and HDR-VDP-2[, [ 2 ] ]3]The provided mass fraction Q is used as an evaluation index. The higher the PU-PSNR value is, the closer the SSIM is to 1, and the higher the Q value is, the closer the high dynamic range image reconstructed by the model is to the original image is, and the higher the reconstruction quality is. Fig. 2-3 compare the visual reconstruction effect comparison of different algorithms on the test picture, and fig. 4 compares whether the mixed loss function is used on the test picture.
TABLE 1
Method | PU-PSNR | SSIM | Q |
Akyuz[4] | 17.82 | 0.649 | 54.81 |
Kuo[5] | 15.76 | 0.599 | 51.99 |
Huo[6] | 13.72 | 0.539 | 47.05 |
Kov[7] | 17.19 | 0.644 | 53.01 |
Masia[8] | 17.82 | 0.655 | 55.29 |
DrTMO[9] | 19.54 | 0.601 | 58.26 |
HDRCNN[10] | 19.41 | 0.719 | 56.56 |
ExpandNet[11] | 19.12 | 0.722 | 55.68 |
Ours | 19.57 | 0.733 | 57.11 |
TABLE 2
Method | PU-PSNR | SSIM | Q |
No mixing loss function | 19.31 | 0.714 | 54.23 |
With mixing loss function | 19.57 | 0.733 | 57.11 |
Table 1 shows the comparison of the present invention with other advanced method indexes, wherein the algorithm with the best experimental results is shown in bold font and underlined, and the second best algorithm is shown only in bold. It can be seen from table 1 that the method of the present invention has better robustness and accuracy in high dynamic range image reconstruction. Table 2 shows the influence of the hybrid loss function provided by the present invention on the high dynamic range image reconstruction, and it can be seen that the hybrid loss function provided by the present invention can greatly improve the quality of the high dynamic range image reconstruction. The experiments show that the high dynamic range image reconstruction method based on the global feature guidance has the advantages of being advanced in indexes and visual effects, low in calculation complexity and good in real-time performance.
Claims (1)
1. A high dynamic range image reconstruction method based on global feature guidance is characterized by comprising the following steps:
step 1, model establishment: designing a high dynamic range image reconstruction model according to the structure of a convolutional neural network self-encoder, wherein the high dynamic range image reconstruction model comprises a local feature extraction branch, a global feature extraction branch and a reconstruction branch;
step 2, image augmentation: the training of the deep learning model depends heavily on a large-scale data set, and the image augmentation technology generates similar training samples by performing a series of simple transformations on training pictures, so that the scale of the training samples is improved; the large-scale data set can effectively improve the robustness and generalization capability of the model, and the scale of the high-dynamic-range image sample is enlarged by using a method of inversion and mirror image data augmentation;
step 3, training set establishment: performing Reinhard tone mapping on the large-scale high dynamic range image sample obtained in the step 2 to obtain a low dynamic range image similar to that shot in reality, respectively cutting the high dynamic range image and the low dynamic range image, and trying the small blocks to construct a data set;
step 4, training a multi-column convolution neural network model: training a high dynamic range image reconstruction model on the training set obtained in the step 3, using a random gradient descent optimization method, carrying out constraint through a mixed loss function, and obtaining a model for reconstructing a low dynamic range image into a high dynamic range image after training is finished;
and 5, reconstructing a high dynamic range image: in the model obtained by training in the step 4, low dynamic range images with different sizes are input, namely, the low dynamic range images can be reconstructed and restored into corresponding high dynamic range images.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2019109980900 | 2019-10-21 | ||
CN201910998090 | 2019-10-21 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111709900A true CN111709900A (en) | 2020-09-25 |
Family
ID=72547985
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010748229.9A Pending CN111709900A (en) | 2019-10-21 | 2020-07-30 | High dynamic range image reconstruction method based on global feature guidance |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111709900A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112801866A (en) * | 2020-12-18 | 2021-05-14 | 展讯通信(天津)有限公司 | Image reconstruction model generation method, image reconstruction method and related equipment |
CN112991209A (en) * | 2021-03-12 | 2021-06-18 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
CN113487524A (en) * | 2021-04-07 | 2021-10-08 | 北京百度网讯科技有限公司 | Image format conversion method, device, equipment, storage medium and program product |
CN114820373A (en) * | 2022-04-28 | 2022-07-29 | 电子科技大学 | Single image reconstruction HDR method based on knowledge heuristic |
CN115103118A (en) * | 2022-06-20 | 2022-09-23 | 北京航空航天大学 | High dynamic range image generation method, device, equipment and readable storage medium |
WO2022266955A1 (en) * | 2021-06-24 | 2022-12-29 | Oppo广东移动通信有限公司 | Image decoding method and apparatus, image processing method and apparatus, and device |
WO2023039863A1 (en) * | 2021-09-17 | 2023-03-23 | 京东方科技集团股份有限公司 | Method for training image processing model, and method for generating high-dynamic-range image |
CN117474816A (en) * | 2023-12-26 | 2024-01-30 | 中国科学院宁波材料技术与工程研究所 | High dynamic range image tone mapping method, system and readable storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120201456A1 (en) * | 2009-10-08 | 2012-08-09 | International Business Machines Corporation | Transforming a digital image from a low dynamic range (ldr) image to a high dynamic range (hdr) image |
CN108681991A (en) * | 2018-04-04 | 2018-10-19 | 上海交通大学 | Based on the high dynamic range negative tone mapping method and system for generating confrontation network |
CN109255755A (en) * | 2018-10-24 | 2019-01-22 | 上海大学 | Image super-resolution rebuilding method based on multiple row convolutional neural networks |
CN110197463A (en) * | 2019-04-25 | 2019-09-03 | 深圳大学 | High dynamic range image tone mapping method and its system based on deep learning |
-
2020
- 2020-07-30 CN CN202010748229.9A patent/CN111709900A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120201456A1 (en) * | 2009-10-08 | 2012-08-09 | International Business Machines Corporation | Transforming a digital image from a low dynamic range (ldr) image to a high dynamic range (hdr) image |
CN108681991A (en) * | 2018-04-04 | 2018-10-19 | 上海交通大学 | Based on the high dynamic range negative tone mapping method and system for generating confrontation network |
CN109255755A (en) * | 2018-10-24 | 2019-01-22 | 上海大学 | Image super-resolution rebuilding method based on multiple row convolutional neural networks |
CN110197463A (en) * | 2019-04-25 | 2019-09-03 | 深圳大学 | High dynamic range image tone mapping method and its system based on deep learning |
Non-Patent Citations (2)
Title |
---|
D. MARNERIDES等: "《ExpandNet: A Deep Convolutional Neural Network for High Dynamic Range Expansion from Low Dynamic Range Content》", 《COMPUTER GRAPHICS FORUM》 * |
JINGHUI LI等: "《HDRNET: Single-Image-based HDR Reconstruction Using Channel Attention CNN》", 《ICMSSP 2019: PROCEEDINGS OF THE 2019 4TH INTERNATIONAL CONFERENCE ON MULTIMEDIA SYSTEMS AND SIGNAL》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112801866A (en) * | 2020-12-18 | 2021-05-14 | 展讯通信(天津)有限公司 | Image reconstruction model generation method, image reconstruction method and related equipment |
CN112991209A (en) * | 2021-03-12 | 2021-06-18 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
CN112991209B (en) * | 2021-03-12 | 2024-01-12 | 北京百度网讯科技有限公司 | Image processing method, device, electronic equipment and storage medium |
CN113487524A (en) * | 2021-04-07 | 2021-10-08 | 北京百度网讯科技有限公司 | Image format conversion method, device, equipment, storage medium and program product |
WO2022213716A1 (en) * | 2021-04-07 | 2022-10-13 | 北京百度网讯科技有限公司 | Image format conversion method and apparatus, device, storage medium, and program product |
WO2022266955A1 (en) * | 2021-06-24 | 2022-12-29 | Oppo广东移动通信有限公司 | Image decoding method and apparatus, image processing method and apparatus, and device |
WO2023039863A1 (en) * | 2021-09-17 | 2023-03-23 | 京东方科技集团股份有限公司 | Method for training image processing model, and method for generating high-dynamic-range image |
CN114820373A (en) * | 2022-04-28 | 2022-07-29 | 电子科技大学 | Single image reconstruction HDR method based on knowledge heuristic |
CN114820373B (en) * | 2022-04-28 | 2023-04-25 | 电子科技大学 | Single image reconstruction HDR method based on knowledge heuristic |
CN115103118B (en) * | 2022-06-20 | 2023-04-07 | 北京航空航天大学 | High dynamic range image generation method, device, equipment and readable storage medium |
CN115103118A (en) * | 2022-06-20 | 2022-09-23 | 北京航空航天大学 | High dynamic range image generation method, device, equipment and readable storage medium |
CN117474816A (en) * | 2023-12-26 | 2024-01-30 | 中国科学院宁波材料技术与工程研究所 | High dynamic range image tone mapping method, system and readable storage medium |
CN117474816B (en) * | 2023-12-26 | 2024-03-12 | 中国科学院宁波材料技术与工程研究所 | High dynamic range image tone mapping method, system and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111709900A (en) | High dynamic range image reconstruction method based on global feature guidance | |
CN110033410B (en) | Image reconstruction model training method, image super-resolution reconstruction method and device | |
CN111161277B (en) | Natural image matting method based on deep learning | |
CN110889813A (en) | Low-light image enhancement method based on infrared information | |
CN110717868B (en) | Video high dynamic range inverse tone mapping model construction and mapping method and device | |
CN110675328A (en) | Low-illumination image enhancement method and device based on condition generation countermeasure network | |
CN113096029A (en) | High dynamic range image generation method based on multi-branch codec neural network | |
CN114897742B (en) | Image restoration method with texture and structural features fused twice | |
CN111127331A (en) | Image denoising method based on pixel-level global noise estimation coding and decoding network | |
CN114283080A (en) | Multi-mode feature fusion text-guided image compression noise removal method | |
CN106709504A (en) | Detail-preserving high fidelity tone mapping method | |
CN110610458A (en) | Method and system for GAN image enhancement interactive processing based on ridge regression | |
CN110689498B (en) | High-definition video optimization method based on hierarchical blurring of non-focus part | |
CN114943894A (en) | ConvCRF-based high-resolution remote sensing image building extraction optimization method | |
CN113379606B (en) | Face super-resolution method based on pre-training generation model | |
CN116363036B (en) | Infrared and visible light image fusion method based on visual enhancement | |
CN116523985A (en) | Structure and texture feature guided double-encoder image restoration method | |
CN116703750A (en) | Image defogging method and system based on edge attention and multi-order differential loss | |
CN116109510A (en) | Face image restoration method based on structure and texture dual generation | |
CN116433516A (en) | Low-illumination image denoising and enhancing method based on attention mechanism | |
CN113888417A (en) | Human face image restoration method based on semantic analysis generation guidance | |
Mohite et al. | Image inpainting with contextual attention and partial convolution | |
CN113688694B (en) | Method and device for improving video definition based on unpaired learning | |
Wu et al. | Semantic image inpainting based on generative adversarial networks | |
CN115457265B (en) | Image defogging method and system based on generation of countermeasure network and multi-scale fusion |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200925 |