CN111340719A - Transient image data enhancement method based on full-connection automatic coding machine - Google Patents
Transient image data enhancement method based on full-connection automatic coding machine Download PDFInfo
- Publication number
- CN111340719A CN111340719A CN202010090793.6A CN202010090793A CN111340719A CN 111340719 A CN111340719 A CN 111340719A CN 202010090793 A CN202010090793 A CN 202010090793A CN 111340719 A CN111340719 A CN 111340719A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- network
- automatic coding
- coding machine
- 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
- 230000001052 transient effect Effects 0.000 title claims abstract description 64
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims description 56
- 230000006870 function Effects 0.000 claims description 40
- 238000006243 chemical reaction Methods 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 12
- 230000009466 transformation Effects 0.000 claims description 12
- 230000002708 enhancing effect Effects 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000013506 data mapping Methods 0.000 claims 1
- 239000000284 extract Substances 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 8
- 238000003384 imaging method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000013501 data transformation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 230000003313 weakening 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a transient image data enhancement method based on a full-connection automatic coding machine. And then, based on the self-designed fully-connected automatic coding machine network structure, the data characteristics of the transient image are learned by utilizing the automatic coding function of the automatic coding machine. And by setting a network structure, a related loss function, a regularization item and the like, the automatic coding machine can not only correctly learn the potential characteristics of the transient image, but also prevent overfitting. The method combines the data distribution of the transient image, utilizes the excellent characteristic of the full-connection automatic coding machine in the aspect of data characteristics, better extracts the data characteristics of the transient image, reduces the noise of original data and enhances the accuracy and stability of the data expressing the transient image.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a transient image data enhancement method based on a full-connection automatic coding machine.
Background
Transient imaging technology is a technology for imaging the transient state of light in a propagation process. With the continuous and deep research on computer graphics and computer vision, and the wide application in various fields, the problems related to illumination are also continuously raised, and it is increasingly difficult to use the traditional image information with only two bits of spatial information to end up. Transient imaging techniques can provide picosecond temporal resolution, provide rich information with high accuracy, provide possibilities for the outcome of these problems, and provide possibilities for better solutions to the problems that have already been solved. Transient imaging techniques have become a major research hotspot.
The key to transient imaging is accurate restoration of the light propagation process in the time dimension, but due to noise generated by various reasons such as Multipath Interference (MPI) problem and signal attenuation problem after multiple photon reflections, the data has a low signal-to-noise ratio. The essence of our invention is: the essence and potential characteristics of transient image data are found through a network, the enhancement of non-noise weak signals and the weakening or removal of noise signals are realized, so that the data enhancement of transient images is realized, and the signal-to-noise ratio of the data is improved.
The extraction of potential data features has been an important study in the field of images, and the use of image key features rather than the whole image generally better solves the practical problem. Such as image enhancement, image denoising, image inpainting, image compression and reconstruction, etc.
An automatic coding machine is a common network structure for feature extraction or self-coding learning. On one hand, the characteristics of the transient image data can be learned by utilizing the automatic coding function of the automatic coding machine, and on the other hand, most of noise of the transient image data can be removed by training the denoising capability of the automatic coding machine, so that the data enhancement of the transient image is realized.
In the prior art, the key of transient imaging is to accurately recover the light propagation process in the time dimension, but due to noise generated by various reasons such as Multipath Interference (MPI) problem and signal attenuation problem after multiple reflections of photons, the data has a low signal-to-noise ratio.
Disclosure of Invention
The invention mainly aims to overcome the defects of the prior art and provide a transient image data enhancement method based on a full-connection automatic coding machine, which is realized by combining the automatic coding function and the denoising learning capability of the automatic coding machine based on the transient image data distribution characteristics and aims to extract the potential characteristics of transient image data, remove most of noise, improve the data signal-to-noise ratio and realize the data enhancement of a transient image. .
In order to achieve the purpose, the invention adopts the following technical scheme:
the transient image data enhancement method based on the full-connection automatic coding machine comprises the following steps:
s1, inputting transient image data, performing numerical value conversion by utilizing logarithmic transformation processing, and inputting the processed data serving as training data into a network model of a fully-connected automatic coding machine for network iterative training;
s2, before network iterative training, setting a network structure of a fully-connected automatic coding machine;
s3, obtaining a trained fully-connected automatic coding machine network model through iterative training;
s4, carrying out numerical value conversion on the original transient image data to be enhanced, wherein the conversion mode is consistent with the operation on training data during training;
s5, inputting the converted data into a trained fully-connected automatic coding machine network model to obtain network output data;
and S6, performing inverse transformation on the network output data, and mapping the data back to the original value domain interval to obtain the enhanced data.
As a preferred technical solution, in the step S1, the numerical value conversion by the logarithm conversion process is implemented in a specific manner:
utilizing functions prior to iterative training of a network
Preprocessing the data, and mapping the processed data to [0, 8]]Within this interval.
As a preferred technical solution, the network structure setting of the fully connected automatic coding machine in step S2 is specifically implemented as:
(1) the fully-connected automatic coding machine network is a fully-connected network;
(2) 13 layers are counted, and the number of neurons in each layer is [4096, 2048, 1024, 512, 256, 128, 64, 128, 256, 512, 1024, 2048, 4096 ];
(3) the initial value of the weight of each layer meets the normal distribution with the standard deviation of 1 and the mean value of 0; setting the initial value of each layer of bias to be 0;
(4) the activation function of each layer is set to leakage Relu
(5) Setting an original loss function as a mean square error function;
(6) during training, use
Mapping raw input data to [0, 8]]Section and use
The network output data is reverse mapped back to the original interval of the transient image data.
As a preferred technical scheme, in the step (5), adding L2 regularization as a penalty term of a loss function, so that the loss function becomes an original loss function + the penalty term; the coefficient of the relevant regularization term is set to 5 e-6; by adding L2 regularization, overfitting is prevented and the model generalization performance is improved; after the loss function is calculated, the data is pruned by the Relu function and then is output as a network result.
As a preferred technical solution, in step S3, the trained network model of the fully-connected automatic coding machine is obtained through iterative training, and the implementation is specifically as follows:
training according to batches, namely dividing the whole training data into a plurality of batches of data, and training the network by using only one batch of data in each iteration;
each iteration target is a value for reducing a mean square error loss function, and when the mean square error is more than or equal to the minimum value MIN of the current record, the count k is added by 1; otherwise, updating the MIN value to be the mean square error of the current iteration, and setting the count k to be zero;
and when the count k is larger than the set threshold, stopping iteration, and finishing training to obtain the trained weight model.
As a preferable technical scheme, in order to ensure the training speed and the training stability, different learning rates are adopted at different iteration times, when the training period is in the interval of [0, 80], the learning rate is set to be 4e-5, when the training period is in the interval of [81, 230], the learning rate is set to be 1e-5, and when the iteration times exceed 230, the learning rate is set to be 5 e-6.
As a preferred technical solution, in step S4, the original transient image data to be enhanced is subjected to numerical value conversion, and the conversion mode is consistent with the operation on the training data during training, and the specific implementation is as follows:
the original data needing data enhancement is also transient images, the distribution characteristics of the transient image data are also met, and functions used in training preprocessing are utilized
Mapping raw data to [0, 8]]And the interval ensures the correct use of the weight model.
As a preferred technical solution, in step S5, the converted data is input into a trained fully-connected automatic coding machine network model to obtain network output data, and the method specifically includes:
reading the preprocessed data, calling out a trained weight model, and obtaining an output result positioned in an interval [0, 8] by the data through a 13-layer network.
As a preferred technical solution, in step S6, inverse transformation is performed on the network output data, and the data is mapped back to the original region to obtain enhanced data, which is specifically implemented as:
before inputting, the data is logarithmically transformed to map to [0, 8]]Interval, after obtaining the original output of the network, needs to pass through the function
Inverse transformation is performed on the data to map the data back to the original interval.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the invention adopts the technical scheme based on the full-connection automatic coding machine, and extracts the potential data characteristics of the transient image by unsupervised learning aiming at the data distribution characteristics of the transient image. The denoising learning capability of the automatic coding machine is combined, a large amount of noise formed by the transient image due to multi-path interference and the like is removed, the technical problem of high denoising cost of the transient image is solved, and the technical effect of data enhancement of the transient image is achieved.
2. The invention uses the logarithm function to preprocess the data aiming at the distribution characteristics of the transient image data, so that the data distribution is more uniform. The problem that the learning result of the original data in the region is only fitted with the maximum peak value is solved, and the fitting effect of the network model on the whole data is improved.
Drawings
FIG. 1 is a schematic diagram of the steps performed by the method of the present invention.
Fig. 2 is a schematic diagram of a network structure of a fully-connected automatic coding machine.
Fig. 3 is a comparison graph of data effects before and after enhancement when enhancing the "bathroom" frame 1114 of the transient image according to the present invention.
Fig. 4 is a comparison graph of data effects before and after enhancement when the transient image "bathroom" is enhanced at 2859.
Fig. 5 is a comparison graph of data effects before and after enhancement when the transient image "bathroom" frame 3681 is enhanced.
Fig. 6 is a comparison diagram of data effects before and after 4096 data on a time axis of a certain pixel point is enhanced when a transient image is enhanced.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Examples
As shown in fig. 1, the method for enhancing transient image data based on a fully connected automatic coding machine of the present invention includes the following steps:
(1) inputting transient image data, performing numerical value conversion by utilizing logarithmic transformation processing, and inputting the processed data serving as training data into a network model of a fully-connected automatic coding machine for network iterative training.
In the step (1), the value of the actually captured transient image data is exponentially attenuated and the variation range is 101To 10-7. Statistically, 90% of the data is at the quantile value 2.0742e-5, while the maximum value of a transient signal is usually greater than 10, the value difference being very large. If raw data is used as the input of the network model, when the averaging error is used as the loss function, the maximum peak value accounts for a very large proportion in error calculation, and when the model is trained by the gradient descent algorithm, only the maximum peak value is fitted. For this purpose, it is necessary to use functions before iterative training of the network
The data are preprocessed, so that the data distribution becomes more uniform, and the overall fitting effect of the network model on the signals is effectively improved. And the transformed data will be mapped to this [0, 8]]Within the interval.
(2) Before network iterative training, setting a network structure of a fully-connected automatic coding machine, and setting the network structure, namely parameters, specifically as follows:
1) the network is a fully connected network;
2) the total number of neurons in each layer is 13 layers [4096, 2048, 1024, 512, 256, 128, 64, 128, 256, 512, 1024, 2048, 4096], as shown in fig. 2.
3) The initial value of the weight of each layer meets the normal distribution with the standard deviation of 1 and the mean value of 0; each layer bias initial value is set to 0.
4) The activation function of each layer is set to leakage Relu
5) The original loss function is set to be the mean square error function. Meanwhile, L2 regularization is added as a penalty term of the loss function, so that the loss function becomes an original loss function + the penalty term. The coefficient of the associated regularization term is set to 5 e-6. By adding L2 regularization, overfitting can be effectively prevented and the model generalization performance can be improved. After the loss function is calculated, the data is pruned by the Relu function and then is output as a network result.
6) During training, use
Mapping raw input data to [0, 8]]Section and use
And reversely mapping the network output data to the original interval.
(3) And obtaining the trained fully-connected automatic coding machine network model through iterative training.
Further, the training is performed in batches, each batch having a training size set to 7000. Each iteration target is a value for reducing a mean square error loss function, and when the mean square error is more than or equal to the minimum value MIN of the current record, the count k is added by 1; otherwise, updating the MIN value to be the mean square error of the current iteration, and setting the count k to be zero; and when the count k is larger than the set threshold, stopping iteration, and finishing training to obtain the trained weight model. In order to ensure the training speed and the training stability, different learning rates are adopted at different iteration times. When the iteration number is in the interval of [0, 80], the learning rate is set to 4e-5, when the iteration number is in the interval of [81, 230], the learning rate is set to 1e-5, and when the iteration number exceeds 230, the learning rate is set to 5 e-6. When the iteration stops, a well-learned model is obtained.
(4) The transient image data needing data enhancement is preprocessed, and the function used in the preprocessing of the training data is also utilized
Mapping raw data to [0, 8]]And the interval ensures the correct use of the weight model. Note that the uppermost curve data in fig. 6 is data before preprocessing, i.e., raw data.
(5) Reading the preprocessed data, calling out a trained weight model, and obtaining an output result positioned in an interval [0, 8] by the data passing through a 13-layer network
(6) Inverse transformation of the output data with an inverse function of F (x), in particular
The data is mapped back to the original interval by inverse transformation of the output data. The second graph of fig. 6 is the output data, and the third graph is the visual comparison of the input and output data.
Fig. 2 is a network structure diagram of a fully connected transcoder. Each layer of the network structure is a full connection layer, the activation function is a leakage Relu,
the transient image used in the invention is three-dimensional data, which are respectively two-dimensional space coordinates (x, y) (0)<＝x<300,0<＝y<300) And time t (0)<＝t<4096). 7000 pixel points of a single image are randomly sampled for each training to serve as batch training input. Each single training input is 4096 data in the time dimension for a single pixel point. Will utilize before data input
And data transformation preprocessing is carried out, so that the distribution of data is more uniform. After the Encoder, 4096 data will be extracted as 64 data features. The Decoder will use these 64 features to recover 4096 data of the pixel in the time dimension. The data is eventually pruned by the Relu function. And outputting the trimmed result as network output. The network output is subjected to inverse data transformation
MappingAnd returning to the original area.
Fig. 3, fig. 4 and fig. 5 are comparison diagrams of data effects before and after transient image "bathroom" enhancement, which are respectively the 1114 th frame, the 2859 th frame and the 3681 th frame. The original data on the left and the enhanced output data on the right. It can be seen that after the characteristics of the transient image are extracted by using the characteristics of the automatic coding machine, the output result is well fitted to the original data, and the information of the structure, the edge, the light intensity and the like of the original image can be correctly expressed. And it can be seen that the right image has much reduced noise relative to the left image, and the whole image appears smoother. In conclusion, the data of the transient image can be enhanced to a certain extent based on the fully connected automatic coding machine.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (9)
1. The transient image data enhancement method based on the full-connection automatic coding machine is characterized by comprising the following steps of:
s1, inputting transient image data, performing numerical value conversion by utilizing logarithmic transformation processing, and inputting the processed data serving as training data into a network model of a fully-connected automatic coding machine for network iterative training;
s2, before network iterative training, setting a network structure of a fully-connected automatic coding machine;
s3, obtaining a trained fully-connected automatic coding machine network model through iterative training;
s4, carrying out numerical value conversion on the original transient image data to be enhanced, wherein the conversion mode is consistent with the operation on training data during training;
s5, inputting the converted data into a trained fully-connected automatic coding machine network model to obtain network output data;
and S6, performing inverse transformation on the network output data, and mapping the data back to the original value domain interval to obtain the enhanced data.
2. The method for enhancing transient image data based on a fully-connected automatic coding machine according to claim 1, wherein in step S1, the numerical value conversion is performed by using logarithmic transformation, and the specific implementation manner is as follows:
utilizing functions prior to iterative training of a network
Preprocessing the data, and mapping the processed data to [0, 8]]Within this interval.
3. The method for enhancing transient image data based on full-connected automatic coding machine according to claim 2, wherein the network structure setting of full-connected automatic coding machine in step S2 is specifically implemented as:
(1) the fully-connected automatic coding machine network is a fully-connected network;
(2) 13 layers are counted, and the number of neurons in each layer is [4096, 2048, 1024, 512, 256, 128, 64, 128, 256, 512, 1024, 2048, 4096 ];
(3) the initial value of the weight of each layer meets the normal distribution with the standard deviation of 1 and the mean value of 0; setting the initial value of each layer of bias to be 0;
(4) the activation function of each layer is set to leakage Relu
(5) Setting an original loss function as a mean square error function;
(6) during training, use
Mapping raw input data to [0, 8]]Section and use
Reversing network output dataMapping back to the original interval of the transient image data.
4. The method for enhancing transient image data based on the fully-connected automatic coding machine according to claim 3, characterized in that in (5), L2 regularization is added as a penalty term of the loss function, so that the loss function becomes the original loss function + the penalty term; the coefficient of the relevant regularization term is set to 5 e-6; by adding L2 regularization, overfitting is prevented and the model generalization performance is improved; after the loss function is calculated, the data is pruned by the Relu function and then is output as a network result.
5. The method for enhancing transient image data based on a fully-connected automatic coding machine according to claim 1, wherein in step S3, the trained fully-connected automatic coding machine network model is obtained through iterative training, and is specifically implemented as:
training according to batches, namely dividing the whole training data into a plurality of batches of data, and training the network by using only one batch of data in each iteration;
each iteration target is a value for reducing a mean square error loss function, and when the mean square error is more than or equal to the minimum value MIN of the current record, the count k is added by 1; otherwise, updating the MIN value to be the mean square error of the current iteration, and setting the count k to be zero;
and when the count k is larger than the set threshold, stopping iteration, and finishing training to obtain the trained weight model.
6. The method of claim 5, wherein different learning rates are used for different iterations to ensure training speed and training stability, the learning rate is set to 4e-5 when the training period is in the interval [0, 80], the learning rate is set to 1e-5 when the training period is in the interval [81, 230], and the learning rate is set to 5e-6 when the iteration number exceeds 230.
7. The method for enhancing transient image data based on fully connected automatic coding machine according to claim 1, wherein in step S4, the original transient image data to be enhanced is subjected to value conversion in a manner consistent with the operation on training data during training, and the method is specifically implemented as follows:
the original data needing data enhancement is also transient images, the distribution characteristics of the transient image data are also met, and functions used in training preprocessing are utilized
Mapping raw data to [0, 8]]And the interval ensures the correct use of the weight model.
8. The method for enhancing transient image data based on full-automatic coding machine of claim 3, wherein in step S5, the converted data is input into the trained network model of full-automatic coding machine to obtain the network output data, and the method is implemented as follows:
reading the preprocessed data, calling out a trained weight model, and obtaining an output result positioned in an interval [0, 8] by the data through a 13-layer network.
9. The method for enhancing transient image data based on fully connected automatic coding machine according to claim 3, wherein said step S6 is implemented by performing inverse transformation on the network output data, and mapping the data back to the original region to obtain the enhanced data, specifically as follows:
before inputting, the data is logarithmically transformed to map to [0, 8]]Interval, after obtaining the original output of the network, needs to pass through the function
Inverse transformation is performed on the data to map the data back to the original interval.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010090793.6A CN111340719B (en) | 2020-02-13 | 2020-02-13 | Transient image data enhancement method based on full-connection automatic coding machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010090793.6A CN111340719B (en) | 2020-02-13 | 2020-02-13 | Transient image data enhancement method based on full-connection automatic coding machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111340719A true CN111340719A (en) | 2020-06-26 |
| CN111340719B CN111340719B (en) | 2023-03-31 |
Family
ID=71183585
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010090793.6A Active CN111340719B (en) | 2020-02-13 | 2020-02-13 | Transient image data enhancement method based on full-connection automatic coding machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111340719B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114722826A (en) * | 2022-04-07 | 2022-07-08 | 平安科技（深圳）有限公司 | Model training method and device, electronic equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104751172A (en) * | 2015-03-12 | 2015-07-01 | 西安电子科技大学 | Method for classifying polarized SAR (Synthetic Aperture Radar) images based on de-noising automatic coding |
| CN110631699A (en) * | 2019-08-27 | 2019-12-31 | 南京理工大学 | Transient spectrometer based on deep network unmixing |
-
2020
- 2020-02-13 CN CN202010090793.6A patent/CN111340719B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104751172A (en) * | 2015-03-12 | 2015-07-01 | 西安电子科技大学 | Method for classifying polarized SAR (Synthetic Aperture Radar) images based on de-noising automatic coding |
| CN110631699A (en) * | 2019-08-27 | 2019-12-31 | 南京理工大学 | Transient spectrometer based on deep network unmixing |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114722826A (en) * | 2022-04-07 | 2022-07-08 | 平安科技（深圳）有限公司 | Model training method and device, electronic equipment and storage medium |
| CN114722826B (en) * | 2022-04-07 | 2024-02-02 | 平安科技（深圳）有限公司 | Model training method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111340719B (en) | 2023-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110599409B (en) | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel | |
| Fan et al. | Multiscale low-light image enhancement network with illumination constraint | |
| CN105046677B (en) | A kind of enhancing treating method and apparatus for traffic video image | |
| CN115661144B (en) | Adaptive medical image segmentation method based on deformable U-Net | |
| CN111861906B (en) | Pavement crack image virtual augmentation model establishment and image virtual augmentation method | |
| CN112651917A (en) | Space satellite low-illumination image enhancement method based on generation countermeasure network | |
| Zhao et al. | A novel Neutrosophic image segmentation based on improved fuzzy C-means algorithm (NIS-IFCM) | |
| Zhang et al. | Segmentation for finger vein image based on PDEs denoising | |
| Li et al. | Densely connected network for impulse noise removal | |
| CN111340719B (en) | Transient image data enhancement method based on full-connection automatic coding machine | |
| Zeng et al. | Progressive feature fusion attention dense network for speckle noise removal in OCT images | |
| CN113191968B (en) | Method for establishing three-dimensional ultrasonic image blind denoising model and application thereof | |
| CN114202473A (en) | Image restoration method and device based on multi-scale features and attention mechanism | |
| Song et al. | A single image dehazing method based on end-to-end cpad-net network in deep learning environment | |
| CN117392036A (en) | Low-light image enhancement method based on illumination amplitude | |
| Zhou et al. | An improved algorithm using weighted guided coefficient and union self‐adaptive image enhancement for single image haze removal | |
| CN116091357A (en) | Low-light image enhancement method for fusion of depth convolution attention and multi-scale features | |
| Yang et al. | Mixed noise removal by residual learning of deep cnn | |
| CN111047537A (en) | System for recovering details in image denoising | |
| Wang et al. | Infrared image edge detection based on improved Canny algorithm | |
| CN112001956A (en) | CNN-based schlieren strong laser far-field focal spot measurement image denoising method | |
| Li et al. | A lightweight and effective deep learning model for Gaussian noise removal | |
| Jiang et al. | Multi-frame image super-resolution algorithm based on small amount of data | |
| CN113962895B (en) | Self-adaptive generation countermeasure network image noise reduction method | |
| CN112508833B (en) | Medical image enhancement method and device based on deep learning |
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 |