CN111161166A - Image moire eliminating method based on depth multi-resolution network - Google Patents
Image moire eliminating method based on depth multi-resolution network Download PDFInfo
- Publication number
- CN111161166A CN111161166A CN201911293089.4A CN201911293089A CN111161166A CN 111161166 A CN111161166 A CN 111161166A CN 201911293089 A CN201911293089 A CN 201911293089A CN 111161166 A CN111161166 A CN 111161166A
- Authority
- CN
- China
- Prior art keywords
- image
- moire
- resolution
- model
- sampling
- 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 42
- 238000005070 sampling Methods 0.000 claims abstract description 46
- 230000008030 elimination Effects 0.000 claims abstract description 38
- 238000003379 elimination reaction Methods 0.000 claims abstract description 38
- 238000011176 pooling Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 29
- 230000004913 activation Effects 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims description 6
- 230000000694 effects Effects 0.000 abstract description 11
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003638 chemical reducing agent Substances 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003595 spectral effect 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
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/77—Retouching; Inpainting; Scratch removal
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- 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 Processing (AREA)
Abstract
The invention aims to provide an image moire elimination method based on a depth multiresolution network, which comprises the steps of adjusting a depth multiresolution model, redesigning a down-sampling module, replacing a down-sampling pooling layer with a convolution layer with the step length of 2, carrying out learnable multiresolution feature sampling on moire in an input image through the convolution layer with the step length of 2 by the down-sampling module, wherein a sampled feature map comprises moire information and position information under different resolutions; the downsampling mode avoids the fact that the backbone network only eliminates moire in a single frequency domain, and since moire exists in a plurality of frequency domains, moire elimination in a single resolution can cause poor elimination effect.
Description
Technical Field
The invention belongs to the field of image processing, relates to an image moire eliminating method, and particularly relates to an image moire eliminating method based on a depth multi-resolution network.
Background
Due to the limitations of the cost and the volume of a mobile phone camera, the shot images are distorted to different degrees, and particularly, on some regular texture images, the stripes of a ripple pattern, namely Moire patterns, are often seen. The sampling frequency of the mobile phone camera is fixed, and when the high-frequency components of scene information are rich and do not meet the sampling law, if high-quality clear imaging is required to be obtained, Moire interference inevitably exists when the mobile phone camera shoots a scene. Moire patterns are often large in area in an image, are obvious in color cast and seriously affect the image quality and the image analysis result; the moire can be regarded as aliasing interference noise, the structure and distribution of the moire are closely related to scene information, sampling frequency of a mobile phone camera and an interpolation algorithm, the moire is different in shape under different scenes, and no obvious distribution rule exists.
Although modern scientific research and application of moire phenomenon begin in the second half of the 19 th century, the moire phenomenon in digital imaging process has been noticed and researched in recent years along with the widespread use of digital imaging devices such as mobile phone cameras. The existing moire elimination methods are mainly classified into three major categories, namely a filtering-based method, an accurate up-sampling interpolation-based method and a professional image processing software-based method. Sidorof and Kokaram propose a spectral model to suppress Moore streaks in transmission from a telecine device. Liu et al propose low rank and sparse matrix decomposition methods to eliminate Moire patterns and preserve high frequency texture. Sur and gredianc suggest the elimination of quasi-periodic noise with the advantage of frequency domain statistics. In general, the algorithms have satisfactory effects, but the algorithms have the disadvantage that the algorithms are complex and are unlikely to be applied to a signal conversion system of a mobile phone camera.
At present, an image moire elimination method is mainly based on a super-resolution restoration technology, the concept of super-resolution restoration was originally proposed by Harris and Goodman in the last 60 th century, and a super-resolution convolutional neural network (SRCNN) is the first successful attempt to perform super-resolution by using only a convolutional layer. Unlike the shallow network architecture used in SRCNN, VeryDeep SuperResolution (VDSR) is based on the deep CNN architecture. IRCNN proposes a set of CNN-based noise reducers that can be used jointly for several low-level visual tasks, such as image denoising, deblurring, and super-resolution. The efficient subpixel convolutional neural network (ESPCN) is a fast SR method that can manipulate images and video in real time. Enhanced deep super-resolution (EDSR) modifies the ResNet architecture originally proposed for image classification for SR tasks, showing a substantial improvement by deleting the bulk normalization layer and ReLU activation. The cascade surplus network (CARN) uses ResNet Blocks to learn the relation between low-resolution input and high-resolution output, and uses local and global cascade modes to learn, thus achieving the best effect at present. The deep convolutional neural network models have good effects when applied to the image super-resolution problem, but the backbone network has single resolution, so that the effect of eliminating moire fringes when directly applied to a digital image is poor.
Since the elimination of moire in digital images is difficult to implement, it has been a challenge to train an efficient moire elimination model.
Disclosure of Invention
The invention aims to provide an image moire eliminating method based on a depth multi-resolution network, and the method is used for solving the problem that the existing super-resolution method based on a depth convolution neural network is poor in image moire eliminating effect.
The invention is realized by the following technical scheme:
an image moire elimination method based on a depth multi-resolution network comprises the following steps:
s1, collecting images, down-sampling the collected images to obtain images with moire fringes, carrying out descreening processing and denoising processing on the collected images to obtain true value labels, respectively carrying out image quantity expansion on the images with moire fringes and the true value labels, and dividing the expanded images into a training set and a test set;
s2, replacing a down-sampling pooling layer in the depth multi-resolution model with a convolution layer with the step length of 2, wherein the feature extraction module consists of three branches with different resolutions, and a residual error module is added on each resolution branch to obtain an adjusted depth multi-resolution model;
s3, taking the image with the moire pattern in the training set as input, taking the truth label in the training set as a truth value, and adopting the adjusted depth multiresolution model to train to obtain an image moire pattern elimination model;
and S4, inputting the images in the test set into the trained image moire elimination model to obtain the images with automatically eliminated moire.
Preferably, in S1, the expansion of the number of images is performed using horizontal flipping, random small degree rotation, and image cropping.
Preferably, in S2, the upsampling deconvolution layer in the depth multi-resolution model is replaced by a sub-pixel convolution layer, and a feature fusion module is added to obtain an adjusted depth multi-resolution model.
Further, in S2, the adjusted depth multiresolution model: the down-sampling module consists of two convolution layers with the step length of 2, four convolution layers with the step length of 1 and six active layers; the characteristic extraction module consists of three branches with different resolutions, each resolution branch consists of five convolution layers and five activation layers, and a residual error module is added on each resolution branch; the up-sampling module consists of two convolution layers, two active layers and two sub-pixel convolution layers, and the feature fusion module consists of one conditioner layer, two convolution layers and one active layer.
Furthermore, the convolution kernel scales of the six convolution layers of the downsampling module are all 3 multiplied by 3, and the six active layers adopt ReLU active functions; the convolution kernel scales of the 5 convolution layers of each resolution branch of the feature extraction module are all 3 multiplied by 3, and the five activation layers adopt ReLU activation functions; the convolution kernel scales of two convolution layers of the up-sampling module are both 3 multiplied by 3, the two activation layers adopt ReLU activation functions, the scaling scale of one sub-pixel convolution layer is 2, and the scaling scale of one sub-pixel convolution layer is 4.
Preferably, S3 specifically includes:
s3.1, performing random ratio down-sampling on-line data enhancement on the images with the moire patterns in the training set;
and S3.2, inputting the image subjected to online data enhancement in the S3.1 into the adjusted depth multiresolution model for training to obtain an image moire elimination model.
Further, in S3.1, the ratio of down-sampling is set to be 0.05-0.4.
Further, in S3.2, an L1 loss function is adopted in the training process, the L1 loss function is enabled to be minimum through training, the training is completed, and the image moire eliminating model is obtained.
Preferably, in S1, the collected image is an image of a PPT projection.
Compared with the prior art, the invention has the following beneficial technical effects:
the image moire eliminating method comprises the steps of adjusting a depth multiresolution model, redesigning a down-sampling module, replacing a down-sampling pooling layer with a convolution layer with the step length of 2, carrying out learnable multiresolution feature sampling on moire fringes in an input image through the convolution layer with the step length of 2 by the down-sampling module, wherein a feature map after sampling comprises moire information and position information under different resolutions; the downsampling mode avoids the fact that the backbone network only eliminates moire in a single frequency domain, and since moire exists in a plurality of frequency domains, moire elimination in a single resolution can cause poor elimination effect.
Furthermore, the method replaces the up-sampling deconvolution layer in the depth multi-resolution model with the sub-pixel convolution layer, adopts the sub-pixel convolution layer to perform up-sampling, solves the problems of black spots and black spots generated in the deconvolution up-sampling process, and ensures that a clean area in the image is not influenced by elimination of moire fringes.
Furthermore, the invention provides an online data enhancement method based on random ratio down-sampling, compared with the traditional image enhancement method, the problem that the varieties of moire data sets are less is solved more effectively, online data enhancement is carried out on the basis of the characteristic that moire changes along with the change of image resolution, and the moire elimination effect of the model is effectively improved.
Drawings
FIG. 1 is a flow chart of an implementation of the image moire elimination method based on a deep multi-resolution network of the present invention;
FIG. 2 is a network structure diagram of the image moire elimination method based on the deep multi-resolution network of the present invention;
FIG. 3 is a block diagram of an upsampling module of the present invention;
FIG. 4 is a graph showing the results of partial image moir é elimination according to the present invention.
Detailed Description
The present invention will now be described in further detail with reference to specific examples, which are intended to be illustrative, but not limiting, of the invention.
The invention provides an image moire eliminating method based on a depth multi-resolution network, which comprises the following specific steps as shown in figure 1:
s1, collecting images, down-sampling the collected images to obtain images with moire fringes, carrying out descreening processing and denoising processing on the collected images to obtain true value labels, respectively carrying out image quantity expansion on the images with moire fringes and the true value labels, and dividing the images into a training set and a test set after the expansion. The specific working process is as follows:
s1.1, shooting images at different angles, different brightness and different distances;
s1.2, down-sampling the image shot in the S1.1 to obtain an image with moire fringes, and carrying out descreening processing and denoising processing on the image shot in the S1.1 to obtain a true value label; obtaining an image data set with Moire patterns and an image data set with truth labels;
and S1.3, expanding the number of images of the two image data sets manufactured in the S1.2 respectively in a horizontal overturning mode, a random small-degree rotation mode and an image cutting mode, and finally dividing the images into a training set and a test set.
S2, adjusting the existing depth multiresolution model, replacing a down-sampling pooling layer in the existing depth multiresolution model with a convolution layer with the step length of 2, replacing an up-sampling deconvolution layer with a sub-pixel convolution layer, and adding a residual error module on each resolution branch to enable the depth multiresolution model to better adapt to the image moire elimination task; as shown in fig. 2. The adjusted depth multiresolution model specifically comprises:
the downsampling module is redesigned, an image with Moire patterns is input, three feature maps with different resolutions are output after passing through the downsampling module, and the feature maps are transmitted to the feature extraction module; since the moire has the characteristic of wide frequency coverage, the down-sampling module is used for generating feature maps with 1 time, 1/4 times and 1/16 times of input image resolution; the down-sampling module consists of two convolution layers with the step length of 2, 4 convolution layers with the step length of 1 and 6 active layers; the feature extraction module is composed of three branches with different resolutions, each resolution branch is composed of 5 convolutional layers and 5 active layers, and a residual module is added on each resolution branch. Replacing the deconvolution layer of the up-sampling module with a sub-pixel convolution layer, and adding a feature fusion module to enable features extracted by a plurality of resolution branches to be better fused; as shown in fig. 3, the upsampling module consists of two convolutional layers, two active layers, and two sub-pixel convolutional layers, and the feature fusion module consists of one conditioner layer, two convolutional layers, and one active layer.
Convolution kernel scales of 6 convolution layers of the down-sampling module are all 3 multiplied by 3, and 6 active layers adopt ReLU active functions; the convolution kernel scales of 5 convolution layers of each resolution branch of the feature extraction module are all 3 multiplied by 3, and 5 active layers adopt ReLU active functions; the convolution kernel scales of two convolution layers of the up-sampling module are both 3 multiplied by 3, the two activation layers adopt ReLU activation functions, the scaling scale of one sub-pixel convolution layer is 2, and the scaling scale of one sub-pixel convolution layer is 4.
And S3, using the depth multiresolution model for eliminating moire fringes of the image, using the image with moire fringes in the training set as input, adding an online data enhancement step of random ratio down-sampling, using a true value label in the training set as a true value, and carrying out end-to-end training to obtain the image moire fringe elimination model. The specific working process is as follows:
and S3.1, taking the training set of S1.3 as input, and performing random ratio down-sampling on-line data enhancement.
S3.2, for the adjusted depth multiresolution model in the S2, taking the image enhanced by the online data in the S3.1 as the input of the adjusted depth multiresolution model for training;
s3.3, adopting an L1 loss function in the training process of the adjusted depth multiresolution model in the S2;
and S3.4, for the adjusted depth multiresolution model in the S2, taking the image enhanced by the online data in the S3.1 as an input, and training to enable the loss function in the S3.3 to be minimum, namely completing the training to obtain the image moire eliminating model.
In the training process in the step 3, random ratio down-sampling on-line data enhancement is used, when each input image is trained, random ratio down-sampling is firstly carried out on the input image, the down-sampling ratio is set to be 0.05-0.4, and then the input image is input into the depth multi-resolution model for training, so that under the condition that the Moire data of a training set is limited, the types of Moire are increased as much as possible, and the Moire eliminating effect of the depth multi-resolution model is further improved.
And S4, for the trained image moire elimination model, taking the images in the test set as input, and obtaining the image with the moire eliminated automatically. The specific working process is as follows:
s4.1, regarding the image moire elimination model in the S3.4, taking the test set in the S1.3 as input to obtain an output image with moire eliminated automatically;
and S4.2, comparing the output image after the moire fringes are automatically eliminated in the S4.1 with the truth value label in the S1.2, and finding that the image moire fringe elimination model in the S3.4 achieves excellent moire fringe elimination effect.
In the embodiment of the invention, the image projected by the PPT is taken as an object, and the processing is performed according to the method, and the obtained result is shown in fig. 4.
The embodiments of the present invention have been described above with reference to the accompanying drawings. It will be appreciated by persons skilled in the art that the present invention is not limited by the embodiments described above. On the basis of the technical solution of the present invention, those skilled in the art can make various modifications or variations without creative efforts and still be within the protection scope of the present invention.
Claims (9)
1. An image moire elimination method based on a depth multi-resolution network is characterized by comprising the following steps:
s1, collecting images, down-sampling the collected images to obtain images with moire fringes, carrying out descreening processing and denoising processing on the collected images to obtain true value labels, respectively carrying out image quantity expansion on the images with moire fringes and the true value labels, and dividing the expanded images into a training set and a test set;
s2, replacing a down-sampling pooling layer in the depth multi-resolution model with a convolution layer with the step length of 2, wherein the feature extraction module consists of three branches with different resolutions, and a residual error module is added on each resolution branch to obtain an adjusted depth multi-resolution model;
s3, taking the image with the moire pattern in the training set as input, taking the truth label in the training set as a truth value, and adopting the adjusted depth multiresolution model to train to obtain an image moire pattern elimination model;
and S4, inputting the images in the test set into the trained image moire elimination model to obtain the images with automatically eliminated moire.
2. The image moire elimination method based on depth multi-resolution network as claimed in claim 1, wherein in S1, the expansion of image number is performed by using horizontal flipping, random small degree rotation and image cropping.
3. The image moire elimination method based on the depth multi-resolution network as claimed in claim 1, wherein in S2, the upsampling deconvolution layer in the depth multi-resolution model is replaced by a sub-pixel convolution layer, and a feature fusion module is added to obtain the adjusted depth multi-resolution model.
4. The method for image moire elimination based on depth multi-resolution network as claimed in claim 3, wherein in S2, said adjusted depth multi-resolution model is: the down-sampling module consists of two convolution layers with the step length of 2, four convolution layers with the step length of 1 and six active layers; the characteristic extraction module consists of three branches with different resolutions, each resolution branch consists of five convolution layers and five activation layers, and a residual error module is added on each resolution branch; the up-sampling module consists of two convolution layers, two active layers and two sub-pixel convolution layers, and the feature fusion module consists of one conditioner layer, two convolution layers and one active layer.
5. The image moire elimination method based on the deep multi-resolution network as claimed in claim 4, wherein the convolution kernel scales of six convolution layers of the down-sampling module are all 3 x 3, and the six activation layers adopt ReLU activation functions; the convolution kernel scales of the 5 convolution layers of each resolution branch of the feature extraction module are all 3 multiplied by 3, and the five activation layers adopt ReLU activation functions; the convolution kernel scales of two convolution layers of the up-sampling module are both 3 multiplied by 3, the two activation layers adopt ReLU activation functions, the scaling scale of one sub-pixel convolution layer is 2, and the scaling scale of one sub-pixel convolution layer is 4.
6. The image moire elimination method based on the deep multi-resolution network as claimed in claim 1, wherein S3 specifically comprises:
s3.1, performing random ratio down-sampling on-line data enhancement on the images with the moire patterns in the training set;
and S3.2, inputting the image subjected to online data enhancement in the S3.1 into the adjusted depth multiresolution model for training to obtain an image moire elimination model.
7. The image moire elimination method based on the deep multi-resolution network as claimed in claim 6, wherein in S3.1, the ratio of down sampling is set to 0.05-0.4.
8. The image moire elimination method based on the deep multi-resolution network as claimed in claim 6, wherein in S3.2, an L1 loss function is adopted in the training process, the L1 loss function is minimized through training, and the training is completed to obtain the image moire elimination model.
9. The depth multi-resolution network-based image moire elimination method as recited in claim 1, wherein in S1, the collected image is an image of PPT projection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911293089.4A CN111161166A (en) | 2019-12-16 | 2019-12-16 | Image moire eliminating method based on depth multi-resolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911293089.4A CN111161166A (en) | 2019-12-16 | 2019-12-16 | Image moire eliminating method based on depth multi-resolution network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111161166A true CN111161166A (en) | 2020-05-15 |
Family
ID=70557181
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911293089.4A Pending CN111161166A (en) | 2019-12-16 | 2019-12-16 | Image moire eliminating method based on depth multi-resolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111161166A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112184591A (en) * | 2020-09-30 | 2021-01-05 | 佛山市南海区广工大数控装备协同创新研究院 | Image restoration method based on deep learning image Moire elimination |
| CN112598602A (en) * | 2021-01-06 | 2021-04-02 | 福建帝视信息科技有限公司 | Mask-based method for removing Moire of deep learning video |
| CN113283453A (en) * | 2021-06-15 | 2021-08-20 | 深圳大学 | Target detection method, target detection device, computer equipment and storage medium |
| CN114363600A (en) * | 2022-03-15 | 2022-04-15 | 视田科技（天津）有限公司 | Remote rapid 3D projection method and system based on structured light scanning |
| CN114596479A (en) * | 2022-01-29 | 2022-06-07 | 大连理工大学 | Image moire removing method and device suitable for intelligent terminal and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108734659A (en) * | 2018-05-17 | 2018-11-02 | 华中科技大学 | A kind of sub-pix convolved image super resolution ratio reconstruction method based on multiple dimensioned label |
| CN108875331A (en) * | 2017-08-01 | 2018-11-23 | 北京旷视科技有限公司 | Face unlocking method, device and system and storage medium |
| CN109460815A (en) * | 2018-10-25 | 2019-03-12 | 浙江科技学院 | A kind of monocular depth estimation method |
| US20190095795A1 (en) * | 2017-03-15 | 2019-03-28 | Samsung Electronics Co., Ltd. | System and method for designing efficient super resolution deep convolutional neural networks by cascade network training, cascade network trimming, and dilated convolutions |
| CN109977832A (en) * | 2019-03-19 | 2019-07-05 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
| CN110322499A (en) * | 2019-07-09 | 2019-10-11 | 浙江科技学院 | A kind of monocular image depth estimation method based on multilayer feature |
| CN111586298A (en) * | 2020-05-09 | 2020-08-25 | 北京华严互娱科技有限公司 | Real-time moire removing method based on multi-scale network and dynamic feature coding |
-
2019
- 2019-12-16 CN CN201911293089.4A patent/CN111161166A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190095795A1 (en) * | 2017-03-15 | 2019-03-28 | Samsung Electronics Co., Ltd. | System and method for designing efficient super resolution deep convolutional neural networks by cascade network training, cascade network trimming, and dilated convolutions |
| CN108875331A (en) * | 2017-08-01 | 2018-11-23 | 北京旷视科技有限公司 | Face unlocking method, device and system and storage medium |
| CN108734659A (en) * | 2018-05-17 | 2018-11-02 | 华中科技大学 | A kind of sub-pix convolved image super resolution ratio reconstruction method based on multiple dimensioned label |
| CN109460815A (en) * | 2018-10-25 | 2019-03-12 | 浙江科技学院 | A kind of monocular depth estimation method |
| CN109977832A (en) * | 2019-03-19 | 2019-07-05 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
| CN110322499A (en) * | 2019-07-09 | 2019-10-11 | 浙江科技学院 | A kind of monocular image depth estimation method based on multilayer feature |
| CN111586298A (en) * | 2020-05-09 | 2020-08-25 | 北京华严互娱科技有限公司 | Real-time moire removing method based on multi-scale network and dynamic feature coding |
Non-Patent Citations (5)
| Title |
|---|
| TIANYU GAO等，: "Moiré Pattern Removal with Multi-scale Feature Enhancing Network", 《2019 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA & EXPO WORKSHOPS (ICMEW)》 * |
| 朱延东，: "复杂场景监控视频***检测", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 朱秀昌 等，: "《数字图像处理与图像通信》", 30 June 2014 * |
| 杨明 等，: "一种基于局部随机子空间的分类集成算法", 《模式识别与人工智能》 * |
| 龙法宁 等，: "基于深层卷积网络的单幅图像超分辨率重建模型", 《广西科学》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112184591A (en) * | 2020-09-30 | 2021-01-05 | 佛山市南海区广工大数控装备协同创新研究院 | Image restoration method based on deep learning image Moire elimination |
| CN112598602A (en) * | 2021-01-06 | 2021-04-02 | 福建帝视信息科技有限公司 | Mask-based method for removing Moire of deep learning video |
| CN113283453A (en) * | 2021-06-15 | 2021-08-20 | 深圳大学 | Target detection method, target detection device, computer equipment and storage medium |
| CN113283453B (en) * | 2021-06-15 | 2023-08-08 | 深圳大学 | Target detection method, device, computer equipment and storage medium |
| CN114596479A (en) * | 2022-01-29 | 2022-06-07 | 大连理工大学 | Image moire removing method and device suitable for intelligent terminal and storage medium |
| CN114363600A (en) * | 2022-03-15 | 2022-04-15 | 视田科技（天津）有限公司 | Remote rapid 3D projection method and system based on structured light scanning |
| CN114363600B (en) * | 2022-03-15 | 2022-06-21 | 视田科技（天津）有限公司 | Remote rapid 3D projection method and system based on structured light scanning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111161166A (en) | Image moire eliminating method based on depth multi-resolution network | |
| CN110033410B (en) | Image reconstruction model training method, image super-resolution reconstruction method and device | |
| CN109118432B (en) | Image super-resolution reconstruction method based on rapid cyclic convolution network | |
| CN110310227A (en) | A kind of image super-resolution rebuilding method decomposed based on high and low frequency information | |
| CN109035146B (en) | Low-quality image super-resolution method based on deep learning | |
| CN112837245B (en) | Dynamic scene deblurring method based on multi-mode fusion | |
| CN111986084A (en) | Multi-camera low-illumination image quality enhancement method based on multi-task fusion | |
| JP2000188680A (en) | Method and system for producing high resolution image | |
| CN112529776B (en) | Training method of image processing model, image processing method and device | |
| CN111462013A (en) | Single-image rain removing method based on structured residual learning | |
| Niu et al. | Low cost edge sensing for high quality demosaicking | |
| CN111340744A (en) | Attention double-flow deep network-based low-quality image down-sampling method and system | |
| CN112419150A (en) | Random multiple image super-resolution reconstruction method based on bilateral up-sampling network | |
| Yang et al. | License plate image super-resolution based on convolutional neural network | |
| He et al. | SRDRL: A blind super-resolution framework with degradation reconstruction loss | |
| CN112884650A (en) | Image mixing super-resolution method based on self-adaptive texture distillation | |
| Shen et al. | Deeper super-resolution generative adversarial network with gradient penalty for sonar image enhancement | |
| CN115526779A (en) | Infrared image super-resolution reconstruction method based on dynamic attention mechanism | |
| CN114897680A (en) | Angle super-resolution method for fusing light field sub-aperture image and macro-pixel image | |
| CN106558021A (en) | Video enhancement method based on super-resolution technique | |
| Liu et al. | Gradient prior dilated convolution network for remote sensing image super-resolution | |
| CN115965527A (en) | Mobile terminal lightweight image super-resolution reconstruction method based on convolutional neural network | |
| Ling et al. | PRNet: Pyramid Restoration Network for RAW Image Super-Resolution | |
| CN108198134B (en) | Single-image super-resolution reconstruction method based on combined average extreme learning machine | |
| Wang et al. | Multi-scale detail enhancement network for image super-resolution |
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 |
Application publication date: 20200515 |