CN114792347A - Image compression method based on multi-scale space and context information fusion - Google Patents

Image compression method based on multi-scale space and context information fusion Download PDF

Info

Publication number
CN114792347A
CN114792347A CN202210224174.0A CN202210224174A CN114792347A CN 114792347 A CN114792347 A CN 114792347A CN 202210224174 A CN202210224174 A CN 202210224174A CN 114792347 A CN114792347 A CN 114792347A
Authority
CN
China
Prior art keywords
model
scale
image compression
super
hidden
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
Application number
CN202210224174.0A
Other languages
Chinese (zh)
Inventor
王瀚漓
刘自毅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongji University
Original Assignee
Tongji University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tongji University filed Critical Tongji University
Priority to CN202210224174.0A priority Critical patent/CN114792347A/en
Publication of CN114792347A publication Critical patent/CN114792347A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/002—Image coding using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06N3/048—Activation functions

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

The invention relates to an image compression method based on multi-scale space and context information fusion, which comprises the following steps: 1) constructing an image compression model based on multi-scale space and context information fusion, extracting hidden features from an original image through a main encoder, and reducing the loss of forward transmission effective information by adopting a multi-scale information fusion module; 2) the super prior module combines the super prior information and the multi-scale context information to obtain parameters and weights of three Gaussian functions, and the parameters and the weights are added to obtain a Gaussian mixture model to obtain probability distribution of hidden features; 3) based on the probability distribution of the hidden features, the arithmetic coder codes and decodes the hidden features; 4) and the main decoder reconstructs the hidden features into pictures to finish image compression. Compared with the prior art, the method can realize more excellent image reconstruction quality under the condition of lower compression code rate.

Description

Image compression method based on multi-scale space and context information fusion
Technical Field
The invention relates to the technical field of image compression, in particular to an image compression method based on multi-scale space and context information fusion.
Background
After the third information revolution, a large amount of digital information was mutually transmitted between the respective terminals. However, most information remains in text information due to the limited manner of obtaining digital information at that time, but with the birth and popularization of various electronic digital products, especially with the arrival of the era of mobile internet, everyone can become a photographer, a large amount of pictures and video information can be transmitted on the internet by electronic equipment which can be conveniently used, and the requirement for data speed and the requirement for storage space are rapidly increased due to a large amount of data. The necessity of data compression is also reflected, and how to compress picture data occupying a considerable part of internet data becomes a hot research topic.
Before the advent of deep learning based image coding methods, there were a number of conventional methods including JPEG, JPEG2000, BPG, etc., which were widely used until now, but which were implemented with many manually designed components, including blocking, linear transformation, quantization and entropy coding in general. Due to the rapid development of deep learning and the wide application in many computer vision fields, a large number of end-to-end image compression methods based on deep learning are proposed. Most of the existing methods are based on relatively mature deep learning models, such as image compression based on a convolutional neural network, image compression based on a generation countermeasure network and image compression based on a graph convolutional neural network. The image compression algorithm based on the generation countermeasure network utilizes the countermeasure training between the generator and the discriminator to improve the human eye sense of image reconstruction under low code rate, but the recovered image has poor performance on indexes of peak signal to noise ratio (PSNR) and multi-level structure similarity (MS-SSIM). Because of the advantage of convolutional neural networks for image feature extraction, most deep learning-based image compression generally uses the structure of convolutional neural networks. The existing convolution neural network-based method firstly extracts hidden features from an image through a main encoder, and then extracts edge information in the hidden features as the super-prior-inspection features by using a super-prior-inspection automatic encoding machine. Then, the probability distribution of the hidden features is estimated by combining the super-prior features with the context features obtained by the context model, so that the hidden features are subjected to arithmetic coding. Finally, the hidden features are restored to an image using the primary encoder. However, the existing methods still have many defects. First, the main encoder removes spatial redundancy in the hidden features while also losing a part of the valid spatial information, especially information of areas with complex textures, in the forward propagation, which limits the quality of image reconstruction. On the other hand, since the content scale of the compressed image is uncertain, the mask convolution kernel with a fixed size in the context model cannot effectively obtain effective correlation information from the hidden features, so that the accuracy of the entropy model of the existing method is still insufficient.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide an image compression method based on multi-scale space and context information fusion.
The purpose of the invention can be realized by the following technical scheme:
an image compression method based on multi-scale space and context information fusion, comprising the following steps:
1) constructing an image compression model based on multi-scale space and context information fusion, extracting hidden features from an original image through a main encoder, and reducing the loss of forward transmission effective information by adopting a multi-scale information fusion module;
2) the super prior module combines the super prior information and the multi-scale context information to obtain parameters and weights of three Gaussian functions, and the parameters and the weights are added to obtain a Gaussian mixture model to obtain probability distribution of hidden features;
3) based on the probability distribution of the hidden features, the arithmetic coder codes and decodes the hidden features;
4) and the main decoder reconstructs the hidden features into pictures to complete image compression.
The step 1) specifically comprises the following steps:
11) the original picture is subjected to feature extraction and down-sampling through a residual block, an attention module and a multi-scale information fusion module to obtain a hidden feature y, and in order to carry out entropy coding on y, quantization with the step length of 1 is carried out on y to obtain a quantized hidden feature
Figure BDA0003538622660000021
Then there are:
Figure BDA0003538622660000022
Figure BDA0003538622660000023
wherein, x is the original picture,
Figure BDA0003538622660000024
for the parameters of the primary encoder, Q (-) denotes the quantization process, g a () represents a primary encoder;
12) down-sampling the i-th feature y by a multi-scale information fusion module in a main encoder (i) And down-sampling feature y by i +2 times (i+2) Through the combination of the attention mechanism, in order to reduce the consumption of computing resources, the main encoder only adopts two multi-scale information modules, and then:
y (i+2) ＝y (i+2) +y (i+2) *sigmoid(Res(y (i) )).
where Res (·) denotes the residual block.
The step 2) specifically comprises the following steps:
21) the super-prior encoder calculates the super-prior characteristic z from the hidden characteristic y, and then obtains the quantized super-prior characteristic through quantization
Figure BDA0003538622660000031
For assisting in extracting spatial redundancy in the hidden features and improving the accuracy of the probability distribution estimation of the hidden features, there are:
Figure BDA0003538622660000032
Figure BDA0003538622660000033
wherein h is a (. cndot.) represents a super-a-priori encoder,
Figure BDA0003538622660000034
parameters of the super-first-check encoder;
22) hidden features from quantization using a multi-scale three-dimensional context module
Figure BDA0003538622660000035
Multi-scale context features derived from
Figure BDA0003538622660000036
Then there are:
Figure BDA0003538622660000037
wherein, downsample is represented by downsample,
Figure BDA0003538622660000038
representing a three-dimensional context model with a convolution kernel size of 5 x 5,
Figure BDA0003538622660000039
representing a three-dimensional context model with convolution kernel size 7 x 7,
Figure BDA00035386226600000310
a three-dimensional context model representing a convolution kernel size of 9 × 9 × 9;
23) characterizing multi-scale context
Figure BDA00035386226600000311
And a super-precedent feature
Figure BDA00035386226600000312
After combination, the model parameters and the weight of the Gaussian mixture model are obtained by resolving through a super-first decoder, and then:
Figure BDA00035386226600000313
wherein, ω is i ，μ i ，
Figure BDA00035386226600000314
Respectively representing the weight, mean and variance of the ith Gaussian model in the Gaussian mixture model,
Figure BDA00035386226600000315
represents the ith super-a decoder;
24) combining three Gaussian functions into a Gaussian mixture model according to the weight to serve as an entropy model, and calculating to obtain the estimation of the probability distribution of the hidden features, wherein the estimation comprises the following steps:
Figure BDA00035386226600000316
wherein,
Figure BDA00035386226600000317
based on the characteristics of the prior
Figure BDA00035386226600000318
Hidden feature of (2)
Figure BDA00035386226600000319
The conditional probability distribution of (a) is,
Figure BDA00035386226600000320
based on a parameter omega i ，μ i The probability distribution of the gaussian distribution of (a),
Figure BDA00035386226600000321
is in the range of
Figure BDA00035386226600000322
To
Figure BDA00035386226600000323
Evenly distributed noise.
The step 3) specifically comprises the following steps:
31) in order to prevent the gradient disappearance phenomenon during model training, in the stage of training an image compression model, the quantization process is replaced by adding independent uniformly distributed uniform noise;
32) during the use of the image compression model, the hidden features are quantized, the probability distribution of the features is calculated based on the entropy model obtained by the prior encoder, and the quantized features are encoded by the arithmetic coding in the entropy encoding.
The step 4) specifically comprises the following steps:
41) the hidden features are converted into pictures again through residual blocks and upsampling in the main decoder, and then:
Figure BDA0003538622660000041
wherein, g s () represents the master decoder and the master decoder,
Figure BDA0003538622660000042
is a parameter of the main decoder and is,
Figure BDA0003538622660000043
is a reconstructed picture;
42) picture to be reconstructed
Figure BDA0003538622660000044
And comparing the compression effect with the reconstruction effect of the model by objective and subjective indexes of the original picture x.
In step 42), the objective and subjective indexes include PSNR and MS-SSIM.
In the step 1), the image compression model based on the fusion of the multi-scale space and the context information comprises a main encoder, a super-prior decoder, a main decoder and a multi-scale three-dimensional context module.
When the image compression model is trained, in order to balance the relation between the code rate and the image reconstruction quality, a trained objective function is set as:
Figure BDA0003538622660000045
wherein, the lambda is a hyper-parameter for balancing code rate and image reconstruction quality,
Figure BDA0003538622660000046
and
Figure BDA0003538622660000047
respectively as quantized hidden features
Figure BDA0003538622660000048
And quantized super-a-priori features
Figure BDA0003538622660000049
The code rate of (1), D (-) represents the difference between the original picture and the reconstructed picture, MSE and MS-SSIM are used as the measuring standards, when an MSE optimization model is adopted, the evaluation standard of the model is PSNR, and when an MS-SSIM optimization model is adopted, the evaluation standard of the model is MS-SSM.
In order to accelerate the convergence speed of the model, pre-training is firstly carried out under high code rate, then the value of lambda is modified, and the code rate of the model is adjusted to other values.
When the pre-training model is trained, the learning rate is reduced along with the iteration times, and when the models with other code rates are trained, the initial value of the learning rate is increased and is reduced along with the iteration times.
Compared with the prior art, the invention has the following advantages:
firstly, a multi-scale information fusion module adopted in a main encoder fuses image features of different scales together by using an attention mechanism, and the method avoids adding extra spatial redundancy in hidden features while preserving spatial information of a complex region.
And secondly, a multi-scale three-dimensional context module adopted in the context model fuses correlation information in different scale spaces in the hidden features together by using mask three-dimensional convolution kernels of different sizes in parallel, so that the accuracy of the entropy model is improved, and the compression efficiency of the model is improved.
Drawings
Fig. 1 is a schematic diagram of an image compression method based on multi-scale spatial and context information fusion.
Fig. 2 is a schematic diagram of a multi-scale information fusion module.
FIG. 3 is a diagram of a multi-scale three-dimensional context module.
FIG. 4 is a graph comparing the effects of the present invention and several other methods.
FIG. 5 is a graph showing the comparison of the effects of the present invention and other methods.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
Examples
The invention provides an image compression method based on multi-scale space and context information fusion, as shown in figure 1, comprising the following steps:
11) in the training stage, a training set of COCO2014 is adopted, all training pictures are randomly cut into 256x256, the batch size during training is set to be 16, and Kodak24 is adopted as a test set of model training;
12) in order to balance the relationship between the code rate and the image reconstruction quality, the objective function of model training is set as:
Figure BDA0003538622660000051
wherein, the lambda is a hyper-parameter balancing code rate and image reconstruction quality,
Figure BDA0003538622660000052
and
Figure BDA0003538622660000053
respectively represent
Figure BDA0003538622660000054
And
Figure BDA0003538622660000055
d (-) denotes the difference between the original picture and the reconstructed picture, and MSE and MS-SSIM can be used as metrics. When an MSE optimization model is used, the evaluation standard of the model is PSNR, and the values of lambda are {0.0035, 0.0067, 0.0130, 0.0250 and 0.0483} according to different code rates. If an MS-SSIM optimization model is used, the evaluation standard of the model is MS-SSM, and the values of lambda are {4.58, 8.73, 16.64, 31.73 and 60.50} according to the difference of code rates. In order to accelerate the convergence speed of the model, the model is firstly pre-trained under a high code rate, then the value of lambda is modified, and the code rate of the model is adjusted to other values. In training the pre-training model, the learning rate is initially set to 10 -5 Then, every 100000 iterations, the learning rate is reduced to 1/2. After that, when training models of other code rates, the learning rate is initially set to 5 × 10 -5 Then, every 100000 iterations, the learning rate drops to 1/2.
13) The original picture is subjected to feature extraction and down-sampling through a residual block, an attention module and a multi-scale information fusion module to obtain a hidden feature y, and in order to carry out entropy coding on y, the hidden feature y is obtained by quantizing y with the step length of 1
Figure BDA0003538622660000068
The above operation can be represented by the following notations:
Figure BDA0003538622660000061
Figure BDA0003538622660000062
14) the multi-scale information fusion module in the main encoder down-samples the i-times characteristic y (i) And down-sampling the feature y i +2 times (i+2) Fused together by means of an attention mechanism, as shown in particular in fig. 2. To reduce the consumption of computational resources, the primary encoder only employs two multi-scale information modules. This operation can be expressed by the following formula:
y (i+2) ＝y (i+2) +y (i+2) *sigmoid(Res(y (i) ))
where Res (·) denotes the residual block.
21) The super-a-priori encoder calculates a super-a-priori feature z from the hidden feature and then obtains the super-a-priori feature z by quantization
Figure BDA0003538622660000069
To help extract spatial redundancy in the hidden features and improve the accuracy of the hidden feature probability distribution estimation. The above steps can be represented by the following formula:
Figure BDA0003538622660000063
Figure BDA0003538622660000064
22) the multi-scale context features calculated from the quantized hidden features by using the multi-scale three-dimensional context model are specifically shown in fig. 3. The above steps can be formulated as:
Figure BDA0003538622660000065
23) after combining the context feature and the super-prior feature, calculating model parameters and weights of the gaussian mixture model by using a super-prior decoder, which can be expressed as:
Figure BDA0003538622660000066
24) combining three Gaussian functions into a Gaussian mixture model according to the weight to serve as an entropy model, and calculating to obtain the estimation of the probability distribution of the hidden features, wherein the estimation can be expressed as follows:
Figure BDA0003538622660000067
31) since the hidden features after quantization are discrete, the derivative of the discrete function is 0 everywhere, which will cause the gradient disappearance in the model training, and in order to train the image compression model, the quantization process will be replaced by adding independent uniformly distributed uniform noise in the training phase.
32) During the use of the model, the hidden features are quantized, the probability distribution of the features can be calculated based on an entropy model obtained by a super-prior automatic coding machine, and the quantized features are coded by entropy coding, generally by arithmetic coding in entropy coding.
41) The hidden features are converted into pictures again through residual blocks and upsampling in the main decoder, and the above steps can be expressed as:
Figure BDA0003538622660000071
42) and comparing the reconstructed picture with the original picture on objective and subjective indexes, thereby evaluating the compression effect and the reconstruction effect of the model.
In order to verify the effectiveness of the method, the method is compared with the traditional methods such as JPEG, JPEG2000, BPG, VCC and the like and partial end-to-end image compression methods. Kodak24 discloses a test set as test data, compares an original picture with a picture reconstructed after algorithm compression, calculates the difference between the two indexes of PSNR and MS-SSIM, and obtains two graphs as shown in fig. 4 and 5.
According to the invention, a multi-scale information fusion module is used in a main encoder, so that the addition of spatial redundancy in hidden features can be avoided while effective spatial information is reserved, and a multi-scale three-dimensional context module adopted in a context model can fuse context information under different scale spaces to enable an entropy model to be more accurate, and the test result of a Kodak24 public test set shows that the effect of the method is 0.15dB higher than that of the latest traditional image compression standard VVC in PSNR indexes.

Claims (10)

1. An image compression method based on multi-scale space and context information fusion is characterized by comprising the following steps:
1) constructing an image compression model based on multi-scale space and context information fusion, extracting hidden features from an original image through a main encoder, and reducing the loss of forward transmission effective information by adopting a multi-scale information fusion module;
2) the super prior module combines the super prior information and the multi-scale context information to obtain parameters and weights of three Gaussian functions, and the parameters and the weights are added to obtain a Gaussian mixture model to obtain probability distribution of hidden features;
3) based on the probability distribution of the hidden features, the arithmetic coder codes and decodes the hidden features;
4) and the main decoder reconstructs the hidden features into pictures to finish image compression.
2. The image compression method based on multi-scale spatial and contextual information fusion according to claim 1, wherein said step 1) specifically comprises the steps of:
11) the original picture is subjected to feature extraction and down-sampling through a residual block, an attention module and a multi-scale information fusion module to obtain a hidden feature y, and in order to carry out entropy coding on y, quantization with the step length of 1 is carried out on y to obtain a quantized hidden feature
Figure FDA0003538622650000011
Then there are:
Figure FDA0003538622650000015
Figure FDA0003538622650000012
wherein, x is the original picture,
Figure FDA0003538622650000013
for the parameters of the primary encoder, Q (-) denotes the quantization process, g a () represents a primary encoder;
12) down-sampling the i-th feature y by a multi-scale information fusion module in a main encoder (i) And down-sampling feature y by i +2 times (i+2) Through the fusion of the forms of attention mechanisms, in order to reduce the consumption of computing resources, the main encoder only adopts two multi-scale information modules, and the following modules are provided:
y (i+2) ＝y (i+2) +y (i+2) *sigmoid(Res(y (i) )).
where Res (·) denotes the residual block.
3. The image compression method based on the fusion of the multi-scale space and the context information as claimed in claim 2, wherein the step 2) specifically comprises the following steps:
21) the super-prior encoder calculates the super-prior characteristic z from the hidden characteristic y, and then obtains the quantized super-prior characteristic through quantization
Figure FDA0003538622650000014
For assisting in extracting spatial redundancy in the hidden features and improving the accuracy of the probability distribution estimation of the hidden features, there are:
Figure FDA0003538622650000021
Figure FDA0003538622650000022
wherein h is a (. cndot.) denotes a super-a-priori coder,
Figure FDA0003538622650000023
parameters of the super-prior encoder;
22) hidden features from quantization using a multi-scale three-dimensional context module
Figure FDA0003538622650000024
Multi-scale context features derived from
Figure FDA0003538622650000025
Then there are:
Figure FDA0003538622650000026
wherein, downsample is represented by downsample,
Figure FDA0003538622650000027
representing a three-dimensional context model with a convolution kernel size of 5 x 5,
Figure FDA0003538622650000028
representing a three-dimensional context model with convolution kernel size 7 x 7,
Figure FDA0003538622650000029
a three-dimensional context model representing a convolution kernel size of 9 × 9 × 9;
23) characterizing multi-scale context
Figure FDA00035386226500000210
And a super-precedent feature
Figure FDA00035386226500000222
After combination, the model parameters and the weight of the Gaussian mixture model are obtained by resolving through a super-first decoder, and then:
Figure FDA00035386226500000211
wherein, ω is i ，μ i ，
Figure FDA00035386226500000212
Respectively representing the weight, mean and variance of the ith Gaussian model in the Gaussian mixture model,
Figure FDA00035386226500000213
represents the ith super-a decoder;
24) combining three Gaussian functions into a Gaussian mixture model according to the weight to serve as an entropy model, and calculating to obtain the estimation of the probability distribution of the hidden features, wherein the estimation comprises the following steps:
Figure FDA00035386226500000214
wherein,
Figure FDA00035386226500000215
based on the characteristics of the prior
Figure FDA00035386226500000221
Hidden feature of (2)
Figure FDA00035386226500000216
The conditional probability distribution of (2) is,
Figure FDA00035386226500000217
based on a parameter omega i ，μ i The probability distribution of the gaussian distribution of (a),
Figure FDA00035386226500000218
is in the range of-
Figure FDA00035386226500000219
To
Figure FDA00035386226500000220
Evenly distributed noise.
4. The image compression method based on the fusion of the multi-scale space and the context information as claimed in claim 2, wherein the step 3) specifically comprises the following steps:
31) in order to prevent the gradient disappearance phenomenon during model training, in the stage of training an image compression model, the quantization process is replaced by adding independent uniformly distributed uniform noise;
32) during the use of the image compression model, the hidden features are quantized, the probability distribution of the features is calculated based on the entropy model obtained by the prior encoder, and the quantized features are encoded by the arithmetic coding in the entropy encoding.
5. The image compression method based on multi-scale spatial and contextual information fusion according to claim 1, wherein said step 4) specifically comprises the steps of:
41) the hidden features are re-transformed into pictures through residual blocks and upsampling in the main decoder, then there are:
Figure FDA0003538622650000031
wherein, g s (. cndot.) denotes a master decoder,
Figure FDA0003538622650000032
is a parameter of the main decoder and is,
Figure FDA0003538622650000033
is a reconstructed picture;
42) picture to be reconstructed
Figure FDA0003538622650000034
And comparing the compression effect with the reconstruction effect of the model by objective and subjective indexes of the original picture x.
6. The method as claimed in claim 5, wherein in step 42), the objective and subjective indicators include PSNR and MS-SSIM.
7. The method according to claim 1, wherein the image compression model based on fusion of multi-scale space and context information in step 1) comprises a main encoder, a super-a-decoder, a main decoder, and a multi-scale three-dimensional context module.
8. The method of claim 7, wherein when the image compression model is trained, in order to balance the relationship between the bit rate and the image reconstruction quality, the trained objective function is set as:
Figure FDA0003538622650000035
wherein, the lambda is a hyper-parameter for balancing code rate and image reconstruction quality,
Figure FDA0003538622650000036
and
Figure FDA0003538622650000037
respectively the quantized hidden featuresSign for
Figure FDA0003538622650000038
And the quantized super-a priori characteristics
Figure FDA0003538622650000039
The code rate of (1), D (-) represents the difference between the original picture and the reconstructed picture, MSE and MS-SSIM are used as the measuring standards, when an MSE optimization model is adopted, the evaluation standard of the model is PSNR, and when an MS-SSIM optimization model is adopted, the evaluation standard of the model is MS-SSM.
9. The image compression method based on multi-scale space and context information fusion of claim 8, characterized in that, in order to accelerate the convergence speed of the model, pre-training is performed at a high code rate, and then the value of λ is modified, and the code rate of the model is adjusted to other values.
10. The image compression method based on multi-scale space and context information fusion of claim 8, characterized in that, when a pre-training model is trained, the learning rate decreases with the number of iterations, and when models with other code rates are trained, the initial learning rate value increases and decreases with the number of iterations.
CN202210224174.0A 2022-03-09 2022-03-09 Image compression method based on multi-scale space and context information fusion Pending CN114792347A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210224174.0A CN114792347A (en) 2022-03-09 2022-03-09 Image compression method based on multi-scale space and context information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210224174.0A CN114792347A (en) 2022-03-09 2022-03-09 Image compression method based on multi-scale space and context information fusion

Publications (1)

Publication Number Publication Date
CN114792347A true CN114792347A (en) 2022-07-26

Family

ID=82460743

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210224174.0A Pending CN114792347A (en) 2022-03-09 2022-03-09 Image compression method based on multi-scale space and context information fusion

Country Status (1)

Country Link
CN (1) CN114792347A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116743182A (en) * 2023-08-15 2023-09-12 国网江西省电力有限公司信息通信分公司 Lossless data compression method
CN117173263A (en) * 2023-10-31 2023-12-05 江苏势通生物科技有限公司 Image compression method for generating countermeasure network based on enhanced multi-scale residual error

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116743182A (en) * 2023-08-15 2023-09-12 国网江西省电力有限公司信息通信分公司 Lossless data compression method
CN116743182B (en) * 2023-08-15 2023-12-08 国网江西省电力有限公司信息通信分公司 Lossless data compression method
CN117173263A (en) * 2023-10-31 2023-12-05 江苏势通生物科技有限公司 Image compression method for generating countermeasure network based on enhanced multi-scale residual error
CN117173263B (en) * 2023-10-31 2024-02-02 江苏势通生物科技有限公司 Image compression method for generating countermeasure network based on enhanced multi-scale residual error

Similar Documents

Publication Publication Date Title
CN111787323B (en) Variable bit rate generation type compression method based on counterstudy
CN113658051B (en) Image defogging method and system based on cyclic generation countermeasure network
CN114792347A (en) Image compression method based on multi-scale space and context information fusion
US11354822B2 (en) Stop code tolerant image compression neural networks
CN111147862B (en) End-to-end image compression method based on target coding
CN110248190B (en) Multilayer residual coefficient image coding method based on compressed sensing
CN111641826B (en) Method, device and system for encoding and decoding data
You et al. Patch-based deep autoencoder for point cloud geometry compression
CN111833277A (en) Marine image defogging method with non-paired multi-scale hybrid coding and decoding structure
CN113132727B (en) Scalable machine vision coding method and training method of motion-guided image generation network
CN113163203A (en) Deep learning feature compression and decompression method, system and terminal
CN110753225A (en) Video compression method and device and terminal equipment
CN116600119B (en) Video encoding method, video decoding method, video encoding device, video decoding device, computer equipment and storage medium
CN115393452A (en) Point cloud geometric compression method based on asymmetric self-encoder structure
CN114422802A (en) Self-encoder image compression method based on codebook
CN114494387A (en) Data set network generation model and fog map generation method
CN117522674A (en) Image reconstruction system and method combining local and global information
CN111107377A (en) Depth image compression method, device, equipment and storage medium
CN114663536B (en) Image compression method and device
CN116437089A (en) Depth video compression algorithm based on key target
CN111343458A (en) Sparse gray image coding and decoding method and system based on reconstructed residual
CN115361555A (en) Image encoding method, image encoding device, and computer storage medium
Luo et al. Neural image compression with quantization rectifier
Mao et al. Learning to predict on octree for scalable point cloud geometry coding
Teng et al. Light Field Compression via a Variational Graph Auto-Encoder

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