CN109376762A - Image processing method and device - Google Patents
Image processing method and device Download PDFInfo
- Publication number
- CN109376762A CN109376762A CN201811067307.8A CN201811067307A CN109376762A CN 109376762 A CN109376762 A CN 109376762A CN 201811067307 A CN201811067307 A CN 201811067307A CN 109376762 A CN109376762 A CN 109376762A
- Authority
- CN
- China
- Prior art keywords
- image
- algorithm
- compression
- target
- compression algorithm
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The disclosure provides a kind of image processing method and device, is related to technical field of image processing, is able to solve the single image compression algorithm problem not good enough to different classes of image compression quality.The specific technical proposal is: obtaining target image;By neural network classification model, the image category of the target image is determined;Target algorithm is determined according to preset mapping data;The target image is compressed according to the target algorithm.The disclosure is used for image procossing.
Description
Technical field
This disclosure relates to technical field of image processing more particularly to image processing method and device.
Background technique
The usual color of natural image is uniformly excessive, is not in the image information of step change type.Therefore logical for natural image
Compression of images is carried out frequently with the mode of prediction plus frequency-domain transform, preferable compression quality can be obtained.
Different from natural image, it is artificially generated the characteristics of image has oneself.It is that noise is controllable first, nothing can be generated and make an uproar
Image;Secondly, including a large amount of high gradient signal, such as text, lines in image.Frequency domain is added to become according to traditional prediction
The mode changed compresses these high gradient images, and relative to natural image, often compression effectiveness is not good enough.
As it can be seen that a kind of image compression algorithm good compression quality available for the image of a certain classification, but for
Then compression quality cannot be guaranteed the image of another category.
Summary of the invention
The embodiment of the present disclosure provides a kind of image processing method and device, is able to solve single image compression algorithm to difference
The problem that the image compression quality of classification is not good enough.The technical solution is as follows:
According to the first aspect of the embodiments of the present disclosure, a kind of image processing method is provided, this method comprises:
Obtain target image;
By neural network classification model, the image category of the target image is determined;
Target algorithm is determined according to preset mapping data;
Wherein, the mapping data are used to indicate K image category and L compression algorithm, 2≤L≤K, the K image
An image category corresponds to a compression algorithm in the L compression algorithm in classification, i-th in the K image category
Compression algorithm corresponding to image category is in the L compression algorithm to i-th of optimal compression of image category compression quality
Algorithm, the target algorithm are compression algorithm corresponding to the image category of the target image, 1≤i≤K;
The target image is compressed according to the target algorithm.
Technical solution provided by the present disclosure determines the image category of target image by neural network classification model, according to
Image category selects one as target algorithm from K compression algorithm, and target algorithm is in K compression algorithm to the image class
The other optimal compression algorithm of compression quality, after determining target algorithm, compresses target image according to target algorithm.I.e.
According to image category, the optimal compression algorithm of compression quality is selected from K compression algorithm, target image is compressed, solve
The problem that single image compression algorithm of having determined is not good enough to different classes of image compression quality.
In one embodiment, further includes:
Training sample is obtained, determines canonical algorithm corresponding to each sample image in the training sample;Wherein, one
Canonical algorithm corresponding to sample image is in the L compression algorithm to one optimal pressure of sample image compression quality
Compression algorithm;
The training neural network classification model, the sample image that same canonical algorithm is corresponded in the training sample is returned
For one kind.
The sample image of the same canonical algorithm of correspondence is classified as one kind, the quantity of image category can be reduced as far as possible, is improved
The classification effectiveness to target image of neural network classification model.
In one embodiment, K=L, the K image category and the L compression algorithm correspond;
It is described to determine target algorithm according to preset mapping data, comprising:
According to the one-to-one K compression algorithm of the K image category, the determining image class with the target image
Not corresponding compression algorithm.
Image category is identical as compression algorithm quantity, corresponds, the quantity of image category reaches minimum value at this time, into one
Step improves the classification effectiveness to target image of neural network classification model.
In one embodiment, canonical algorithm corresponding to each sample image in the determination training sample, packet
It includes:
Target sample image is compressed by each compression algorithm in the L compression algorithm respectively, is determined each
The corresponding compression quality parameter of compression algorithm;
According to the corresponding compression quality parameter of each compression algorithm, determine that standard corresponding to the target sample image is calculated
Method.
For a sample image, is compressed respectively using different compression algorithms, it is corresponding to compare algorithms of different
Compression quality determines the optimal compression algorithm of compression quality.Before product export, it can pass through when training neural network classification model
This mode determines the corresponding canonical algorithm of sample image.After product export during user's use, in this way
Further training optimization can be done to neural network classification model.
In one embodiment, the corresponding compression quality parameter of each compression algorithm of the determination, comprising:
Determine the corresponding compression ratio of each compression algorithm and Y-PSNR it is therein at least one.
In one embodiment, the acquisition target image, comprising:
Image to be compressed is obtained, the image to be compressed is divided into image macro, by the image of the image to be compressed
Macro block is as the target image.
Image to be compressed is divided into macro block, then selects suitable compression algorithm to be compressed for each macro block, can be mentioned
Compression quality of the height to image entirety to be compressed.
According to the second aspect of an embodiment of the present disclosure, a kind of image processing apparatus is provided, comprising:
Input module, for obtaining target image;
Categorization module, for determining the image category of the target image by neural network classification model;
Chosen module, for determining target algorithm according to preset mapping data;
Wherein, the mapping data are used to indicate K image category and L compression algorithm, 2≤L≤K, the K image
An image category corresponds to a compression algorithm in the L compression algorithm in classification, i-th in the K image category
Compression algorithm corresponding to image category is in the L compression algorithm to i-th of optimal compression of image category compression quality
Algorithm, the target algorithm are compression algorithm corresponding to the image category of the target image, K >=2,1≤i≤K;
Compression module, for being compressed according to the target algorithm to the target image.
In one embodiment, further includes:
Base modules determine standard corresponding to each sample image in the training sample for obtaining training sample
Algorithm;Wherein, canonical algorithm corresponding to a sample image is in the L compression algorithm to one sample image pressure
The optimal compression algorithm of contracting quality;
Training module will correspond to same standard and calculate for training the neural network classification model in the training sample
The sample image of method is classified as one kind.
In one embodiment, K=L, the K image category and the L compression algorithm correspond, described selected
Module includes:
Corresponding submodule, for according to and the one-to-one K compression algorithm of the K image category, determine with it is described
The corresponding compression algorithm of the image category of target image.
In one embodiment, the base modules include:
Computational submodule, for respectively by each compression algorithm in the L compression algorithm to target sample image into
Row compression, determines the corresponding compression quality parameter of each compression algorithm;
Submodule is screened, for determining the target sample figure according to the corresponding compression quality parameter of each compression algorithm
As corresponding canonical algorithm.
In one embodiment, the computational submodule includes:
Parameters unit, for determine the corresponding compression ratio of each compression algorithm and Y-PSNR it is therein at least one.
In one embodiment, the input module includes:
Divide submodule, for obtaining image to be compressed, the image to be compressed be divided into image macro, will it is described to
The image macro of image is compressed as the target image.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure
Example, and together with specification for explaining the principles of this disclosure.
Fig. 1 is a kind of flow chart for image processing method that the embodiment of the present disclosure provides;
Fig. 2 is that the image category that the embodiment of the present disclosure provides and compression algorithm corresponding relationship illustrate schematic diagram;
Fig. 3 is a kind of flow chart for image processing method that the embodiment of the present disclosure provides;
Fig. 4 be the embodiment of the present disclosure provide schematic diagram is illustrated to neural network classification model training process;
Fig. 5 is that the image category that the embodiment of the present disclosure provides and compression algorithm corresponding relationship illustrate schematic diagram;
Fig. 6 be the embodiment of the present disclosure provide schematic diagram is illustrated to image macro treatment process;
Fig. 7 is a kind of structural schematic diagram for image processing apparatus that the embodiment of the present disclosure provides;
Fig. 8 is a kind of structural schematic diagram for image processing apparatus that the embodiment of the present disclosure provides;
Fig. 9 is a kind of structural schematic diagram for image processing apparatus that the embodiment of the present disclosure provides;
Figure 10 is a kind of structural schematic diagram for image processing apparatus that the embodiment of the present disclosure provides;
Figure 11 is a kind of structural schematic diagram for image processing apparatus that the embodiment of the present disclosure provides;
Figure 12 is a kind of structural schematic diagram for image processing apparatus that the embodiment of the present disclosure provides.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
It is different that the compression quality that same image obtains is handled using different compression algorithms.Embodiment of the disclosure provides one kind
Image processing method, for an image to be compressed, the compression that can select compression quality optimal from several compression algorithm is calculated
Method is compressed, and has solved the problems, such as that single image compression algorithm is not good enough to different classes of image compression quality.
Image processing method is applied to image processing apparatus, which can be applied in the equipment such as mobile phone, computer, TV.
As shown in Figure 1, image processing method the following steps are included:
101, target image is obtained.
Target image is the image that need to be compressed by image compression algorithm.
102, by neural network classification model, the image category of target image is determined.
Neural network classification model is trained image classification model.Target image is neural network classification model
Input, neural network classification model export the image category of target image.Input picture is divided into K by neural network classification model
A image category, K >=2.
In one embodiment, neural network classification model is by L compression algorithm, and wherein a certain compression algorithm can for use
The image for obtaining optimal compression quality incorporates one kind, 2≤L≤K into.
Wherein, compression algorithm includes but is not limited to that the compression calculation of compression of images is carried out by way of prediction plus frequency-domain transform
Method, for compression algorithm of high gradient image etc..Compression quality can be measured by compression ratio or Y-PSNR.
Referring to shown in Fig. 2, with L=3, for the situation of K=4, compression algorithm includes algorithm A1, algorithm A2And algorithm A3,
Image category includes classification C1, classification C2, classification C3With classification C4。
Classification C1To use algorithm A1The image category of optimal compression quality can be obtained.
Classification C2To use algorithm A2The image category of optimal compression quality can be obtained.
Classification C3To use algorithm A3The image category of optimal compression quality can be obtained.
Classification C4To use algorithm A3The image category of optimal compression quality can be obtained.
103, target algorithm is determined according to preset mapping data.
Referring to shown in Fig. 2, mapping data are used to indicate K image category and L compression algorithm, one in K image category
A image category corresponds to a compression algorithm in L compression algorithm, in K image category corresponding to i-th of image category
Compression algorithm is in L compression algorithm to i-th of optimal compression algorithm of image category compression quality, 1≤i≤K.Shown in Fig. 2
For L=3, the situation of K=4.
Target algorithm is compression algorithm corresponding to the image category of target image.Referring to shown in Fig. 2, when target image
Image category is classification C1When, target algorithm is algorithm A1。
104, target image is compressed according to target algorithm.
Referring to shown in Fig. 2, determine that the image category of target image is C by neural network classification model1When, then according to calculation
Method A1Target image is compressed.Relative to algorithm A2With algorithm A3, according to algorithm A1It is available that compression is carried out to target image
Better compression quality.
The image processing method that the embodiment of the present disclosure provides, the image of target image is determined by neural network classification model
Classification selects one as target algorithm according to image category from K compression algorithm, and target algorithm is in K compression algorithm
A compression algorithm optimal to the image category compression quality, after determining target algorithm, according to target algorithm to target image
It is compressed.I.e. according to image category, the optimal compression algorithm of compression quality is selected from K compression algorithm, to target image
It is compressed, solves the problems, such as that single image compression algorithm is not good enough to different classes of image compression quality.
Based on the image processing method that the corresponding embodiment of above-mentioned Fig. 1 provides, another embodiment of the disclosure provides a kind of figure
As processing method.The present embodiment is illustrated so that target image is the situation of image macro as an example, interior in part of step
The step held in embodiment corresponding with Fig. 1 is same or like, only elaborates below to difference in step.
Referring to shown in Fig. 3, image processing method provided in this embodiment the following steps are included:
301, training sample is obtained, determines canonical algorithm corresponding to each sample image in training sample.
Training sample includes several sample images, is carried out so that sample image is the situation of image macro as an example in the present embodiment
Illustrate, without limitation to the size of image macro.
Canonical algorithm corresponding to one sample image is optimal to the sample image compression quality in L compression algorithm
Compression algorithm.L compression algorithm include but is not limited to H.264, H.265, JPEG, YUV etc..
For purposes of illustration only, a sample image in training sample is known as target sample image.
Target sample image is compressed by each compression algorithm in L compression algorithm respectively, determines each compression
The corresponding compression quality parameter of algorithm.Further according to the corresponding compression quality parameter of each compression algorithm, target sample image is determined
Corresponding canonical algorithm.
Referring to shown in Fig. 4, target sample image is passed through respectively after L compression algorithm compressed, each pressure is obtained
The compression quality parameter of compression algorithm.Algorithm A1Corresponding compression quality parameter is with Q1It indicates, algorithm A2Corresponding compression quality
Parameter is with Q2It indicates, and so on.The corresponding L compression quality parameter meter of L compression algorithm is { Q1, Q2, Q3... QL}
Compression quality parameter may include compression ratio and Y-PSNR it is therein at least one.Include with compression quality parameter
For the situation of compression ratio or Y-PSNR, compression ratio is higher, and expression compression quality is higher, the higher expression pressure of Y-PSNR
Contracting quality is higher.By { Q1, Q2, Q3... QLIn algorithm corresponding to maximum value be determined as standard corresponding to target sample image
Algorithm.For example, working as Q2For { Q1, Q2, Q3... QLIn maximum value when, determine algorithm A2For mark corresponding to target sample image
Quasi- algorithm.
When compression quality parameter includes compression ratio and Y-PSNR, compression quality parameter can be compression ratio and peak value
The weighted sum of signal-to-noise ratio.Compression quality parameter is higher, and expression compression quality is higher, by { Q1, Q2, Q3... QLIn maximum value institute
Corresponding algorithm is determined as canonical algorithm corresponding to target sample image.
302, training neural network classification model, is classified as one for the sample image for corresponding to same canonical algorithm in training sample
Class.
Neural network classification model includes but is not limited to the models such as VGG16, VGG19.In training Holy Bible network class model
When, the sample image of the same canonical algorithm of correspondence is classified as one kind, the quantity of image category can be reduced, as far as possible to improve nerve
The classification effectiveness to target image of network class model.
Referring to shown in Fig. 4, the feature of neural network classification model extraction target sample image is classified according to feature,
The image category for determining target sample image determines image referring to the corresponding relationship of image category shown in Fig. 2 and compression algorithm
The corresponding target algorithm of classification.
When the target algorithm of target sample image and canonical algorithm are inconsistent, the ginseng of neural network classification model is corrected
Number does further training optimization to neural network classification model.When the target algorithm of target sample image is consistent with canonical algorithm
When, indicate that neural network classification model is correct to the classification of target sample image.
For example, the image category for the target sample image that neural network classification model determines is classification C3, it is determined that target
Algorithm is algorithm A3.And the canonical algorithm of the target sample image determined in step 301 is algorithm A2.Target algorithm and standard are calculated
Method is inconsistent, then neural network classification model does further training optimization.
When the image category for the target sample image that neural network classification model determines is classification C2When, determining target is calculated
Method is algorithm A2.The canonical algorithm of the target sample image determined in step 301 is algorithm A2.Target algorithm and canonical algorithm one
It causes, neural network classification model is correct to the classification of target sample image.
Input picture is divided into K image category, the smaller neural network classification mould of the value of K by neural network classification model
Type is higher to the classification effectiveness of input picture.
The minimum value of K is L, i.e. K=L, and the quantity of image category and the quantity of compression algorithm are equal at this time, are K (L)
A, K (L) a image category and K (L) a compression algorithm correspond.
Referring to Figure 5, by taking the situation of K=L=3 as an example, compression algorithm includes algorithm A1, algorithm A2And algorithm A3, figure
As classification includes classification C1, classification C2With classification C3.3 image categories are corresponded with 3 compression algorithms.
Step 301- step 302 is the process being trained to neural network classification model.
For target sample image, is compressed respectively using different compression algorithms, it is corresponding to compare algorithms of different
Compression quality determines the optimal compression algorithm of compression quality, i.e. canonical algorithm.
By canonical algorithm compared with the target algorithm according to determined by neural network classification model, the two unanimously then indicates mind
Correct through network class category of model, the two is inconsistent, does further training optimization to neural network classification model.
Step 301- step 302 can be executed before factory, i.e., prefabricated before product export to have already passed through trained nerve net
Network disaggregated model.When executing before factory, training sample is usually the image macro that developer selectes.
Step 301- step 302 can be executed before factory, can be executed in the use process of user after product export, i.e.,
Further training optimization is done to neural network classification model in user's use process.When being executed in user's use process, instruction
Practicing sample is usually actual use process extracted image macro from image to be compressed.
303, image to be compressed is obtained, image to be compressed is divided into image macro.
Step 303 and subsequent step are the mistake for selecting compression algorithm based on neural network classification model and completing compression of images
Journey.
According to scheduled macroblock size, image to be compressed is divided into several image macros, by subsequent step to each
Macro block completes compression of images.
304, by neural network classification model, the image category of target image is determined.
Referring to shown in Fig. 6, using the image macro of image to be compressed as target image, i.e. mesh neural network classification model
Input, the image category of image macro is determined by neural network classification model.
305, target algorithm is determined according to preset mapping data.
In one embodiment, K=L, i.e. K image category are corresponded with K compression algorithm.Determining image category
Afterwards, according to image category, determine a compression algorithm corresponding with the image category as target algorithm.
For example, referring to shown in Fig. 5, neural network classification model determines that the image category of image macro is classification C1, then scheme
As the corresponding target algorithm of macro block is algorithm A1。
306, target image is compressed according to target algorithm.
By taking an image macro as an example, referring to Fig. 5, the image category of the image macro is classification C1, then corresponding target
Algorithm is algorithm A1, according to algorithm A1The image macro is compressed.
Referring to shown in Fig. 6, each image macro respectively corresponds to a target algorithm, macro to each image according to target algorithm
Block is compressed.
The image processing method that the embodiment of the present disclosure provides, the image of target image is determined by neural network classification model
Classification selects the optimal compression algorithm of compression quality from K compression algorithm, presses target image according to image category
Contracting, solves the problems, such as that single image compression algorithm is not good enough to different classes of image compression quality.
The compression algorithm for being suitable for image macro is determined by neural network classification model, without for each compression
Algorithm goes to determine that its compression quality makes a choice again one by one.Huge, the compression algorithm that determines that compression quality will lead to calculation amount one by one
Quantity increase (value of L is bigger) and will lead to calculation amount and increase suddenly.And it can be efficiently from a variety of by neural network classification model
Optimal compression algorithm is selected in compression algorithm, greatlys save computing resource, when the quantity of compression algorithm increases, calculation amount increase
It is relatively small, therefore the quantity of compression algorithm is more, the effect for saving computing resource is better.
Compression algorithm quantity is bigger, carry out compression of images when alternative compression algorithm it is more, compression of images effect is got over
It is good.The image processing method favorable expandability that the disclosure provides only need to be to neural network classification when increasing more compression algorithms
Re -training after model is finely tuned, product renewing upgrade cost are small, high-efficient.
In addition, can further train neural network point in user's use process using technical solution provided by the present disclosure
Class model continues to optimize neural network classification model, improves the accuracy to image classification, so that the compression improved to image is imitated
Fruit.
Further, provided training sample often has differences under different application scene, the skill that the disclosure provides
In art scheme, neural network classification model can constantly training optimizes in specific application scenarios, obtains so that the disclosure provided
Scheme can be widely applied to plurality of application scenes, obtain optimal compression quality for different crowd, different demands.
Image processing method described in based on the above embodiment, following is embodiment of the present disclosure, be can be used for
Execute embodiments of the present disclosure.
The embodiment of the present disclosure provides a kind of image processing apparatus, as shown in fig. 7, the image processing apparatus includes:
Input module 71, for obtaining target image.
Categorization module 72, for determining the image category of target image by neural network classification model.
Chosen module 73, for determining target algorithm according to preset mapping data.
Wherein, mapping data are used to indicate K image category and L compression algorithm, 2≤L≤K, one in K image category
A image category corresponds to a compression algorithm in L compression algorithm, in K image category corresponding to i-th of image category
Compression algorithm is in L compression algorithm to i-th of optimal compression algorithm of image category compression quality, and target algorithm is target
Compression algorithm corresponding to the image category of image, K >=2,1≤i≤K.
Compression module 74, for being compressed according to target algorithm to target image.
As shown in figure 8, in one embodiment, further includes:
Base modules 75 determine that standard corresponding to each sample image is calculated in training sample for obtaining training sample
Method.Wherein, canonical algorithm corresponding to a sample image is best to a sample image compression quality in L compression algorithm
Compression algorithm.
Training module 76 will correspond to the sample of same canonical algorithm for training neural network classification model in training sample
This image is classified as one kind.
As shown in figure 9, in one embodiment, K=L, K image category is corresponded with L compression algorithm, mould is selected
Block 73 includes:
Corresponding submodule 731, for basis and the one-to-one K compression algorithm of K image category, determining and target figure
The corresponding compression algorithm of the image category of picture.
As shown in Figure 10, in one embodiment, base modules 75 include:
Computational submodule 751, for being carried out respectively by each compression algorithm in L compression algorithm to target sample image
Compression, determines the corresponding compression quality parameter of each compression algorithm.
Submodule 752 is screened, for determining target sample image according to the corresponding compression quality parameter of each compression algorithm
Corresponding canonical algorithm.
As shown in figure 11, in one embodiment, computational submodule 751 includes:
Parameters unit 753, for determining the corresponding compression ratio of each compression algorithm and Y-PSNR therein at least one
It is a.
As shown in figure 12, in one embodiment, input module 71 includes:
It divides submodule 711 and image to be compressed is divided into image macro for obtaining image to be compressed, it will be to be compressed
The image macro of image is as target image.
The image processing apparatus that the embodiment of the present disclosure provides, the image of target image is determined by neural network classification model
Classification selects one as target algorithm according to image category from K compression algorithm, and target algorithm is in K compression algorithm
A compression algorithm optimal to the image category compression quality, after determining target algorithm, according to target algorithm to target image
It is compressed.I.e. according to image category, the optimal compression algorithm of compression quality is selected from K compression algorithm, to target image
It is compressed, solves the problems, such as that single image compression algorithm is not good enough to different classes of image compression quality.
Based on image processing method described in the corresponding embodiment of above-mentioned Fig. 1-Fig. 6, the embodiment of the present disclosure is also provided
A kind of computer readable storage medium.
The computer readable storage medium can be non-transitory computer-readable storage medium.For example, non-transitory calculates
Machine readable storage medium storing program for executing can be read-only memory (English: Read Only Memory, ROM), random access memory (English
Text: Random Access Memory, RAM), CD-ROM, tape, floppy disk and optical data storage devices etc..On the storage medium
It is stored with computer instruction, when computer instruction is performed, it can be achieved that described in the corresponding embodiment of above-mentioned Fig. 1-Fig. 6
Data transmission method, details are not described herein again.
Those skilled in the art will readily occur to its of the disclosure after considering specification and practicing disclosure disclosed herein
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by following
Claim is pointed out.
Claims (12)
1. a kind of image processing method characterized by comprising
Obtain target image;
By neural network classification model, the image category of the target image is determined;
Target algorithm is determined according to preset mapping data;
Wherein, the mapping data are used to indicate K image category and L compression algorithm, 2≤L≤K, the K image category
In an image category correspond to a compression algorithm in the L compression algorithm, i-th of image in the K image category
Compression algorithm corresponding to classification is to calculate in the L compression algorithm the optimal compression of i-th of image category compression quality
Method, the target algorithm are compression algorithm corresponding to the image category of the target image, 1≤i≤K;
The target image is compressed according to the target algorithm.
2. the method according to claim 1, wherein further include:
Training sample is obtained, determines canonical algorithm corresponding to each sample image in the training sample;Wherein, a sample
Canonical algorithm corresponding to image is to calculate in the L compression algorithm the optimal compression of one sample image compression quality
Method;
The training neural network classification model, is classified as one for the sample image for corresponding to same canonical algorithm in the training sample
Class.
3. according to the method described in claim 2, it is characterized in that, K=L, the K image category and the L compression are calculated
Method corresponds;
It is described to determine target algorithm according to preset mapping data, comprising:
According to the one-to-one K compression algorithm of the K image category, the determining image category pair with the target image
The compression algorithm answered.
4. according to the method described in claim 2, it is characterized in that, each sample image institute in the determination training sample
Corresponding canonical algorithm, comprising:
Target sample image is compressed by each compression algorithm in the L compression algorithm respectively, determines each compression
The corresponding compression quality parameter of algorithm;
According to the corresponding compression quality parameter of each compression algorithm, canonical algorithm corresponding to the target sample image is determined.
5. according to the method described in claim 4, it is characterized in that, the corresponding compression quality ginseng of each compression algorithm of the determination
Number, comprising:
Determine the corresponding compression ratio of each compression algorithm and Y-PSNR it is therein at least one.
6. method according to claim 1-5, which is characterized in that the acquisition target image, comprising:
Image to be compressed is obtained, the image to be compressed is divided into image macro, by the image macro of the image to be compressed
As the target image.
7. a kind of image processing apparatus characterized by comprising
Input module, for obtaining target image;
Categorization module, for determining the image category of the target image by neural network classification model;
Chosen module, for determining target algorithm according to preset mapping data;
Wherein, the mapping data are used to indicate K image category and L compression algorithm, 2≤L≤K, the K image category
In an image category correspond to a compression algorithm in the L compression algorithm, i-th of image in the K image category
Compression algorithm corresponding to classification is to calculate in the L compression algorithm the optimal compression of i-th of image category compression quality
Method, the target algorithm are compression algorithm corresponding to the image category of the target image, K >=2,1≤i≤K;
Compression module, for being compressed according to the target algorithm to the target image.
8. device according to claim 7, which is characterized in that further include:
Base modules determine canonical algorithm corresponding to each sample image in the training sample for obtaining training sample;
Wherein, canonical algorithm corresponding to a sample image is to compress matter to one sample image in the L compression algorithm
Measure optimal compression algorithm;
Training module will correspond to same canonical algorithm for training the neural network classification model in the training sample
Sample image is classified as one kind.
9. device according to claim 8, which is characterized in that K=L, the K image category and the L compression are calculated
Method corresponds, and the chosen module includes:
Corresponding submodule, for basis and the one-to-one K compression algorithm of the K image category, the determining and target
The corresponding compression algorithm of the image category of image.
10. device according to claim 8, which is characterized in that the base modules include:
Computational submodule, for being pressed respectively by each compression algorithm in the L compression algorithm target sample image
Contracting, determines the corresponding compression quality parameter of each compression algorithm;
Submodule is screened, for determining the target sample image institute according to the corresponding compression quality parameter of each compression algorithm
Corresponding canonical algorithm.
11. device according to claim 10, which is characterized in that the computational submodule includes:
Parameters unit, for determine the corresponding compression ratio of each compression algorithm and Y-PSNR it is therein at least one.
12. device according to claim 7, which is characterized in that the input module includes:
It divides submodule and the image to be compressed is divided into image macro for obtaining image to be compressed, it will be described to be compressed
The image macro of image is as the target image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811067307.8A CN109376762A (en) | 2018-09-13 | 2018-09-13 | Image processing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811067307.8A CN109376762A (en) | 2018-09-13 | 2018-09-13 | Image processing method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109376762A true CN109376762A (en) | 2019-02-22 |
Family
ID=65405107
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811067307.8A Pending CN109376762A (en) | 2018-09-13 | 2018-09-13 | Image processing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109376762A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110475117A (en) * | 2019-09-26 | 2019-11-19 | 北京迈格威科技有限公司 | Method, apparatus, electronic equipment and the computer storage medium of compression of images |
CN110572652A (en) * | 2019-09-04 | 2019-12-13 | 锐捷网络股份有限公司 | Static image processing method and device |
CN111009017A (en) * | 2019-10-18 | 2020-04-14 | 珠海格力电器股份有限公司 | Picture compression method and device, storage medium and computer equipment |
CN115037937A (en) * | 2022-06-02 | 2022-09-09 | 北京新唐思创教育科技有限公司 | Image compression method, apparatus, device and medium |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101106709A (en) * | 2006-07-10 | 2008-01-16 | 逐点半导体(上海)有限公司 | Image compression method |
CN102957906A (en) * | 2011-08-29 | 2013-03-06 | 广州九游信息技术有限公司 | Method and system for image classifying compression |
CN105430394A (en) * | 2015-11-23 | 2016-03-23 | 小米科技有限责任公司 | Video data compression processing method, apparatus and equipment |
CN106385592A (en) * | 2016-08-31 | 2017-02-08 | 苏睿 | Image compression method and device |
CN107103632A (en) * | 2017-05-03 | 2017-08-29 | 西安万像电子科技有限公司 | Method for compressing image and device |
US20170249536A1 (en) * | 2016-02-29 | 2017-08-31 | Christopher J. Hillar | Self-Organizing Discrete Recurrent Network Digital Image Codec |
CN107301668A (en) * | 2017-06-14 | 2017-10-27 | 成都四方伟业软件股份有限公司 | A kind of picture compression method based on sparse matrix, convolutional neural networks |
CN108062780A (en) * | 2017-12-29 | 2018-05-22 | 百度在线网络技术(北京)有限公司 | Method for compressing image and device |
-
2018
- 2018-09-13 CN CN201811067307.8A patent/CN109376762A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101106709A (en) * | 2006-07-10 | 2008-01-16 | 逐点半导体(上海)有限公司 | Image compression method |
CN102957906A (en) * | 2011-08-29 | 2013-03-06 | 广州九游信息技术有限公司 | Method and system for image classifying compression |
CN105430394A (en) * | 2015-11-23 | 2016-03-23 | 小米科技有限责任公司 | Video data compression processing method, apparatus and equipment |
US20170249536A1 (en) * | 2016-02-29 | 2017-08-31 | Christopher J. Hillar | Self-Organizing Discrete Recurrent Network Digital Image Codec |
CN106385592A (en) * | 2016-08-31 | 2017-02-08 | 苏睿 | Image compression method and device |
CN107103632A (en) * | 2017-05-03 | 2017-08-29 | 西安万像电子科技有限公司 | Method for compressing image and device |
CN107301668A (en) * | 2017-06-14 | 2017-10-27 | 成都四方伟业软件股份有限公司 | A kind of picture compression method based on sparse matrix, convolutional neural networks |
CN108062780A (en) * | 2017-12-29 | 2018-05-22 | 百度在线网络技术(北京)有限公司 | Method for compressing image and device |
Non-Patent Citations (2)
Title |
---|
吴婧 等: "基于块分类的计算机桌面图像压缩算法", 《浙江大学学报(工学版)》 * |
马锐: "基于分类的图像压缩算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110572652A (en) * | 2019-09-04 | 2019-12-13 | 锐捷网络股份有限公司 | Static image processing method and device |
CN110572652B (en) * | 2019-09-04 | 2021-07-16 | 锐捷网络股份有限公司 | Static image processing method and device |
CN110475117A (en) * | 2019-09-26 | 2019-11-19 | 北京迈格威科技有限公司 | Method, apparatus, electronic equipment and the computer storage medium of compression of images |
CN110475117B (en) * | 2019-09-26 | 2022-03-29 | 北京迈格威科技有限公司 | Image compression method and device, electronic equipment and computer storage medium |
CN111009017A (en) * | 2019-10-18 | 2020-04-14 | 珠海格力电器股份有限公司 | Picture compression method and device, storage medium and computer equipment |
CN111009017B (en) * | 2019-10-18 | 2021-08-17 | 珠海格力电器股份有限公司 | Picture compression method and device, storage medium and computer equipment |
CN115037937A (en) * | 2022-06-02 | 2022-09-09 | 北京新唐思创教育科技有限公司 | Image compression method, apparatus, device and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109376762A (en) | Image processing method and device | |
JP7413400B2 (en) | Skin quality measurement method, skin quality classification method, skin quality measurement device, electronic equipment and storage medium | |
US11095925B2 (en) | Artificial intelligence based resolution improvement system | |
CN111131889B (en) | Method and system for adaptively adjusting images and sounds in scene and readable storage medium | |
CN110637460B (en) | Visual quality preserving quantitative parameter prediction using deep neural networks | |
JP2022519469A (en) | Image quality evaluation method and equipment | |
CN109784153A (en) | Emotion identification method, apparatus, computer equipment and storage medium | |
CN107680077A (en) | A kind of non-reference picture quality appraisement method based on multistage Gradient Features | |
CN108830823B (en) | Full-reference image quality evaluation method based on spatial domain combined frequency domain analysis | |
CN108564066A (en) | A kind of person recognition model training method and character recognition method | |
US11259029B2 (en) | Method, device, apparatus for predicting video coding complexity and storage medium | |
CN108198177A (en) | Image acquiring method, device, terminal and storage medium | |
CN110517225B (en) | AI image recognition method, apparatus, device and storage medium | |
CN109062715B (en) | Method and device for determining memory clock frequency and terminal | |
WO2020181706A1 (en) | Plant species identification method and apparatus | |
CN111047543A (en) | Image enhancement method, device and storage medium | |
CN110120034B (en) | Image quality evaluation method related to visual perception | |
CN110298405A (en) | Classification recognition methods and device, storage medium and terminal | |
CN111630854A (en) | System and method for image signal processor tuning | |
Morzelona | Human visual system quality assessment in the images using the IQA model integrated with automated machine learning model | |
JP2014056442A (en) | Image quality estimation device, image quality estimation method and image quality estimation program | |
CN112116102A (en) | Method and system for expanding domain adaptive training set | |
CN110766631A (en) | Face image modification method and device, electronic equipment and computer readable medium | |
JP7362924B2 (en) | Data augmentation-based spatial analysis model learning device and method | |
CN115457015A (en) | Image no-reference quality evaluation method and device based on visual interactive perception double-flow network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190222 |