CN110084843A - A kind of method for compressing image based on deep learning applied to furniture 3 D-printing - Google Patents
A kind of method for compressing image based on deep learning applied to furniture 3 D-printing Download PDFInfo
- Publication number
- CN110084843A CN110084843A CN201910327972.4A CN201910327972A CN110084843A CN 110084843 A CN110084843 A CN 110084843A CN 201910327972 A CN201910327972 A CN 201910327972A CN 110084843 A CN110084843 A CN 110084843A
- Authority
- CN
- China
- Prior art keywords
- image
- furniture
- network model
- image data
- neural network
- 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.)
- Withdrawn
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B29—WORKING OF PLASTICS; WORKING OF SUBSTANCES IN A PLASTIC STATE IN GENERAL
- B29C—SHAPING OR JOINING OF PLASTICS; SHAPING OF MATERIAL IN A PLASTIC STATE, NOT OTHERWISE PROVIDED FOR; AFTER-TREATMENT OF THE SHAPED PRODUCTS, e.g. REPAIRING
- B29C64/00—Additive manufacturing, i.e. manufacturing of three-dimensional [3D] objects by additive deposition, additive agglomeration or additive layering, e.g. by 3D printing, stereolithography or selective laser sintering
- B29C64/30—Auxiliary operations or equipment
- B29C64/386—Data acquisition or data processing for additive manufacturing
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B33—ADDITIVE MANUFACTURING TECHNOLOGY
- B33Y—ADDITIVE MANUFACTURING, i.e. MANUFACTURING OF THREE-DIMENSIONAL [3-D] OBJECTS BY ADDITIVE DEPOSITION, ADDITIVE AGGLOMERATION OR ADDITIVE LAYERING, e.g. BY 3-D PRINTING, STEREOLITHOGRAPHY OR SELECTIVE LASER SINTERING
- B33Y50/00—Data acquisition or data processing for additive manufacturing
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
- H04N13/10—Processing, recording or transmission of stereoscopic or multi-view image signals
- H04N13/106—Processing image signals
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
- H04N13/10—Processing, recording or transmission of stereoscopic or multi-view image signals
- H04N13/106—Processing image signals
- H04N13/161—Encoding, multiplexing or demultiplexing different image signal components
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Materials Engineering (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Manufacturing & Machinery (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Mathematical Physics (AREA)
- Mechanical Engineering (AREA)
- Optics & Photonics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention discloses a kind of method for compressing image based on deep learning applied to furniture 3 D-printing, comprising: acquisition furniture surface texture image data;Quantization encoding is carried out to the furniture surface texture image data;Construct image entropy neural network model;Image data input picture entropy neural network model after the quantization encoding is trained, compression of images network model is obtained;Furniture surface texture image data to be printed is inputted into described image compression network model, obtains compressed furniture surface texture image data;Compressed image data is passed to three printers again and executes decompression, printing process;Neural network model of the invention can effectively compress the furniture texture image with abundant structural information and unique geometric error modeling, reduce the mean square error between original image and decompressed image to the maximum extent;When the present invention is applied to furniture 3 D-printing, memory footprint can be effectively reduced, improve the resolution ratio of print image image.
Description
Technical field
The invention belongs to Computer Image Processing field, it is specifically a kind of applied to furniture 3 D-printing based on depth
The method for compressing image of habit.
Background technique
With three-dimensional printing technology constantly improve and the quick hair of the digital image processing techniques based on machine learning
Exhibition constructs the 3 D-printing of object by layer-by-layer printing with can bond new material based on mathematical model
Technology gradually applies to all trades and professions.And in the process flow of China's modern production furniture, 3 D-printing by
Step is applied to Furniture modeling design innovation, furniture material innovation etc..However since 3 D-printing itself technique limits, and family
Have texture image color and vein and geometric error modeling is more complicated, image fining requires height, and resolution ratio is inadequate, often leads to print
Output image has obvious machine printing trace.Meanwhile the image of high resolution often needs to occupy because of 3 D-printing memory problem
A large amount of memory headrooms are so as to cause printing failure.The technology for replacing Wood surface texture to create limitation by hand using machine printing is difficult
Topic has perplexed the Furniture manufacturing industry several years.
In existing method for compressing image, JPEG will appear apparent blocking artifact in low rate encoding, and encounter
Picture quality will receive badly damaged when bit error;JPEG2000 encodes computational complexity height and to text image and composite diagram
The Lossless Image Compression performance of picture is low.
In recent years, the full resolution image compress technique based on DNNS, generallys use context model and recurrent neural net
Network (RNN) makes network by that can reduce the mean square error (MSE) between original image and decompressed image after training to greatest extent.Base
It is higher for the adaptability of particular target domain (i.e. finite field) in the study compressibility of DNN, it can be realized in these domains higher
Compression ratio.The image compress processing method of the prior art lacks the coding method towards three-dimensional even multidimensional image, tradition pressure
Contracting method is not able to satisfy the performance requirement of furniture textured pattern compression;Lack and is instructed simultaneously using context model as rate distorterence term
Practice the network that two encoders improve depth image compression performance.
Summary of the invention
In response to the problems existing in the prior art, the purpose of the present invention is to provide it is a kind of applied to furniture 3 D-printing based on
The method for compressing image of deep learning, after improving textured pattern compression reconfiguration by training autocoder and entropy model mutually
Resolution ratio, there is the furniture texture image of abundant structural information and unique texture by compressing, reduce memory footprint, mention
The resolution ratio of high print image.
To achieve the above object, the technical solution adopted by the present invention is that:
A kind of method for compressing image based on deep learning applied to furniture 3 D-printing, comprising the following steps:
S1 acquires furniture surface texture image data;
S2 carries out quantization encoding to the furniture surface texture image data;
S3 constructs image entropy neural network model;
Image data input picture entropy neural network model after the quantization encoding is trained, obtains image by S4
Compression network model;
Furniture surface texture image data to be printed is inputted described image compression network model, after obtaining compression by S5
Furniture surface texture image data.
Specifically, in step S1, the furniture surface texture image is color image or gray level image;
The size of described image is n*n, wherein n=2k, k >=5, k are integer;
The size of described image is 100k~500k.
Specifically, in step S2, the method for the quantization encoding are as follows: define L quantization central point C={ c on the image1,
c2,…,cL, image is quantified using pixel each on range image nearest quantization central point, specific formula is as follows:
Wherein, (1, L) j ∈, xiFor the pixel value of each point in image,For the quantized value that image hard quantization directly obtains,
For the quantized value obtained after the soft quantization of image, the quantized value that two amounts method obtains is different, can be with using different quantification manners
It reduces image quantization and is distorted ratio.
In step S2, to the furniture surface texture image data carry out quantization encoding after, it is also necessary to image data into
Row classification, respectively training data and test data;The training data is for training neural network model;The test data
For testing trained neural network model, the low neural network model of distortion rate is obtained.
Specifically, in step S3, the method for building described image entropy neural network model are as follows:
Based on PixelRNN and factorization assignment itemConstruct entropyModel, formula are as follows:
Context model is constituted using convolution autocoder and lightweight 3D-CNN, the image after inputting as quantization encoding
Data export as compressed image data.
Specifically, in step S4, during training neural network model, ladder is taken to autocoder and quantizer
Algorithm tradeoff distortion is spent, entropy model is updated;The distortion is the mean square error between original image and compression image;It loses
True cost weighs formula are as follows:
Wherein,For amount distortion, α is with reference to coefficient, no fixed value, and training pattern will update its iteration, and tradeoff is lost
True cost makes the value of above-mentioned formula reach minimum to reach the smallest scheme of distortion cost.
In step S4, after obtaining described image compression network model, need input test data to compression of images network mould
Type is tested, using compression ratio and distortion than measuring image compression quality, to obtain lesser distortion ratio.
Compared with prior art, the beneficial effects of the present invention are: the present invention passes through the deep learning side based on compression of images
Method, effectively study compression has the furniture surface coating pattern of abundant structural information and color, geometric error modeling, to make this hair
Bright neural network model can effectively compress the furniture texture image with abundant structural information and unique geometric error modeling, maximum
Reduce to limit the mean square error between original image and decompressed image;When the present invention is applied to furniture 3 D-printing, Ke Yiyou
Effect reduces memory footprint, improves the resolution ratio of print image image.
Detailed description of the invention
Fig. 1 is that a kind of process of the method for compressing image based on deep learning applied to furniture 3 D-printing of the present invention is shown
Meaning block diagram;
Fig. 2 is the structural schematic block diagram of compression of images network model in the present invention.
Specific embodiment
Below in conjunction with the attached drawing in the present invention, technical solution of the present invention is clearly and completely described, it is clear that
Described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on the implementation in the present invention
Example, those of ordinary skill in the art's all other embodiment obtained under the conditions of not making creative work belong to
The scope of protection of the invention.
As shown in Figure 1, 2, a kind of image based on deep learning applied to furniture 3 D-printing is present embodiments provided
Compression method, specifically includes the following steps:
S1, acquisition is largely with the image data of furniture textured pattern;
S2 carries out quantization encoding to acquired image data and convolution initializes;
S3 constructs neural network model, the image data small lot after quantization encoding is iterated in neural network model
It is trained, takes gradient algorithm tradeoff distortion in each iteration for autocoder and quantizer, constantly update nerve net
Network model;
Test data is inputted and carries out compressed and decompressed test in trained neural network model, utilizes compression ratio by S4
With distortion than measuring image compression quality, to obtain sufficiently small distortion ratio;
S6 inputs the furniture surface textured pattern to be printed, neural network output compression in the neural network debugged
Good image data;
The image data of compression is passed to 3D printing equipment by S7, executes decompression, printing process.
Specifically, in step S1, the furniture surface texture image is color image or gray level image, and colored phase can be used
Machine or gray scale camera acquire image data;
The size of described image is n*n, wherein n=2k, k >=5, k are integer;
The size of described image is 100k~500k.
Specifically, in step S2, the method for the quantization encoding are as follows: define L quantization central point C={ c on the image1,
c2,…,cL, image is quantified using pixel each on range image nearest quantization central point, specific formula is as follows:
The formula relies on softening amount that can be micro- come the gradient after calculating into transmittance process;Wherein, (1, L) j ∈, xiFor figure
The pixel value of each point as in,For the quantized value that image hard quantization directly obtains,For the quantized value obtained after the soft quantization of image,
The quantized value that two amounts method obtains is different, using different quantification manners, can reduce image quantization distortion ratio.
In step S2, to the furniture surface texture image data carry out quantization encoding after, it is also necessary to image data into
Row classification, respectively training data and test data;The training data is for training neural network model;The test data
For testing trained neural network model, the low neural network model of distortion rate is obtained;The test data is extremely
10000 images are needed less.
Specifically, in step S3, the method for building described image entropy neural network model are as follows:
Based on PixelRNN and factorization assignment itemConstruct entropyModel, formula are as follows:
It, can be with using known cross entropy property as Coding cost when replacing true distribution p using Fault Distribution q
Obtain above-mentioned calculating, the estimation that we can regard CE as.Therefore, when being trained to autocoder, friendship can be passed through
Fork entropy CE is minimized indirectly
In the formula, using the known properties of cross entropy as Coding cost when Fault DistributionThe correct distribution of substitution
CostConvolution autocoder and lightweight 3D-CNN is recycled to constitute neural network context model, the model
Input is the image data after quantization encoding, is exported as compressed image data.
Specifically, in step S4, during training neural network model, ladder is taken to autocoder and quantizer
Algorithm tradeoff distortion is spent, entropy model is updated;The distortion is the mean square error between original image and compression image;It loses
True cost weighs formula are as follows:
Wherein,For amount distortion, α is with reference to coefficient, and no fixed value, training pattern will update its iteration, to reach
To distortion cost minimum programme, the value of above-mentioned formula is made to reach minimum.
The present embodiment carries out compression study using the context model of deep learning and establishes the compressed web towards furniture image
Network, compared to conventional method, the technique upgrading space and industrial production value after network foundation are larger, can be effectively reduced interior
Occupied space is deposited, the resolution ratio of print image image is improved.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (7)
1. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing, which is characterized in that including following
Step:
S1 acquires furniture surface texture image data;
S2 carries out quantization encoding to the furniture surface texture image data;
S3 constructs image entropy neural network model;
Image data input picture entropy neural network model after the quantization encoding is trained, obtains compression of images by S4
Network model;
Furniture surface texture image data to be printed is inputted described image compression network model, obtains compressed family by S5
Have skin texture images data.
2. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing according to claim 1,
It is characterized in that, the furniture surface texture image is color image or gray level image in step S1;
The size of described image is n*n, wherein n=2k, k >=5, k are integer;
The size of described image is 100k~500k.
3. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing according to claim 1,
It is characterized in that, in step S2, the method for the quantization encoding are as follows: define L quantization central point C={ c on the image1,
c2,…,cL, image is quantified using pixel each on range image nearest quantization central point, specific formula is as follows:
Wherein, (1, L) j ∈, xiFor the pixel value of each point in image,For the quantized value that image hard quantization directly obtains,For figure
As the quantized value obtained after soft quantization.
4. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing according to claim 1,
It is characterized in that, in step S2, after carrying out quantization encoding to the furniture surface texture image data, it is also necessary to image data
Classify, respectively training data and test data.
5. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing according to claim 1,
It is characterized in that, in step S3, the method for building described image entropy neural network model are as follows:
Based on PixelRNN and factorization assignment itemConstruct entropyModel, formula are as follows:
Wherein, Coding cost when p is Fault Distribution, Coding cost when q is correct distribution, CE areEstimated value,
When being trained to autocoder, minimized indirectly by cross entropy CEValue;
Neural network context model is constituted using convolution autocoder and lightweight 3D-CNN, is inputted as after quantization encoding
Image data exports as compressed image data.
6. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing according to claim 1,
It is characterized in that, during training neural network model, taking gradient to calculate autocoder and quantizer in step S4
Right weighing apparatus distortion, is updated entropy model;The distortion is the mean square error between original image and compression image;It is distorted generation
Valence weighs formula are as follows:
Wherein,For amount distortion, α is with reference to coefficient, no fixed value.
7. a kind of method for compressing image based on deep learning applied to furniture 3 D-printing according to claim 1,
It is characterized in that, after obtaining described image compression network model, needing input test data to compression of images network in step S4
Model is tested.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910327972.4A CN110084843A (en) | 2019-04-23 | 2019-04-23 | A kind of method for compressing image based on deep learning applied to furniture 3 D-printing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910327972.4A CN110084843A (en) | 2019-04-23 | 2019-04-23 | A kind of method for compressing image based on deep learning applied to furniture 3 D-printing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110084843A true CN110084843A (en) | 2019-08-02 |
Family
ID=67416223
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910327972.4A Withdrawn CN110084843A (en) | 2019-04-23 | 2019-04-23 | A kind of method for compressing image based on deep learning applied to furniture 3 D-printing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110084843A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110411308A (en) * | 2019-08-14 | 2019-11-05 | 浙江德尔达医疗科技有限公司 | A kind of accuracy checking method of customized type 3D printing model |
CN112659548A (en) * | 2020-11-06 | 2021-04-16 | 西安交通大学 | Surface exposure 3D printing process optimization method based on genetic algorithm and BP neural network |
CN113079377A (en) * | 2021-04-01 | 2021-07-06 | 中国科学技术大学 | Training method for depth image/video compression network |
-
2019
- 2019-04-23 CN CN201910327972.4A patent/CN110084843A/en not_active Withdrawn
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110411308A (en) * | 2019-08-14 | 2019-11-05 | 浙江德尔达医疗科技有限公司 | A kind of accuracy checking method of customized type 3D printing model |
CN110411308B (en) * | 2019-08-14 | 2021-08-24 | 浙江德尔达医疗科技有限公司 | Precision detection method for customized 3D printing model |
CN112659548A (en) * | 2020-11-06 | 2021-04-16 | 西安交通大学 | Surface exposure 3D printing process optimization method based on genetic algorithm and BP neural network |
CN113079377A (en) * | 2021-04-01 | 2021-07-06 | 中国科学技术大学 | Training method for depth image/video compression network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106157339B (en) | The animated Mesh sequence compaction method extracted based on low-rank vertex trajectories subspace | |
CN110084843A (en) | A kind of method for compressing image based on deep learning applied to furniture 3 D-printing | |
CN109685743B (en) | Image mixed noise elimination method based on noise learning neural network model | |
CN106254879B (en) | A kind of application encodes the Image Lossy Compression method of neural network certainly | |
CN108154499B (en) | Woven fabric texture flaw detection method based on K-SVD learning dictionary | |
CN108038503B (en) | Woven fabric texture characterization method based on K-SVD learning dictionary | |
CN101835048A (en) | By carry out the method and apparatus of video coding based on the just noticeable difference model of ABT | |
CN110248190A (en) | A kind of compressed sensing based multilayer residual error coefficient image encoding method | |
CN103700074B (en) | Based on the self-adapting compressing perception method of sampling of discrete cosine transform coefficient distribution | |
CN108898568B (en) | Image synthesis method and device | |
Fan et al. | D-dpcc: Deep dynamic point cloud compression via 3d motion prediction | |
CN107018410B (en) | A kind of non-reference picture quality appraisement method based on pre- attention mechanism and spatial dependence | |
CN108111852A (en) | Towards the double measurement parameter rate distortion control methods for quantifying splits' positions perceptual coding | |
Gao et al. | Rate-distortion modeling for bit rate constrained point cloud compression | |
Harell et al. | Rate-distortion in image coding for machines | |
Joshua et al. | Comparison of DCT and DWT image compression | |
CN104299256B (en) | Almost-lossless compression domain volume rendering method for three-dimensional volume data | |
CN111083498B (en) | Model training method and using method for video coding inter-frame loop filtering | |
Bletterer et al. | Point cloud compression using depth maps | |
CN107146260A (en) | A kind of compression of images based on mean square error perceives the method for sampling | |
Gao et al. | Quality constrained compression using DWT-based image quality metric | |
CN107203991A (en) | A kind of half reference image quality appraisement method based on spectrum residual error | |
CN111161363A (en) | Image coding model training method and device | |
CN115294010A (en) | Method for evaluating quality of reference point cloud based on support vector machine | |
CN107705249A (en) | Image super-resolution method based on depth measure study |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190802 |
|
WW01 | Invention patent application withdrawn after publication |