CN101668204A - Immune clone image compression method - Google Patents
Immune clone image compression method Download PDFInfo
- Publication number
- CN101668204A CN101668204A CN 200910024059 CN200910024059A CN101668204A CN 101668204 A CN101668204 A CN 101668204A CN 200910024059 CN200910024059 CN 200910024059 CN 200910024059 A CN200910024059 A CN 200910024059A CN 101668204 A CN101668204 A CN 101668204A
- Authority
- CN
- China
- Prior art keywords
- image
- code book
- average
- vector
- residual vector
- 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
Images
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses an immune clone image compression method mainly solving the problems of low compression speed and compression ratio and unclear compression images of the prior art. The immuneclone image compression method comprises the following implementation steps: (1) selecting a natural image with clear texture and high contrast as a training sample, dividing the training sample intoblocks and extracting an average value and a residential vector of the training sample; (2) respectively using an immune clone clustering algorithm to the average value and the residential vector of the training sample to generate an average value code book and a residential vector code book; (3) dividing a testing image into blocks and solving an average value and a residential vector of the testing image; (4) using the code book generated in the step (2) to quantize and encode the average vector value and the residential vector of the testing image; (5) transmitting quantized and encoded data to a client; (6) decoding the data received at the client according to the code books; and (7) recovering the decoded data into the image. The invention can be used for the fields of image compression, image transmission, and the like under a constrained bandwidth condition in third-generation mobile communication.
Description
Technical field
The invention belongs to artificial intelligence technology and digital image processing field, relate to of the application of a kind of immune clone clustering method, specifically a kind of immune clone image compression method in digital image processing field.This method can be used for solving in 3G (Third Generation) Moblie and the network service, under network bandwidth condition of limited, how effectively to realize the problem of image transmission.
Background technology
Along with the maturation gradually of 3G (Third Generation) Moblie technology, standardization and the development of 3G enter substantial phase, and image and video transmission will become the important application in the 3G (Third Generation) Moblie.This has just proposed new challenge to existing digital image compression technology.Existing Image Compression as the JPEG technology, adopts the method for transform domain coding, earlier image is carried out discrete cosine transform, carry out compressed encoding then, because discrete cosine transform needs a large amount of operation time, thereby the speed of image compression encoding decoding is very slow.And traditional employing M/RVQ Methods for Coding is used LBG, KMEANS, and methods such as SOM produce code book, then the average and the residual vector of image are carried out coding and decoding, though arithmetic speed has been accelerated, the image compression rate is all relative with image compression quality relatively poor.These methods mainly are to be optimized at internet transmission and terminal, reckon without wireless network and pass bandwidth more, the fact that the cell-phone customer terminal computational speed is slower makes traditional method for compressing image, in the time of on being applied to 3G network and mobile phone, effect is unsatisfactory.This just impels Many researchers to attempt some new schemes, proposes new method for compressing image, solves many-sides such as image compression rate, image compression speed, image decompression speed, computational complexity and requires to be difficult to satisfied simultaneously contradiction
Summary of the invention
The objective of the invention is to overcome the deficiency of above-mentioned prior art, propose a kind of immune clone image compression method, the time that increase the image compression rate to reach, reduces image compression and decompress(ion), and reduce computational complexity.
Technical scheme of the present invention is to select earlier some representative images as training image, and training image is divided into the fritter of n*n, produces code book with M/RVQ and immune clone point symmetry clustering method then.For the image of to be compressed/decompress(ion), it is divided into piece after, the contrast code book carries out Code And Decode.Code book is can training in advance good, is cured in the coding decoder, and can realizes with hardware, thereby realize the high spped coding decoding, and concrete steps comprise as follows:
(1) selects the high natural image of clean mark and contrast as training sample, and this training sample is divided into piece, be converted into vector, extract average and residual vector;
(2) use the immune clone clustering algorithm to generate average code book and residual vector code book respectively to the average and the residual vector of training sample;
(3) test pattern is divided into piece, every as a vector, obtains test pattern average and residual vector;
(4) code book that uses (2) to produce carries out quantization encoding to the vectorial average and the residual vector of test pattern;
(5) give client with the data behind the quantization encoding by Channel Transmission;
(6) in client the data based code book that receives is decoded;
(7) decoded reduction of data is become image.
Because the present invention extracted average and residual vector, make the brightness of image separate, and make cluster centre, thereby make edge of image protect better more near the edge with textural characteristics, make the image border after the compression more clear; Because the present invention uses the immune clone clustering algorithm to find out forced coding, effectively utilize the symmetry of image, thereby obtained bigger image compression rate and better pictures compression quality simultaneously; In addition because the present invention has used image average and residual vector coding, avoided the big mathematic(al) manipulation of amount of calculation such as discrete cosine transform, wavelet transformation in the Image Compression commonly used, image compression and decompress(ion) speed have faster been obtained, the time of having reduced the image compression decompress(ion).
Description of drawings
Fig. 1 is a FB(flow block) of the present invention;
Fig. 2 is original test pattern lena, boat and the airplane that uses among the contrast experiment.
Fig. 3 is with existing method and the inventive method compression effectiveness comparison diagram to image lena;
Fig. 4 is with existing method and the inventive method compression effectiveness comparison diagram to image boat;
Fig. 5 is with existing method and the inventive method compression effectiveness comparison diagram to image airplane.
Embodiment
With reference to Fig. 1 and foregoing performing step, the present invention mainly comprises three parts: code book design, image encoding and picture decoding, and in the performing step, the code book design is carried out in step (1), (2), image encoding is carried out in step (3), (4), and picture decoding is carried out in step (6), (7).Introduce the concrete implementation step of code book design, image encoding and picture decoding below respectively.
Code book comprises average code book codebook1 and two parts of residual vector code book codebook2, and the size of these two code books is respectively b1 and b2, and b1 gets 16 in the experiment, and b2 gets 256.
1. code book designs performing step
1.1) establish the big or small r*c of being of image, the size that training image is divided into non-overlapping copies is the fritter of n*n, n gets 4 in the experiment, if the not enough n*n in image border then replenishes 0, total m=r/n*c/n fritter.The corresponding matrix of the gray value of every fritter is converted into the gray scale vector with this matrix by from top to bottom permanent order from left to right;
1.2) calculate the average and the residual vector of each gray scale vector.Wherein average is the mean value of gray scale vector, and residual vector deducts mean value for the gray scale vector;
1.3) to the average x of all gray scale vectors
1, x
2..., x
m, ask the minimum value of cluster exponential function with existing immune clone algorithm, the independent variable when the cluster exponential function is obtained minimum value is average code book codebook1.The computing formula of cluster exponential function is:
C wherein
1, c
2..., c
B1Be the average code book that to find the solution, dist (x
i, c
j) be x
iTo c
jThe point symmetry distance;
1.4) to the residual vector y of all gray scale vectors
1, y
2..., y
m, ask the minimum value of cluster exponential function with existing immune clone algorithm, the independent variable when the cluster exponential function is obtained minimum value is the residual vector code book, and the computing formula of cluster exponential function is:
C wherein
1', c
2' ..., c
B2' be the residual vector code book that will find the solution, dist (y
i, c
j) be y
iTo c
j' the point symmetry distance.
2. image encoding performing step
2.1) size that image division is become non-overlapping copies is the fritter of n*n, n gets 4 in the experiment, if the not enough n*n in image border then replenishes 0.The corresponding matrix of the gray value of every fritter is converted into the gray scale vector with this matrix by from top to bottom permanent order from left to right;
2.2) calculate the average x and the residual vector y of each gray scale vector, wherein average x is a gray scale vector mean value, residual vector y deducts mean value for the gray scale vector;
2.3) to the average x and the residual vector y of each gray scale vector, in codebook1, search near the coding cx of x, in codebook2, search coding cy near y; With cx, cy connects into [cx, cy] coding as this vector;
2.4) coding of each vector is merged, as the coding of entire image.
3. picture decoding performing step
3.1) according to the code length of each gray scale vector, the coding of image is decomposed into the coding [cx, cy] of each gray scale vector;
3.2) coding [cx, cy] of each gray scale vector is decomposed into average coding cx and residual vector coding cy, in codebook1, search the average x of cx correspondence, in codebook2, search the residual vector y of cy correspondence;
3.3) with x and y addition, obtain gray scale vector x+y, the gray scale vector is reduced into the matrix of n*n, the i.e. gray value sequence of each small images by from left to right from top to bottom the permanent order in when coding;
3.4) with each image fritter with the tiling of each image fritter non-overlapping copies ground, and the edge mended 0 of 0 fritter and removes the image after obtaining compressing.
Emulation experiment
Effect of the present invention can further specify by following experiment:
1. simulated conditions:
At CPU is to use MATLAB to carry out emulation in core22.4HZ, internal memory 2G, the WINDOWS XP system.
2. emulation content:
Select three width of cloth images in the accompanying drawing 2 to be used as test, and contrast with three kinds of method for compressing image such as existing LBG, SOM, MOD-KMEANS.Three width of cloth test patterns are respectively the airplane images shown in the boat image shown in the lena image shown in Fig. 2 (a), Fig. 2 (b) and Fig. 2 (c).
Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) and Fig. 3 (d) are respectively LBG, SOM, MOD-KMEANS and the inventive method compression result to test pattern lena, can see, Fig. 3 (a) granular sensation is more serious, Fig. 3 (b) blur margin is clear, Fig. 3 (c) character facial is jagged, and Fig. 3 as a result (d) of the inventive method compression, edge clear, burr is few, more approaches original image.
Fig. 4 (a), Fig. 4 (b), Fig. 4 (c) and Fig. 4 (d) are respectively LBG, SOM, MOD-KMEANS and the inventive method compression result to test pattern boat, can see, Fig. 4 (a) distortion is apparent in view, mast zigzag fashion in Fig. 4 (b) image is very outstanding, the literal of Fig. 4 (c) fore is very fuzzy, and Fig. 4 as a result (d) of the inventive method compression, the mast edge keeps better, the fore literal is also more clear, and is more approaching with original image.
Fig. 5 (a), Fig. 5 (b), Fig. 5 (c) and Fig. 5 (d) are respectively LBG, SOM, MOD-KMEANS and the inventive method compression result to test pattern airplane, can see, Fig. 5 (a) lacks unity and coherence clear, Fig. 5 (b) cloud layer details has loses, F16 sign on Fig. 5 (c) aircraft is very fuzzy, and Fig. 5 as a result (d) of the inventive method compression, F16 sign on the aircraft is obviously good than other three kinds of methods, details is also clear than other three kinds of methods, and integral image also more approaches original image.
Evaluation map is a Y-PSNR PSNR value as the most frequently used technical indicator of compression effectiveness, and the PSNR value is big more, show that the difference of image after the compression and original image is more little, thereby the compression effectiveness of image is good more.At above-mentioned three width of cloth test patterns, to existing LBG, SOM, MOD-KMEANS method and the inventive method, tested its Y-PSNR PSNR value respectively, provide the PSNR value table of four kinds of methods below to each image compression.
Each method of table 1 is compressed the PSNR (unit: db) that obtains to 3 width of cloth test patterns.
| ??LBG | ??SOM | ??MOD-KMEANS | The inventive method | |
| ??Lena | ??25.86549 | ??26.26231 | ??26.06883 | ??26.6117 |
| ??Boat | ??24.22942 | ??24.2959 | ??26.276384 | ??27.86095 |
| ??airplane | ??23.85141 | ??23.96298 | ??27.30211 | ??27.54689 |
As can be seen from Table 1, the inventive method is all bigger than the Y-PSNR PSNR value of other three kinds of methods.Illustrate that the relative original image distortion of image after this method compression is less, go up that the inventive method is more superior than other three kinds of methods from technical indicator PSNR.
Claims (6)
1, a kind of immune clone image compression method comprises the steps:
(1) selects the high natural image of clean mark and contrast as training sample, and this training sample is divided into piece, be converted into vector, extract average and residual vector;
(2) use the immune clone clustering algorithm to generate average code book and residual vector code book respectively to the average and the residual vector of training sample;
(3) test pattern is divided into piece, every as a vector, obtains test pattern average and residual vector;
(4) code book that uses step (2) to produce carries out quantization encoding to the vectorial average and the residual vector of test pattern;
(5) give client with the data behind the quantization encoding by Channel Transmission;
(6) in client the data based code book that receives is decoded;
(7) decoded reduction of data is become image.
2, immune clone image compression method according to claim 1 wherein is divided into piece with training image and test pattern, is to be the fritter of equal sizes with image division, with the gray value sequence of each little image as a vector.
3, immune clone image compression method according to claim 1, wherein use the immune clone clustering algorithm to generate code book, be to use the immune clone clustering algorithm to obtain cluster centre, as average code book and residual vector code book, this average code book and vectorial code book are solidificated in the encoder with these centers.
4, immune clone image compression method according to claim 1, quantization encoding wherein is to select to approach most the coding of the index of this average as this average in the average code book; In the residual vector code book, select near the index of this residual vector coding as this residual vector.
5, immune clone image compression method according to claim 1, wherein the data based code book that receives is decoded in client, be meant average coding and the residual vector coding of client, be converted to data corresponding in average code book and the residual vector code book respectively receiving.
6, immune clone image compression method according to claim 1, wherein decoded reduction of data being become image, is average and the residual vector addition that will obtain after will decoding earlier, obtains the gray value of a little image block, then these little images are merged into big image, form the image after compressing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910024059 CN101668204A (en) | 2009-09-25 | 2009-09-25 | Immune clone image compression method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910024059 CN101668204A (en) | 2009-09-25 | 2009-09-25 | Immune clone image compression method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101668204A true CN101668204A (en) | 2010-03-10 |
Family
ID=41804605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910024059 Pending CN101668204A (en) | 2009-09-25 | 2009-09-25 | Immune clone image compression method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101668204A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101847263A (en) * | 2010-06-04 | 2010-09-29 | 西安电子科技大学 | Unsupervised image division method based on multi-target immune cluster integration |
| CN102970510A (en) * | 2012-11-23 | 2013-03-13 | 清华大学 | Method for transmitting human face video |
| CN103051900A (en) * | 2013-01-05 | 2013-04-17 | 东华大学 | Image compression method based on wavelet transform and clonal selection algorithm |
| CN113766237A (en) * | 2021-09-30 | 2021-12-07 | 咪咕文化科技有限公司 | Encoding method, decoding method, device, equipment and readable storage medium |
| WO2023222313A1 (en) * | 2022-05-20 | 2023-11-23 | Nokia Technologies Oy | A method, an apparatus and a computer program product for machine learning |
-
2009
- 2009-09-25 CN CN 200910024059 patent/CN101668204A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101847263A (en) * | 2010-06-04 | 2010-09-29 | 西安电子科技大学 | Unsupervised image division method based on multi-target immune cluster integration |
| CN101847263B (en) * | 2010-06-04 | 2012-02-08 | 西安电子科技大学 | Unsupervised image division method based on multi-target immune cluster integration |
| CN102970510A (en) * | 2012-11-23 | 2013-03-13 | 清华大学 | Method for transmitting human face video |
| CN102970510B (en) * | 2012-11-23 | 2015-04-15 | 清华大学 | Method for transmitting human face video |
| CN103051900A (en) * | 2013-01-05 | 2013-04-17 | 东华大学 | Image compression method based on wavelet transform and clonal selection algorithm |
| CN103051900B (en) * | 2013-01-05 | 2016-01-20 | 东华大学 | A kind of method for compressing image based on wavelet transformation and clonal selection algorithm |
| CN113766237A (en) * | 2021-09-30 | 2021-12-07 | 咪咕文化科技有限公司 | Encoding method, decoding method, device, equipment and readable storage medium |
| WO2023222313A1 (en) * | 2022-05-20 | 2023-11-23 | Nokia Technologies Oy | A method, an apparatus and a computer program product for machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cheng et al. | Learned image compression with discretized gaussian mixture likelihoods and attention modules | |
| Li et al. | Learning convolutional networks for content-weighted image compression | |
| Li et al. | An efficient deep convolutional neural networks model for compressed image deblocking | |
| Liu et al. | Data-driven soft decoding of compressed images in dual transform-pixel domain | |
| US6671413B1 (en) | Embedded and efficient low-complexity hierarchical image coder and corresponding methods therefor | |
| CN102088606B (en) | Sparse representation-based deblocking method | |
| CN103533458B (en) | The embedding of a kind of video watermark and extracting method | |
| Wang et al. | Hybrid image coding based on partial fractal mapping | |
| KR100809409B1 (en) | Decoding apparatus, inverse quantization method, and computer readable medium recorded with the program | |
| WO2003058975A2 (en) | Method of compressing images of arbitrarily shaped objects | |
| CN104023216A (en) | Face video compression method | |
| CN101668204A (en) | Immune clone image compression method | |
| Miya et al. | Wavelet techniques for medical images performance analysis and observations with EZW and underwater image processing | |
| Niu et al. | Edge-based perceptual image coding | |
| Jiang et al. | A hybrid image compression algorithm based on human visual system | |
| CN101267557B (en) | A method for image compression based on compound vector quantification | |
| Selim et al. | A simplified fractal image compression algorithm | |
| CN1822050A (en) | Full phase anticosine double orthogonal transformation and its improving method for JPEG | |
| Chen et al. | Research on image compression algorithm based on Rectangle Segmentation and storage with sparse matrix | |
| Poolakkachalil et al. | Comparative analysis of lossless compression techniques in efficient DCT-based image compression system based on Laplacian Transparent Composite Model and An Innovative Lossless Compression Method for Discrete-Color Images | |
| Kale et al. | Visually improved image compression by combining EZW encoding with texture modeling using Huffman Encoder | |
| Ngan et al. | Hybrid image coding scheme in incorporating human visual system characteristics | |
| Song et al. | Compressed image restoration via external-image assisted band adaptive pca model learning | |
| JP2017092801A (en) | Encoding method, encoding apparatus, and encoding program | |
| Bhatia et al. | A novel image enhancement technique based on statistical analysis of DCT coefficients for JPEG compressed images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Open date: 20100310 |