CN105072446B - The compression sampling and restructing algorithm of a kind of coloured image - Google Patents
The compression sampling and restructing algorithm of a kind of coloured image Download PDFInfo
- Publication number
- CN105072446B CN105072446B CN201510442038.9A CN201510442038A CN105072446B CN 105072446 B CN105072446 B CN 105072446B CN 201510442038 A CN201510442038 A CN 201510442038A CN 105072446 B CN105072446 B CN 105072446B
- Authority
- CN
- China
- Prior art keywords
- parts
- passages
- sample rate
- channel
- coloured image
- 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.)
- Active
Links
Landscapes
- Compression Of Band Width Or Redundancy In Fax (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention relates to the compression sampling and restructing algorithm of a kind of coloured image, coloured image is decomposed into tri- passages of R, G, B and carries out wavelet transformation respectively by coding side, the wavelet conversion coefficient of different layers is measured using different sample rates, and the sparse reconstruct of group is carried out using the correlation of three passages in decoding end.During present invention processing image, the data volume of required sampling is few, and the picture quality of recovery is good.The traditional algorithm independently reconstructed relative to RGB, in low sampling rate, Y-PSNR about improves 1 2dB.
Description
Technical field
The present invention relates to the compression sampling and restructing algorithm of a kind of coloured image.
Background technology
Compressed sensing is a kind of new information processing theory, it breaches nyquist sampling theorem in traditional sampling
Limitation so that signal can be to be sampled far below nyquist sampling rate, and reconstructs original with remaining to high probability in decoding end
Beginning signal.
Current many scholars have studied application of the compressive sensing theory in gray level image, and its in coloured image should
With more rare.Most scholars when compressed sensing is applied to coloured image, using by coloured image transform to RGB or
On YUV color spaces, so as to be broken down into three autonomous channels, recycle the compressive sensing theory of gray level image logical to three
Road is compressed and reconstructs respectively.But carry out compression reconfiguration coloured image in this way, can not be using between three passages
High correlation, cause sampling waste.
Nagesh proposes EJSM algorithms, this is to utilize the phase between three passages during color image compression perceives for the first time
Closing property reconstructs, and the method that article proposes the wavelet coefficient joint sparse of three passages extracts the common portion between them,
The degree of rarefication of coloured image is reduced, so as to obtain more preferable quality reconstruction.Majumdar propose using the sparse principle of group come
Coloured image is reconstructed, is effectively improved the reconstruction quality of coloured image, but calculation matrix is too big, is well in simulations
Computer running memory, may be only available for the coloured image of small size.
The content of the invention
In view of this, it is an object of the invention to provide the compression sampling and restructing algorithm of a kind of coloured image, further
The sampling efficiency of coloured image is improved, improves reconstruction quality, the natural image compressed sensing for low sampling rate condition.
To achieve the above object, the present invention adopts the following technical scheme that:A kind of compression sampling of coloured image and reconstruct are calculated
Method, it is characterised in that:
Programming process includes the following steps:
Step A1:One coloured image is decomposed into R passages, G passages and channel B;
Step A2:The R passages, G passages and channel B are subjected to three layers of wavelet decomposition respectively and obtain corresponding wavelet transformation
Coefficient, wherein the R channel decompositions are R1 parts, R2 parts, R3 parts and R4 parts;The G channel decompositions are G1 parts, G2
Partly, G3 parts and G4 parts;The channel B is decomposed into B1 parts, B2 parts, B3 parts and B4 parts;
Step A3:According to the openness distribution sample rate of the wavelet conversion coefficient, the R passages, G passages and B are carried out
The block sampling of passage;
Decoding process comprises the following steps:
Step B1:Group1 is merged into R1 parts, G1 parts, B1 parts, by R2 parts, G2 parts, B2
Part merge into group2, by the R3 part, G3 part, B3 part merge into group3, by the R4 part, G4 part,
B4 merges into group4 in part;
Step B2:Carry out the sparse reconstruct of group respectively to described group1, group2, group3 and group4;
Step B3:Three are isolated in group1, group2, group3 and group4 after the sparse reconstruct of group is carried out to lead to
Four parts in road, and resequence, the wavelet coefficient of the R passages, G passages and the channel B that are reconstructed;
Step B4:Carry out the picture content that wavelet inverse transformation is reconstructed respectively to the R passages, G passages and channel B;
Step B5:The picture content combination of the reconstruct is recovered into the coloured image.
Further, sample rate is allocated as follows in the step A3:Assuming that the coloured image is N × N coloured images,
Its sample rate is ratio, and R1 parts account for the 1/64 of total pixel in R passages, make sample rate ratio1=0.9, the R passage of R1 parts
Middle R2 parts account for the 3/64 of total pixel, make R3 parts in sample rate ratio2=0.8, the R passage of R2 parts account for the 3/ of total pixel
16, R4 parts account for the 3/4 of total pixel, and the sample rate for making R3 parts is 2 times of R4 fractional-sample rates, i.e., ratio3=2 ×
Ratio4, to ensure that the whole-sample rate of image is constant, meets following formula:
As available from the above equation:,
The sample rate method of salary distribution as R passages is performed to the G passages and channel B.
The present invention has the advantages that compared with prior art:During present invention processing coloured image, required sampling
Data volume is few, and the picture quality of recovery is good, the natural image compressed sensing of low sampling rate condition is particularly suitable for, compared to RGB
The traditional algorithm independently reconstructed, in low sampling rate, Y-PSNR about improves 1-2dB.
Brief description of the drawings
Fig. 1 is overview flow chart of the present invention.
Fig. 2 is three layers of wavelet decomposition schematic diagram of one embodiment of the invention R passages.
Fig. 3 is the group1 definition procedure figures of one embodiment of the invention.
Fig. 4 is the group2 definition figures of one embodiment of the invention.
Fig. 5 is the group3 definition figures of one embodiment of the invention.
Fig. 6 is the group4 definition figures of one embodiment of the invention.
Fig. 7 is SL20 restructing algorithms flow chart of the present invention.
Embodiment
The present invention will be further described with reference to the accompanying drawings and embodiments.
Fig. 1 is refer to, the present invention provides a kind of compression sampling and restructing algorithm of coloured image, it is characterised in that:
Programming process includes the following steps:
Step A1:One coloured image is decomposed into R passages, G passages and channel B;
Step A2:The R passages, G passages and channel B are subjected to three layers of wavelet decomposition respectively and obtain corresponding wavelet transformation
Coefficient, the wherein low frequency of each layer of R passages are distributed as shown in Fig. 2, other passages are also similar with high fdrequency component;The R channel decompositions
For R1 parts, R2 parts, R3 parts and R4 parts;The G channel decompositions are G1 parts, G2 parts, G3 parts and G4 parts;Institute
State channel B and be decomposed into B1 parts, B2 parts, B3 parts and B4 parts;
Step A3:According to the openness distribution sample rate of the wavelet conversion coefficient, the R passages, G passages and B are carried out
The block sampling of passage, the detailed process for distributing sample rate are as follows:Assuming that the coloured image is N × N coloured images, its sample rate
For ratio, R1 parts account for the 1/64 of total pixel in R passages, make R2 parts in sample rate ratio1=0.9, the R passage of R1 parts
The 3/64 of total pixel is accounted for, makes the part of R3 in sample rate ratio2=0.8, the R passage of R2 parts account for 3/16, the R4 parts of total pixel
The 3/4 of total pixel is accounted for, the sample rate for making R3 parts is 2 times, i.e. ratio3=2 × ratio4 of R4 fractional-sample rates, to ensure to scheme
The whole-sample rate of picture is constant, meets following formula:
As available from the above equation:,
The sample rate method of salary distribution as R passages is performed to the G passages and channel B.
Decoding process comprises the following steps:
Step B1:Group1 is merged into R1 parts, G1 parts, B1 parts, by R2 parts, G2 parts, B2
Part merge into group2, by the R3 part, G3 part, B3 part merge into group3, by the R4 part, G4 part,
B4 merges into group4 in part;
Step B2:Carry out the sparse reconstruct of group respectively to described group1, group2, group3 and group4;
The sparse principle of group is that the same position of three passages is assigned to be identically numbered, and utilizes the correlation between them
Property, found in solution and number identical position, make them in this position with certain contact.Using size as 16 × 16 cromograms
As exemplified by, if piecemeal size is 2 × 2, then image can be divided into 64 blocks.4 × 1 row for being by each image block scan are sweared
Amount, then can show that the definition procedure of group1 is as shown in Figure 3.
Fig. 3 is refer to, first extracts the wavelet coefficient first row in three passages respectively(Preceding 1/64 part is R1, G1, B1),
Stacked again in the form of column vector, form 12 × 1 matrixes from R11 to B14, R11, R12, R13, R14 are R1 in wherein Fig. 3
4 elements, G11, G12, G13, G14 are 4 elements of G1, and B11, B12, B13, B14 are 4 elements of B1, group1 be by
The identical coordinate definition of relative position into identical numeral, such as R11, G11, B11 in 12 × 1 matrixes that three passages are built up
1 is all defined as, R12, G12, B12 are defined as 2.When organizing sparse reconstruct, to reconstruct first pixel, need to only find
Group1 matrixes intermediate value is 1 coordinate(1;5;9), i.e., R, G, B triple channel composition reconstructed image middle position be set to(1;5;9)Picture
Element has correlation, their value can be made similar, and here it is the principle for organizing sparse reconstruct.Similarly, 16 × 16 coloured images
The definition of group2, group3 and group4 are respectively as shown in Fig. 4, Fig. 5, Fig. 6.
Then, the data of group1, group2, group3 and group4 are reconstructed respectively using SL20 algorithms.Restructuring procedure
That is solving-optimizing problem, wherein A is sensing matrix, and y is measured value, and x is wavelet coefficient,Attach most importance to
The wavelet coefficient of structure, restructing algorithm flow are as shown in Figure 7.
Step B3:Three are isolated in group1, group2, group3 and group4 after the sparse reconstruct of group is carried out to lead to
Four parts in road, and resequence, the wavelet coefficient of the R passages, G passages and the channel B that are reconstructed;Wherein group1
R1 parts, G1 parts and B1 parts are isolated, wherein group2 isolates R2 parts, G2 parts and B2 parts, wherein group3
R3 parts, G3 parts and B3 parts are isolated, wherein group4 isolates R4 parts, G4 parts and B4 parts(Refer to Fig. 1);
Step B4:Carry out the picture content that wavelet inverse transformation is reconstructed respectively to the R passages, G passages and channel B;
Step B5:The picture content combination of the reconstruct is recovered into the coloured image.
In order to verify the effect of the method for the present invention, by taking 512 × 512 coloured image Peppers and Lena as an example, it is assumed that point
Block size is 32 × 32, and calculation matrix is bernoulli matrix, to the color image compression sampling of the present invention and restructing algorithm and biography
The RGB independence reconstructing methods of system compare, with Y-PSNR PSNR(Unit is dB)Reconstruct evaluation as algorithm refers to
Mark.Table 1 gives PSNR of the color image compression reconstructing method under low sampling rate.
Table 1:
As it can be seen from table 1 for coloured image Peppers and Lena, in the case of equal sample rate to image into
Row sampling, method PSNR proposed by the present invention are higher about 1-2dB than RGB independence reconstructing methods.It can thus be appreciated that coloured silk proposed by the present invention
The compression sampling and restructing algorithm of color image in the case of identical compression ratio, can effectively improve the measurement efficiency of coloured image
And reconstruction quality.
The foregoing is merely presently preferred embodiments of the present invention, all equivalent changes done according to scope of the present invention patent with
Modification, should all belong to the covering scope of the present invention.
Claims (2)
1. the compression sampling and restructing algorithm of a kind of coloured image, it is characterised in that:
Programming process includes the following steps:
Step A1:One coloured image is decomposed into R passages, G passages and channel B;
Step A2:The R passages, G passages and channel B are subjected to three layers of wavelet decomposition respectively and obtain corresponding wavelet transformation system
Number, wherein the R channel decompositions are R1 parts from low to high, R2 parts, R3 parts and R4 parts;The G passages point
Solve as G1 parts from low to high, G2 parts, G3 parts and G4 parts;The channel B is decomposed into from low to high
B1 parts, B2 parts, B3 parts and B4 parts;
Step A3:According to the openness distribution sample rate of the wavelet conversion coefficient, higher sample rate is set for low frequency part,
Relatively low sample rate is set for high frequency section, carries out the block sampling of the R passages, G passages and channel B;
Decoding process comprises the following steps:
Step B1:Group1 is merged into R1 parts, G1 parts, B1 parts, by R2 parts, G2 parts, B2 parts
Group2 is merged into, group3 is merged into R3 parts, G3 parts, B3 parts, by R4 parts, G4 parts, B4 portions
Divide and merge into group4;
Step B2:Carry out the sparse reconstruct of group respectively to described group1, group2, group3 and group4;
Step B3:Three passages are isolated in group1, group2, group3 and group4 after the sparse reconstruct of group is carried out
Four parts, and resequence, the wavelet coefficient of the R passages, G passages and the channel B that are reconstructed;
Step B4:Carry out the picture content that wavelet inverse transformation is reconstructed respectively to the R passages, G passages and channel B;
Step B5:The picture content combination of the reconstruct is recovered into the coloured image.
2. the compression sampling and restructing algorithm of coloured image according to claim 1, it is characterised in that:In the step A3
Sample rate is allocated as follows:Assuming that the coloured image is N × N coloured images that length and width are N number of pixel, its sample rate is
R1 parts account for the 1/64 of total pixel in ratio, R passage, make R2 parts in sample rate ratio1=0.9, the R passage of R1 parts account for
The 3/64 of total pixel, makes 3/16, the R4 parts that R3 parts in sample rate ratio2=0.8, the R passage of R2 parts account for total pixel account for
The 3/4 of total pixel, the sample rate for making R3 parts is 2 times, i.e. ratio3=2 × ratio4 of R4 fractional-sample rates, to ensure image
Whole-sample rate it is constant, meet following formula:
As available from the above equation:,
The sample rate method of salary distribution as R passages is performed to the G passages and channel B.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510442038.9A CN105072446B (en) | 2015-07-27 | 2015-07-27 | The compression sampling and restructing algorithm of a kind of coloured image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510442038.9A CN105072446B (en) | 2015-07-27 | 2015-07-27 | The compression sampling and restructing algorithm of a kind of coloured image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105072446A CN105072446A (en) | 2015-11-18 |
CN105072446B true CN105072446B (en) | 2018-05-04 |
Family
ID=54501712
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510442038.9A Active CN105072446B (en) | 2015-07-27 | 2015-07-27 | The compression sampling and restructing algorithm of a kind of coloured image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105072446B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106485760B (en) * | 2016-09-30 | 2019-05-14 | 电子科技大学 | A kind of color image Downsapling method based on minimum interpolation error quadratic sum |
US10666937B2 (en) * | 2016-12-21 | 2020-05-26 | Qualcomm Incorporated | Low-complexity sign prediction for video coding |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103346798A (en) * | 2013-06-05 | 2013-10-09 | 中国科学院微电子研究所 | Method for acquiring signal at sampling frequency lower than Nyquist frequency |
CN103841583A (en) * | 2014-01-16 | 2014-06-04 | 华南理工大学 | Wireless network optimized mass signaling data collecting method based on compressed sensing |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101365444B1 (en) * | 2007-11-19 | 2014-02-21 | 삼성전자주식회사 | Method and apparatus for encoding/decoding moving image efficiently through adjusting a resolution of image |
-
2015
- 2015-07-27 CN CN201510442038.9A patent/CN105072446B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103346798A (en) * | 2013-06-05 | 2013-10-09 | 中国科学院微电子研究所 | Method for acquiring signal at sampling frequency lower than Nyquist frequency |
CN103841583A (en) * | 2014-01-16 | 2014-06-04 | 华南理工大学 | Wireless network optimized mass signaling data collecting method based on compressed sensing |
Also Published As
Publication number | Publication date |
---|---|
CN105072446A (en) | 2015-11-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10834415B2 (en) | Devices for compression/decompression, system, chip, and electronic device | |
JP7431742B2 (en) | Method and apparatus for encoding/decoding a point cloud representing a three-dimensional object | |
CN102630011A (en) | Compressive perceptual coding and decoding method and system in video sensor network | |
CN111179144B (en) | Efficient information hiding method for multi-embedding of multi-system secret information | |
RU2767771C1 (en) | Method and equipment for encoding/decoding point cloud representing three-dimensional object | |
CN105357536B (en) | The soft method of multicasting of video based on residual distribution formula compressed sensing | |
CN102970540A (en) | Multi-view video code rate control method based on key frame code rate-quantitative model | |
CN103281537B (en) | A kind of dynamic range of images compression method and device | |
CN104704825A (en) | Lossless compression of fragmented image data | |
CN105072446B (en) | The compression sampling and restructing algorithm of a kind of coloured image | |
CN106941609A (en) | The video-frequency compression method perceived based on self adaptation splits' positions | |
CN105554502A (en) | Distributed compressed sensing video encoding and decoding method based on foreground-background separation | |
CN103096079A (en) | Multi-view video rate control method based on exactly perceptible distortion | |
CN106056638A (en) | Low-complexity compressive sensing image coding method adapted to texture contrast | |
CN109671010A (en) | It is a kind of based on multiple two-dimensional histograms modification reversible information hide and extracting method | |
CN101827268B (en) | Object-based fractal video compression and decompression method | |
Shi et al. | Multi-scale deep networks for image compressed sensing | |
CN103414903A (en) | Compressing method and device for Bayer format images | |
CN105163122B (en) | A kind of compression of images and decompression method based on image block similarity | |
CN104683818B (en) | Method for compressing image based on biorthogonal invariant set m ultiwavelet | |
CN103763566A (en) | Color halftone image compressing method based on three-dimensional matrix WDCT transformation | |
CN113112557A (en) | Block self-adaptive carton image compression method based on dictionary learning | |
CN117152019A (en) | Low-illumination image enhancement method and system based on double-branch feature processing | |
CN110324638A (en) | Intraframe prediction algorithm based on multiple dimensioned neural network in hybrid video coding standard | |
CN116980611A (en) | Image compression method, apparatus, device, computer program product, and medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |