CN108093264B - Core image compression, decompressing method and system based on splits' positions perception - Google Patents
Core image compression, decompressing method and system based on splits' positions perception Download PDFInfo
- Publication number
- CN108093264B CN108093264B CN201711473464.4A CN201711473464A CN108093264B CN 108093264 B CN108093264 B CN 108093264B CN 201711473464 A CN201711473464 A CN 201711473464A CN 108093264 B CN108093264 B CN 108093264B
- Authority
- CN
- China
- Prior art keywords
- observation
- block
- image
- rate
- core 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.)
- Expired - Fee Related
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/63—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The present invention relates to a kind of core image compression, decompressing method and systems based on splits' positions perception, wherein, compression method includes: to calculate the wavelet transformation rate of sampling at different levels according to the target sampling rate, decomposed class and code rate of setting for core image to be compressed；Gaussian random matrix at different levels is obtained according to the piecemeal size of setting；Wavelet transform is carried out to core image to be compressed using discrete small wave converting method and obtains image block according to subband and piecemeal size by each subband piecemeal of every level-one；Each image block of every level-one is observed using gaussian random matrix, obtains observation block；Lowest frequency subband in observation block is encoded using DPCM method；Lifting wavelet transform is carried out to high-frequency sub-bands at different levels in observation block, then is encoded using set partitioning embedding space matrix method.The experimental results showed that the present invention under the conditions of high compression ratio, can be effectively retained the textural characteristics of core image, improve image reconstruction quality.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of core image compressions based on splits' positions perception
Method and system, and core image decompressing method and system based on splits' positions perception.
Background technique
Rock core is basic geological data important in oilfield prospecting developing, and observation description is determining lithology, inferring and sink
Play a significant role in product environment and source-reservoir-seal assemblage research.With being widely used for core image acquisition equipment, by rock core
Sample is stored in digital form by scanning mode, it has also become the important content of oil field digital Construction.Due to the accumulation of many years
With continuous new coring, cause core data amount extremely huge.Therefore, it studies and the compression algorithm of core image feature is suitble to be very
Significant.By being found to a large amount of typical core image analyses, since special geological environment and complicated geology transition are led
Core image is caused generally to have the characteristics that texture information is abundant, contrast is weaker.Traditional image compression algorithm based on small echo
Since the unusual limitation of the higher-dimensions such as wavelet representation edge, profile is extremely difficult to ideal compression effectiveness.By Donoho,
CS (Compressed Sensing, compressed sensing) theory that Candes and scientist Tao of Chinese origin et al. are proposed shows by asking
Optimization problem is solved, sparse signal can obtain the Exact recovery of high probability from a small amount of non-adaptive linear projection, this is
It further promotes compression of images performance and provides new technological means, cause the extensive attention of researcher.
Quite huge in view of the calculation amount for directlying adopt CS method reconstruct entire image, Gan proposes BCS (Block
Compressed Sensing, splits' positions perception) method, entire image is divided into the block of equidimension, independently to image block
It is observed and reconstructs, greatly reduce storage and calculate cost, but blocking artifact can be generated in low bit- rate.Mun et al. is mentioned
BCS-SPL (Smooth Projected Landweber, smoothly project Landweber) algorithm out, passes through gaussian random matrix
It realizes sampling, is realized and reconstructed using Wiener filtering combination Landweber iterative algorithm, improve blocking artifact, but reconstructed image is thin
Section becomes more fuzzy.So James et al. proposes MS-BCS-SPL (Multiscale on the basis of BCS-SPL algorithm
Block Compressed Sensing with Smoothed Projected Landweber, multiple dimensioned piecemeal variable sampling rate
Compressed sensing algorithm) algorithm, by transformation point in DWT (Discrete Wavelet Transform, wavelet transform) domain
The every level subbands obtained after solution carry out piecemeal, then calculate the sub- rate of sampling and are sampled, then pass through Wiener filtering and combine
Landweber iteration realizes reconstruct, further improves the reconstruction quality of image.
According to stringent foundations of information theory frame, mainly quantization and the cataloged procedure of data compression task are undertaken, therefore
Carrying out compression processing to observation using coding method appropriate can be further improved compression ratio.In the base of splits' positions perception
On plinth, Sungkwang et al. is DPCM (Differential Pulse-Code Modulation, Differential Pulse Code Modulation)
It is combined with SQ (Scalar Quantization, uniform scalar quantization) and compression of images perception observation is quantified, money journey etc.
People combines DPCM and non-uniform scalar quantification treatment compression of images perceives observation, but is similar to this kind of texture letter of rock core in processing
Effect is not ideal enough when the image that breath is abundant and contrast is weaker.
Summary of the invention
The technical problem to be solved by the present invention is to for existing method for compressing image, when handling core image, effect is not
Enough ideal defects provide a kind of core image compression, decompressing method and system based on splits' positions perception.
First aspect present invention provides a kind of core image compression method based on splits' positions perception, comprising:
Step 1 is directed to core image to be compressed, according to the target sampling rate S of settingt, decomposed class L and code rate BrIt calculates
The sub- rate S of sampling of the l grade of wavelet transforml, l ∈ [1, L]；
Step 2, the piecemeal size B according to settingl, pass through [SlBlBl+1/2]×BlBlSize obtain l grades Gauss with
Machine matrix Φl；
Step 3 carries out L grade transformation to the core image to be compressed using discrete small wave converting method, according to subband with
For piecemeal size by each subband piecemeal of every level-one, obtaining size is Bl×BlImage block xi；
Step 4, each image block x to every level-oneiUse gaussian random matrix ΦlIt is observed, obtains observation block
yi；
Step 5, to observation block yiMiddle lowest frequency subband observation is encoded using DPCM method；
Step 6, to observation block yiIn high-frequency sub-band observation at different levels carry out lifting wavelet transforms, then use set partitioning
Embedding space matrix method is encoded.
In the core image compression method according to the present invention based on splits' positions perception, it is preferable that the step
Rapid 1 the following steps are included:
Step 1-1, it is calculated by the following formula compressed sensing sample rate S:
Wherein N is the matrix dimension of core image to be compressed；
Step 1-2, compressed sensing decimation factor S is acquired by following formula transformationf,
Where it is assumed that the sub- rate S of the base band of wavelet transform0For fully sampled rate, then S0=1, and S1=1, to avoid generation
One or more SlThe case where 1 >, make it is l grades all in the case of Sl≤1；WlFor the weighting coefficient of l grades of subbands, Wl=16L-l+1；
Step 1-3, for l grades of subbands, according to its piecemeal size BlUse matrix ΦlIt is sampled, required sampling
Sub- rate SlIt can be calculated by the following formula:
Sl=WlSf。
In the core image compression method according to the present invention based on splits' positions perception, it is preferable that the step
It is compiled after also standardizing with tile cutting techniques to observation in rapid 6 using set partitioning embedding space matrix method
Code.
In the core image compression method according to the present invention based on splits' positions perception, it is preferable that the step
The step of standardizing with tile cutting techniques to observation described in rapid 6 include:
Step 6-1, observation block y is determinediSize be r × c, calculate d0=r × c；
Step 6-2, to d0Evolution is rounded, and obtains d；
Step 6-3, judge the value range of d:
(1) if d=2j, then by observation block yiDimensional transformations be d × d；
(2) if 2j< d < 2j+1, then observation block yiDimensional transformations be r0×c0, wherein r0=2j, c0=2j+1, and
Execute step 6-4；
Step 6-4, by r0As side length, tile format is r0×r0, judge r0With c0Whether the remainder that is divided by is zero, is to sentence
Without common portion between fixed tile to be cut, in observation block yiOn that set partitioning is respectively adopted is embedding after fragment cutting from left to right
Enter block coding method to be encoded；If r0With c0The remainder that is divided by is not zero, then determines there is common portion between tile to be cut, seeing
Measured value block yiOn from left to right fragment cut, record common portion position, be respectively adopted set partitioning embedding space matrix method into
Row coding, takes common portion to do mean value, then be integrated into coding result.
Second aspect of the present invention provides a kind of core image compressibility based on splits' positions perception, comprising:
Sub- rate computing module is sampled, for being directed to core image to be compressed, according to the target sampling rate S of settingt, decomposition level
Number L and code rate BrCalculate the sub- rate S of sampling of the l grade of wavelet transforml, l ∈ [1, L]；
Matrix computing module, for the piecemeal size B according to settingl, pass through [SlBlBl+1/2]×BlBlSize obtain l
The gaussian random matrix Φ of gradel；
Wavelet transformation module, for carrying out L grades of transformation to the core image to be compressed using discrete small wave converting method,
According to subband and piecemeal size by each subband piecemeal of every level-one, obtaining size is Bl×BlImage block xi；
Observation computing module, for each image block x to every level-oneiUse gaussian random matrix ΦlIt is observed,
Obtain observation block yi,
First coding module, for observation block yiMiddle lowest frequency subband observation uses Differential Pulse Code Modulation side
Method is encoded；
Second coding module, for observation block yiIn high-frequency sub-band observation at different levels carry out lifting wavelet transforms, then
It is encoded using set partitioning embedding space matrix method.
In the core image compressibility according to the present invention based on splits' positions perception, it is preferable that described to adopt
Appearance rate computing module includes:
First computing unit, for being calculated by the following formula compressed sensing sample rate S:
Wherein N is the matrix dimension of core image to be compressed；
Second computing unit, for acquiring compressed sensing decimation factor S by following formula transformationf,
Where it is assumed that the sub- rate S of the base band of wavelet transform0For fully sampled rate, then S0=1, and S1=1, to avoid generation
One or more SlThe case where 1 >, make it is l grades all in the case of Sl≤1；WlFor the weighting coefficient of l grades of subbands, Wl=16L-l+1；
Third computing unit is used for for l grades of subbands, according to its piecemeal size BlUse matrix ΦlIt is sampled, institute
The sub- rate S of the sampling neededlIt can be calculated by the following formula:
Sl=WlSf。
In the core image compressibility according to the present invention based on splits' positions perception, it is preferable that described the
Two coding modules include normalization unit, for using set partitioning after standardizing with tile cutting techniques to observation
Embedding space matrix method is encoded.
In the core image compressibility according to the present invention based on splits' positions perception, it is preferable that the rule
Generalized unit includes:
First processing subelement, for determining observation block yiSize be r × c, calculate d0=r × c；
Second processing subelement, for d0Evolution is rounded, and obtains d；
Third handles subelement, for judging the value range of d:
(1) if d=2j, then by observation block yiDimensional transformations be d × d；
(2) if 2j< d < 2j+1, then observation block yiDimensional transformations be r0×c0, wherein r0=2j, c0=2j+1, and
Start fourth process subelement；
Fourth process subelement, by r0As side length, tile format is r0×r0, judge r0With c0The remainder that is divided by whether be
Zero, it is without common portion between determining tile to be cut, in observation block yiOn from left to right fragment cutting after be respectively adopted
Set partitioning embedding space matrix method is encoded；If r0With c0The remainder that is divided by is not zero, then determines to have between tile to be cut public
Part, in observation block yiOn from left to right fragment cut, record common portion position, be respectively adopted set partitioning embedded block volume
Code method is encoded, and takes common portion to do mean value, then be integrated into coding result.
Third aspect present invention provides a kind of core image decompressing method based on splits' positions perception, comprising:
Step 1 carries out DPCM decoding to the lowest frequency subband observation of compression image；
Step 2 carries out SPECK decoding and Lifting Wavelet inverse transformation to the high-frequency sub-band observations at different levels of compression image；
The observation of step 3, each sub-block of at different levels high-frequency sub-bands of the integration through SPECK decoding specification, keeps its extensive
Restore the size of beginning core image；
Step 4 seeks approximate solution using Minimum Mean Squared Error estimation, to obtain the initial solution of reconstructed image
Step 5, to the initial solution of the reconstructed imageOptimize to obtain optimal solution.
Fourth aspect present invention provides a kind of core image decompression system based on splits' positions perception, comprising:
First decoder module carries out DPCM decoding for the lowest frequency subband observation to compression image；
Second decoder module carries out SPECK decoding for the high-frequency sub-band observations at different levels to compression image and is promoted small
Wave inverse transformation.
Size restoration module, the sight of each sub-block for integrating the high-frequency sub-bands at different levels through SPECK decoding specification
Measured value makes it restore the size of original core image；
Initial solution computing module, for seeking approximate solution using Minimum Mean Squared Error estimation, to obtain the first of reconstructed image
Begin solution
Optimization module, for the initial solution to the reconstructed imageOptimize to obtain optimal solution.
Implementing above-mentioned technical proposal of the invention has the advantages that the present invention using wavelet transform to core image
Rarefaction representation is carried out, multiple dimensioned piecemeal is carried out to each subband, subband not at the same level distributes different sample rates, with orthogonal gaussian random
Matrix is observed the image block of corresponding level, and lowest frequency subband observation is encoded using DPCM, high-frequency sub-band observation warp
Lifting wavelet transform carries out SPECK coding, realizes compression and reconstruct by Wiener filtering combination Landweber iteration；Experiment knot
Fruit shows under conditions of high compression ratio, uniform scalar quantization and directlys adopt uniform scalar quantization side in conjunction with using DPCM
Method is compared, and method proposed by the present invention effectively remains the textural characteristics of core image, improves the view of reconstruct core image
Feel effect and Y-PSNR.
Detailed description of the invention
Fig. 1 is the process according to the core image compression method based on splits' positions perception of the preferred embodiment of the present invention
Figure；
Fig. 2 a and Fig. 2 b are respectively the core image compression according to the preferred embodiment of the invention based on splits' positions perception
The two kinds of tile slit modes used in method；
Fig. 3 is the module frame chart of the core image compressibility according to the present invention based on splits' positions perception；
Fig. 4 a~4d gives the quality reconstruction pair that 3 width core images when code rate is 0.0841bpp apply 3 kinds of compression methods
Than figure；
Fig. 5 a~5d gives the quality reconstruction pair that 3 width core images when code rate is 0.0802bpp apply 3 kinds of compression methods
Than figure；
Fig. 6 a~6d gives the quality reconstruction pair that 3 width core images when code rate is 0.0919bpp apply 3 kinds of compression methods
Than figure；
Fig. 7 a~7c is respectively the property of core image 1, core image 2 and core image 3 using 3 kinds of methods compressions and reconstruct
It can comparison.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described.Obviously, described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
The characteristics of present invention is according to core image proposes the DPCM (Differential perceived based on splits' positions
Pulse-Code Modulation, Differential Pulse Code Modulation) and SPECK (Set Partitioned Embedded
Block, set partitioning embedding space matrix) Compression Strategies that combine of method.In DWT (Discrete Wavelet
Transform) in domain, the every level-one low frequency and high-frequency sub-band obtained after transformation is decomposed carries out piecemeal, then according to sample rate
Observation is obtained with observing matrix.DPCM coding is individually carried out to lowest frequency subband observation, high-frequency sub-band observation is carried out
Lifting wavelet transform (Lifting Wavelet Transform), carries out SPECK encoding and decoding again after concentrating energy further.
The compression and reconstruct of core image are realized finally by Wiener filtering combination Landweber iteration.
Referring to Fig. 1, for according to the core image compression method based on splits' positions perception of the preferred embodiment of the present invention
Flow chart.As shown in Figure 1, the core image compression method based on splits' positions perception that the embodiment provides includes following step
It is rapid:
Firstly, in step s101, for core image to be compressed, according to the target sampling rate S of settingt, decomposed class L
With code rate BrCalculate the sub- rate S of sampling of the l grade of wavelet transforml, wherein l ∈ [1, L].
Then, in step s 102, according to the piecemeal size B of settingl, calculate the sub- rate S of sampling at different levelsl, pass through [SlBlBl
+1/2]×BlBlObtain with its etc. sizes l grade gaussian random matrix Φl。
Then, in step s 103, L is carried out to the core image to be compressed using discrete small wave converting method (DWT)
Grade transformation, according to subband and piecemeal size by each subband piecemeal of every level-one, obtaining size is Bl×BlImage block xi。
Then, in step S104, to each image block x of every level-oneiUse gaussian random matrix ΦlIt is observed, obtains
To observation block yi, i.e. yi=Φlxi。
Then, in step s105, to observation block yiMiddle lowest frequency subband observation uses Differential Pulse Code Modulation
Method (DPCM) is encoded, and the lowest frequency subband observation of compression image is obtained
Then, in step s 106, to observation block yiIn high-frequency sub-band observation at different levels carry out lifting wavelet transforms, then
It is encoded using set partitioning embedding space matrix method (SPECK), obtains the high-frequency sub-band observations at different levels of compression imageThis method make full use of wavelet coefficient energy collection neutralize energy with scale increase and the characteristics of decay, by quaternary tree division and
Bit-plane coding method combines, compression performance with higher.Preferably, Lifting Wavelet change is being carried out in step S106
After changing, encoded again using SPECK after also standardizing with tile cutting techniques to observation.The present invention does not limit
The sequence of step S105 and step S106.
Preferably, above-mentioned steps S101 the following steps are included:
(1) it is calculated by the following formula compressed sensing sample rate S:
Wherein N is the matrix dimension of core image to be compressed；
(2) compressed sensing decimation factor S is acquired by following formula transformationf,
Where it is assumed that the sub- rate S of the base band of wavelet transform0For fully sampled rate, then S0=1, and S1=1, to avoid generation
One or more SlThe case where 1 >, make it is l grades all in the case of Sl> 1.WlFor the weighting coefficient of l grades of subbands, and:
Wl=16L-l+1； (3)
(3) for l grades of subbands, according to its piecemeal size BlUse matrix ΦlIt is sampled, the required sub- rate of sampling
SlIt can be calculated by the following formula:
Sl=WlSf。 (4)
Table 1 gives in different target sampling rate StUnder, the sub- rate S of sampling for the l grade that the DWT transformation of L=3 is realizedlSystem
Meter.
Table 1
The principle to the core image compression method of the invention based on splits' positions perception and process carry out detailed below
Explanation.
1, the multiple dimensioned splits' positions perception based on small echo
According to compressive sensing theory, it is assumed that x is the signal that the length obtained from M sampled signal y is N, and M < < N,
It is possible to restore signal x from formula (5).
Y=Φ x, (5)
Wherein, x ∈ RN, y ∈ RM, i.e. x is a N-dimensional vector, and y is a M dimensional vector (i.e. observation), and Φ is a tool
Having sample rate is M × N-dimensional observing matrix of S=M/N.
Computationally intensive to solve the problems, such as to directly adopt CS method reconstructed image, Gan proposes splits' positions cognitive method.It is false
If the image x that a width size is N × N is divided into the image block that size is B × B, the vector expression of i-th of image block is denoted as xi,
Use observing matrix ΦBIt is sampled, obtains observation:
yi=ΦBxi, (6)
Wherein, the size of B is according to the rate of image reconstruction and the comprehensive decision of the quality requirement of reconstruct: when B is lesser,
EMS memory occupation is few and calculating speed is fast；When B is larger, the quality reconstruction of image is relatively good；I=1 ... n, n=N2/B2；ΦB
Be size be MB×B2Orthogonal observing matrix, MB=(M × B2)/N2, M is the observed samples value number of entire image.
Since core image texture information is abundant, to retain more edges and detailed information, the present invention is proposed based on more
Scale splits' positions perceive observation code compression method, i.e., first carry out wavelet transform to core image before piecemeal, in this way
Observing matrix ΦBIt is divided into two parts: DWT multi-scale transform matrix Ω and multiple dimensioned piecemeal observing matrix Φl, i.e. yi=Φl
Ωxi.Assuming that Ω is L grades of DWT decomposition transform matrixes, then, ΦlIt is made of L+1 different observing matrixes.At this moment, image x
L grade be divided into size be Bl×BlImage block, subband pass through observing matrix ΦlIt is sampled, obtaining corresponding size isObservation block yi,Wherein, [1, L] l ∈.On this basis, to the core image of acquisition
Splits' positions perception observation encode and then realizes the compression of core image.
2, the compression of the core image observation based on the perception of multiple dimensioned splits' positions
2.1 compressed sensing observation standardization processings
Core image is after multiple dimensioned splits' positions perceive sampling observation, between lowest frequency subband observation adjacent block
Correlation is stronger, i.e. the data redundancy of lowest frequency subband observation is larger.And most energy of core image concentrate on
Therefore can lowest frequency subband carry out efficient coding to lowest frequency subband observation, will largely influence image coding
Quality.And correlation is weaker between high-frequency sub-band observation adjacent block, data redundancy is little, and the energy contained is also relatively
It is few, but include more detailed information.Therefore the present invention considers to encode lowest frequency subband observation using two dimension DPCM, and right
High-frequency sub-band observation is then encoded using the SPECK algorithm with preferable compression performance based on block structure.
The observation that usual core image is perceived through multiple dimensioned splits' positions is that irregular, traditional SPECK is calculated
Method can only handle the matrix of rule, carry out again after carrying out standardization processing present invention preferably employs tile cutting techniques thus
SPECK coding.The operation of tile models coupling piecemeal can effectively reduce data volume to be treated, make up the deficiency of wavelet transformation,
Better effects can be reached for operations such as compression of images.The basic step for observation of standardizing in step S106 is as follows:
Step 1: determining observation block yiSize be r × c, calculate d0=r × c；
Step 2: to d0Evolution is rounded, and obtains d；
Step 3: judge the value range of d:
(1) if d=2j, then by observation block yiDimensional transformations be d × d；It can carry out SPECK encoding operation；
(2) if 2j< d < 2j+1, then observation block yiDimensional transformations be r0×c0, wherein r0=2j, c0=2j+1, j ∈
Z+；
Step 4: if the case where meeting above-mentioned steps 3 (2), by r0As side length, tile format is r0×r0:
(1) if r0With c0The remainder that is divided by is zero, then can be determined that between tile to be cut without common portion, in observation
SPECK coding is done respectively after fragment cutting from left to right.Sliced fashion is as shown in Figure 2 a.
(2) if r0With c0The remainder that is divided by is not zero, then can be determined that between tile to be cut there is common portion, in observation
Fragment is cut from left to right, records common portion position, is carried out SPECK coding respectively, is taken common portion to do mean value, then be integrated into
SPECK coding resultSliced fashion is as shown in Figure 2 b.
2.2 core image compressed sensing observation DPCM+SPECK coding
Using the graded characteristics of wavelet conversion coefficient, multiple dimensioned piecemeal is carried out to each subband of core image, and keeping mesh
In the case that mark sample rate is constant, different fractions match different sample rates, calculate corresponding observing matrix.With observing matrix to phase
It answers the image block of level to be observed, obtains observation.DPCM coding is carried out to lowest frequency subband observation, high-frequency sub-band is seen
Measured value carries out tile dividing processing first and carries out lifting wavelet transform, is then encoded using SPECK.Finally by Wiener filtering
The compression and reconstruct to core image are realized in conjunction with Landweber iterative operation.Therefore, core image observation pressure of the invention
Contracting method and step such as abovementioned steps S101~S106.
Referring to Fig. 3, for the module frame chart of the core image compressibility according to the present invention based on splits' positions perception.
As shown in figure 3, the core image compressibility based on splits' positions perception that the embodiment provides includes sampling sub- rate to calculate mould
Block 301, matrix computing module 302, wavelet transformation module 303, observation computing module 304, the first coding module 305 and second
Coding module 306.
Wherein, sub- rate computing module 301 is sampled for for core image to be compressed, according to the target sampling rate of setting
St, decomposed class L and code rate BrCalculate the sub- rate S of sampling of the l grade of wavelet transforml, wherein l ∈ [1, L].
Matrix computing module 302 is used for the piecemeal size B according to settingl, pass through [SlBlBl+1/2]×BlBlObtain and its
Etc. the l grade of sizes gaussian random matrix Φl。
Wavelet transformation module 303 is used to carry out L grades of changes to the core image to be compressed using discrete small wave converting method
It changes, according to subband and piecemeal size by each subband piecemeal of every level-one, obtaining size is Bl×BlImage block xi。
Observation computing module 304 is used for each image block x to every level-oneiUse gaussian random matrix ΦlIt is seen
It surveys, obtains observation block yi；
First coding module 305 is connected with the observation computing module 304, for observation block yiMiddle lowest frequency
Band observation is encoded using DPCM method.
Second coding module 306 is connected with the observation computing module 304, for observation block yiIn high frequencies at different levels
Subband observation carries out lifting wavelet transform, then is encoded using set partitioning embedding space matrix method.
Preferably, sampling sub- rate computing module 301 includes:
First computing unit, for being calculated by the following formula compressed sensing sample rate S:
Wherein N is the matrix dimension of core image to be compressed；
Second computing unit acquires compressed sensing decimation factor S by following formula transformationf,
Where it is assumed that the sub- rate S of the base band of wavelet transform0For fully sampled rate, then S0=1, and S1=1, to avoid generation
One or more SlThe case where 1 >, make it is l grades all in the case of Sl< 1；WlFor the weighting coefficient of l grades of subbands, Wl=16L-l+1；
Third computing unit, for in l grades of subbands, according to its piecemeal size BlUse matrix ΦlIt is sampled,
The required sub- rate S of samplinglIt can be calculated by the following formula:
Sl=WlSf。
Preferably, the second coding module 306 includes normalization unit, for carrying out with tile cutting techniques to observation
It is encoded again using set partitioning embedding space matrix method after standardization.
Preferably, the normalization unit includes:
First processing subelement, for determining observation block yiSize be r × c, calculate d0=r × c；
Second processing subelement, for d0Evolution is rounded, and obtains d；
Third handles subelement, for judging the value range of d:
(1) if d=2j, then by observation block yiDimensional transformations be d × d；
(2) if 2j< d < 2j+1, then observation block yiDimensional transformations be r0×c0, wherein r0=2j, c0=2j+1, and
Start fourth process subelement；
Fourth process subelement, when third processing subelement judges that the value range of d plants situation for (2), being used for will
r0As side length, tile format is r0×r0, judge r0With c0Whether the remainder that is divided by is zero, is to determine no common portion,
Observation block yiOn set partitioning embedding space matrix method be respectively adopted encoded after fragment cutting from left to right；If r0With c0
The remainder that is divided by is not zero, then is determined with common portion, in observation block yiOn from left to right fragment cut, record common portion position
It sets, set partitioning embedding space matrix method is respectively adopted and is encoded, takes common portion to do mean value, then be integrated into coding result
The present invention also provides a kind of core image decompressing methods based on splits' positions perception, and decompression procedure is compressed
The inverse process of journey is decoded operation to observation first, and restores full size size, then in conjunction with Wiener filtering and
Landweber iteration realizes core image reconstruct, the specific steps are as follows:
Step 1, to compression image lowest frequency subband observationCarry out DPCM decoding；
Step 2, to compression image high-frequency sub-band observations at different levelsCarry out SPECK decoding (the inverse mistake of SPECK coding
Journey) and Lifting Wavelet inverse transformation；
The observation of step 3, each sub-block of at different levels high-frequency sub-bands of the integration through SPECK decoding specification, keeps its extensive
The size of beginning core image is restored, i.e.,Size；
Step 4 solves x using MMSE (Minimum Mean Square Error, Minimum Mean Squared Error estimation)iIt is close
Like solution, to obtain the initial solution of reconstructed image
Step 5, to the initial solution of the reconstructed imageOptimize to obtain optimal solution.
Preferably, optimize that detailed process is as follows in the step 5:
(1) blocking artifact brought by the adaptive wiener filter removal image block with 3 × 3 neighborhoods；
(2) it by filtered image projection in convex set, since observing matrix is orthogonal matrix, can be obtained by following formula:
Wherein,For the initial solution of the reconstructed image, ΦBFor orthogonal observing matrix, yiFor observation；
(3) projection result is carried out wavelet transformation, projection result is filtered using bivariate shrinkage function contraction in wavelet field
Wave；
(4) inverse wavelet transform is carried out to filter result, it will be on filtered image reprojection to convex set；
(5) judge and terminate iteration, until obtaining optimal solution.
The present invention further correspondingly provides a kind of core image decompression system based on splits' positions perception, comprising:
First decoder module, for the lowest frequency subband observation to compression imageCarry out DPCM decoding；
Second decoder module, for the high-frequency sub-band observations at different levels to compression imageCarry out SPECK decoding (SPECK
The inverse process of coding) and Lifting Wavelet inverse transformation；
Size restoration module, the sight of each sub-block for integrating the high-frequency sub-bands at different levels through SPECK decoding specification
Measured value makes it restore the size of original core image；
Initial solution computing module, for seeking approximate solution using Minimum Mean Squared Error estimation, to obtain the first of reconstructed image
The solution that begins xi；
Optimization module, for the initial solution x to the reconstructed imageiOptimize to obtain optimal solution.Wherein optimization module
Optimization method is identical as step 5 in aforementioned decompressing method, and details are not described herein.
The present invention is by following comparative experiments to core image compression, decompressing method and the system perceived based on splits' positions
It is verified.
Wherein, the core image compression method of the invention based on splits' positions perception is as previously mentioned, i.e. at multiple dimensioned point
It combines DPCM and SPECK algorithm to carry out encoding and decoding to core image observation on the basis of block compressed sensing algorithm, realizes rock core
The compression and reconstruct of image.Control methods 1 is using referring in background technique DPCM (Differential Pulse-Code
Modulation, Differential Pulse Code Modulation) and the side that combines SQ (Scalar Quantization, uniform scalar quantization)
Method is labeled as MS-BCS-SPL-DPCM+SQ；Control methods 2 directlys adopt uniform scalar quantization (SQ) using without prediction
Method is labeled as MS-BCS-SPL-SQ.
3 512 × 512 more representational core images are tested respectively using above-mentioned 3 kinds of methods in experiment,
Selected wavelet transform filter is 9/7 integer wavelet, and transformation series is 3, and observing matrix uses orthogonal gaussian random matrix, sparse
Base sets wavelet transform (Duel-tree Discrete Wavelet Transform, DDWT) using 3 grades pairs, every level-one
Image block size is respectively 8,16,32.Under same quantization method under identical reconstruction model, the height of image reconstruction quality
It is low to depend on sample rate and quantizing bit number, sample rate and quantizing bit number are reasonably set, different samplings are tested out
The Y-PSNR of core image is reconstructed under rate and quantizing bit number.
2 core image of table applies the PSNR (dB) of 3 kinds of method reconstructed images
Note: S in table 2tFor target sampling rate.
Table 2 gives the PSNR value that 3 width core images apply 3 kinds of algorithms to realize compression with reconstruct respectively.As can be seen that
In the case where high compression ratio, mentioned using the compression of the method for the present invention core image and the PSNR value ratio of reconstruct using other 2 kinds of methods
High 0.3~1.0dB.Fig. 4 a~4d gives the weight that 3 width core images when code rate is 0.0841bpp apply 3 kinds of compression methods
Structure effect contrast figure.Wherein Fig. 4 a is original core image 1；It is that MS-BCS-SPL-SQ is obtained that Fig. 4 b, which is using control methods 2,
Reconstructed image, PSNR=26.64；Fig. 4 c is the reconstructed image obtained using control methods 1 i.e. MS-BCS-SPL-DPCM+SQ,
PSNR=26.65；Fig. 4 d is the reconstructed image obtained using the method for the present invention, PSNR=27.16.Fig. 5 a~5d gives code rate
3 width core images apply the quality reconstruction comparison diagram of 3 kinds of compression methods when for 0.0802bpp.Wherein Fig. 5 a is original rock core
Image 2；Fig. 5 b is the reconstructed image obtained using control methods 2 i.e. MS-BCS-SPL-SQ, PSNR=23.25；Fig. 5 c is to use
Control methods 1 is the reconstructed image that MS-BCS-SPL-DPCM+SQ is obtained, PSNR=23.25；Fig. 4 d is to use the method for the present invention
Obtained reconstructed image, PSNR=23.86.Fig. 6 a~6d gives 3 width core images when code rate is 0.0919bpp and applies 3 kinds
The quality reconstruction comparison diagram of compression method.Wherein Fig. 6 a is original core image 3；It is MS- that Fig. 6 b, which is using control methods 2,
The reconstructed image that BCS-SPL-SQ is obtained, PSNR=20.03；It is MS-BCS-SPL-DPCM+SQ that Fig. 6 c, which is using control methods 1,
Obtained reconstructed image, PSNR=20.03；Fig. 6 d is the reconstructed image obtained using the method for the present invention, PSNR=20.38.Its
For middle core image 3 compared with 2 texture-rich of core image, and for core image 1,2 texture of core image is again relatively abundant.It can
To find out, better vision is shown when handling the image of texture information relative abundance using method proposed by the present invention and is imitated
Fruit remains more texture informations.
Fig. 7 a~7c is respectively the property of core image 1, core image 2 and core image 3 using 3 kinds of methods compressions and reconstruct
Can comparison, can more intuitively find out PSNR value of 3 kinds of methods under different compression ratios variation tendency and side of the present invention
Promotion degree of the method compared to the PSNR value of other 2 kinds of methods.
In conclusion the characteristics of herein according to core image, on the basis of multiple dimensioned splits' positions perceive framework, sufficiently
The characteristics of using SPECK and DPCM encryption algorithm, carry out a large amount of emulation experiments and analysis, realize to core image observation into
The compression of one step and reconstruct, obtained core image visual effect and Y-PSNR are promoted, and are tied with using SQ and DPCM
It closes SQ to compare, PSNR value averagely improves 0.6dB and 0.5dB respectively, and in the case where high compression ratio, has been effectively retained rock core
The textural characteristics of image.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (6)
1. a kind of core image compression method based on splits' positions perception characterized by comprising
Step 1 is directed to core image to be compressed, according to the target sampling rate S of settingt, decomposed class L and code rate BrIt calculates discrete
The sub- rate S of sampling of the l grade of wavelet transformationl, wherein l ∈ [1, L]；
Step 2, the piecemeal size B according to settingl, pass through [SlBlBl+1/2]×BlBlSize obtain l grades of gaussian random square
Battle array Φl；
Step 3 carries out L grades of transformation to the core image to be compressed using discrete small wave converting method, according to subband and piecemeal
For size by each subband piecemeal of every level-one, obtaining size is Bl×BlImage block xi；
Step 4, each image block x to every level-oneiUse gaussian random matrix ΦlIt is observed, obtains observation block yi；
Step 5, to observation block yiMiddle lowest frequency subband observation is encoded using DPCM method；
Step 6, to observation block yiIn high-frequency sub-band observation at different levels carry out lifting wavelet transforms, then be embedded in using set partitioning
Block coding method is encoded；
The step 1 the following steps are included:
Step 1-1, it is calculated by the following formula compressed sensing sample rate S:
Wherein N is the matrix dimension of core image to be compressed；
Step 1-2, compressed sensing decimation factor S is acquired by following formula transformationf,
Where it is assumed that the sub- rate S of the base band of wavelet transform0For fully sampled rate, then S0=1, and S1=1, to avoid generating one
Or multiple SlThe case where 1 >, make it is l grades all in the case of Sl≤1；WlFor the weighting coefficient of l grades of subbands, Wl=16L-l+1；
Step 1-3, for l grades of subbands, according to its piecemeal size BlUse matrix ΦlIt is sampled, the required sub- rate of sampling
SlIt is calculated by the following formula:
Sl=WlSf。
2. the core image compression method according to claim 1 based on splits' positions perception, which is characterized in that the step
It is compiled after also standardizing with tile cutting techniques to observation in rapid 6 using set partitioning embedding space matrix method
Code.
3. the core image compression method according to claim 2 based on splits' positions perception, which is characterized in that the step
The step of standardizing with tile cutting techniques to observation described in rapid 6 include:
Step 6-1, observation block y is determinediSize be r × c, calculate d0=r × c；
Step 6-2, to d0Evolution is rounded, and obtains d；
Step 6-3, judge the value range of d:
(1) if d=2j, then by observation block yiDimensional transformations be d × d；
(2) if 2j< d < 2j+1, then observation block yiDimensional transformations be r0×c0, wherein r0=2j, c0=2j+1, j is positive whole
Number, and execute step 6-4；
Step 6-4, by r0As side length, tile format is r0×r0, judge r0With c0Whether the remainder that is divided by is zero, be then determine to
Without common portion between cutting tile, in observation block yiOn from left to right fragment cutting after set partitioning embedded block is respectively adopted
Coding method is encoded；If r0With c0The remainder that is divided by is not zero, then determines there is common portion between tile to be cut, in observation
Block yiOn from left to right fragment cut, record common portion position, set partitioning embedding space matrix method is respectively adopted and is compiled
Code, takes common portion to do mean value, then be integrated into coding result.
4. a kind of core image compressibility based on splits' positions perception characterized by comprising
Sub- rate computing module is sampled, for being directed to core image to be compressed, according to the target sampling rate S of settingt, decomposed class L and
Code rate BrCalculate the sub- rate S of sampling of the l grade of wavelet transforml, l ∈ [1, L]；
Matrix computing module, for the piecemeal size B according to settingl, pass through [SlBlBl+1/2]×BlBlSize obtain l grades
Gaussian random matrix Φl；
Wavelet transformation module, for carrying out L grades of transformation to the core image to be compressed using discrete small wave converting method, according to
By each subband piecemeal of every level-one, obtaining size is B for subband and piecemeal sizel×BlImage block xi；
Observation computing module, for each image block x to every level-oneiUse gaussian random matrix ΦlIt is observed, obtains
Observation block yi；
First coding module, for observation block yiMiddle lowest frequency subband observation using DPCM method into
Row coding；
Second coding module, for observation block yiIn high-frequency sub-band observation at different levels carry out lifting wavelet transforms, then using collection
Division embedding space matrix method is closed to be encoded；
The sub- rate computing module of the sampling includes:
First computing unit, for being calculated by the following formula compressed sensing sample rate S:
Wherein N is the matrix dimension of core image to be compressed；
Second computing unit, for acquiring compressed sensing decimation factor S by following formula transformationf,
Where it is assumed that the sub- rate S of the base band of wavelet transform0For fully sampled rate, then S0=1, and S1=1, to avoid generating one
Or multiple SlThe case where 1 >, make it is l grades all in the case of Sl≤1；WlFor the weighting coefficient of l grades of subbands, Wl=16L-l+1；
Third computing unit is used for for l grades of subbands, according to its piecemeal size BlUse matrix ΦlIt is sampled, required for
The sub- rate S of samplinglIt is calculated by the following formula:
Sl=WlSf。
5. the core image compressibility according to claim 4 based on splits' positions perception, which is characterized in that described the
Two coding modules include normalization unit, for using set partitioning after standardizing with tile cutting techniques to observation
Embedding space matrix method is encoded.
6. the core image compressibility according to claim 5 based on splits' positions perception, which is characterized in that the rule
Generalized unit includes:
First processing subelement, for determining observation block yiSize be r × c, calculate d0=r × c；
Second processing subelement, for d0Evolution is rounded, and obtains d；
Third handles subelement, for judging the value range of d:
(1) if d=2j, then by observation block yiDimensional transformations be d × d；
(2) if 2j< d < 2j+1, then observation block yiDimensional transformations be r0×c0, wherein r0=2j, c0=2j+1, j is positive whole
Number, and start fourth process subelement；
Fourth process subelement, by r0As side length, tile format is r0×r0, judge r0With c0Whether the remainder that is divided by is zero, is
No common portion is then determined, in observation block yiOn from left to right fragment cutting after set partitioning embedding space matrix is respectively adopted
Method is encoded；If r0With c0The remainder that is divided by is not zero, then is determined with common portion, in observation block yiOn divide from left to right
Piece cutting, records common portion position, set partitioning embedding space matrix method is respectively adopted and is encoded, common portion is taken to do
Value, then it is integrated into coding result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711473464.4A CN108093264B (en) | 2017-12-29 | 2017-12-29 | Core image compression, decompressing method and system based on splits' positions perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711473464.4A CN108093264B (en) | 2017-12-29 | 2017-12-29 | Core image compression, decompressing method and system based on splits' positions perception |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108093264A CN108093264A (en) | 2018-05-29 |
| CN108093264B true CN108093264B (en) | 2019-03-08 |
Family
ID=62181212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711473464.4A Expired - Fee Related CN108093264B (en) | 2017-12-29 | 2017-12-29 | Core image compression, decompressing method and system based on splits' positions perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108093264B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109559357B (en) * | 2018-09-17 | 2020-01-24 | 衡阳师范学院 | Wavelet packet threshold-based image block compressed sensing reconstruction method |
| CN110175965B (en) * | 2019-05-30 | 2020-12-18 | 齐齐哈尔大学 | Block compressed sensing method based on self-adaptive sampling and smooth projection |
| CN113382255B (en) * | 2019-06-21 | 2022-05-20 | 杭州海康威视数字技术股份有限公司 | Encoding and decoding method, device and storage medium |
| CN111193925B (en) * | 2019-12-25 | 2021-12-07 | 杭州中威电子股份有限公司 | Image compressed sensing coding and normalization method based on block vector inner product |
| CN111882489A (en) * | 2020-05-15 | 2020-11-03 | 东北石油大学 | Super-resolution graph recovery method for simultaneously enhancing underwater images |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056638A (en) * | 2016-05-18 | 2016-10-26 | 信阳师范学院 | Low-complexity compressive sensing image coding method adapted to texture contrast |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103037212B (en) * | 2011-10-08 | 2016-02-10 | 太原科技大学 | The adaptive block compressed sensing method for encoding images of view-based access control model perception |
| CN103886050B (en) * | 2014-03-12 | 2017-05-24 | 佛山科学技术学院 | Image feature storing method, image searching method and device based on compressive sensing |
-
2017
- 2017-12-29 CN CN201711473464.4A patent/CN108093264B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056638A (en) * | 2016-05-18 | 2016-10-26 | 信阳师范学院 | Low-complexity compressive sensing image coding method adapted to texture contrast |
Non-Patent Citations (1)
| Title |
|---|
| 基于压缩感知的岩心图像压缩与重构技术研究;刘彦彤;《东北石油大学硕士研究生学位论文》;20161231;第27-43页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108093264A (en) | 2018-05-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108093264B (en) | Core image compression, decompressing method and system based on splits' positions perception | |
| Kumar et al. | A review: DWT-DCT technique and arithmetic-Huffman coding based image compression | |
| CN110248190A (en) | A kind of compressed sensing based multilayer residual error coefficient image encoding method | |
| Elamaran et al. | Comparison of DCT and wavelets in image coding | |
| Singh et al. | Review of image compression techniques | |
| Ghazel | Adaptive fractal and wavelet image denoising | |
| Kumari et al. | Image quality estimation by entropy and redundancy calculation for various wavelet families | |
| Mohanty et al. | A high performance modified SPIHT for scalable image compression | |
| Guz | A novel image compression method based on classified energy and pattern building blocks | |
| Annadurai et al. | Wavelet based color image compression using vector quantization and morphology | |
| Hong et al. | An edge-preserving subband coding model based on non-adaptive and adaptive regularization | |
| Pandian et al. | Effects of decomposition levels of wavelets in image compression algorithms | |
| Banjanin et al. | Entropy-threshold method for best basis selection | |
| Prashanth et al. | Fractal image compression for hd images with noise using wavelet transforms | |
| Kumari | A wavelet based approach for compression of color images | |
| Tyagi et al. | Enhanced EZW Technique for Compression of Image by Setting Detail Retaining Pass Number | |
| SVEC et al. | Image compression using transform coding methods | |
| Reddy et al. | Compression of Hyperspectral Images by Using DWT and SPIHT | |
| Yonghui et al. | A static image coding algorithm based on contourlet Classified Hidden Markov Tree model | |
| Vrindavanam et al. | Wavelet and JPEG based image compression: an experimental analysis | |
| Mohsenian et al. | Edge-based subband image coding technique for encoding the upper-frequency bands | |
| KR20010077752A (en) | Image compressing method and device by using the discrete wavelet transform applied for fuzzy logics considering the human vision system | |
| Babu et al. | Image compression and reconstruction using improved Stockwell transform for quality enhancement. | |
| Tang et al. | Visual Weighted SPECK Image Coding Based on WBCT | |
| Dongmei et al. | Complexity scalable audio coding algorithm based on wavelet packet decomposition |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190308 Termination date: 20191229 |