CN103065336B - Directional pyramid coding method of image - Google Patents
Directional pyramid coding method of image Download PDFInfo
- Publication number
- CN103065336B CN103065336B CN201310024432.1A CN201310024432A CN103065336B CN 103065336 B CN103065336 B CN 103065336B CN 201310024432 A CN201310024432 A CN 201310024432A CN 103065336 B CN103065336 B CN 103065336B
- Authority
- CN
- China
- Prior art keywords
- pyramid
- image
- tree
- export
- coefficient
- 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
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 47
- 230000015556 catabolic process Effects 0.000 claims description 5
- 230000000694 effects Effects 0.000 abstract description 11
- 238000000638 solvent extraction Methods 0.000 abstract description 3
- 230000004438 eyesight Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 14
- 230000009466 transformation Effects 0.000 description 12
- 238000000354 decomposition reaction Methods 0.000 description 11
- 238000001914 filtration Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 9
- 238000005070 sampling Methods 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 8
- 238000007906 compression Methods 0.000 description 7
- 230000006835 compression Effects 0.000 description 7
- 238000012360 testing method Methods 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000004907 flux Effects 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 235000007926 Craterellus fallax Nutrition 0.000 description 1
- 240000007175 Datura inoxia Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 210000000481 breast Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000015271 coagulation Effects 0.000 description 1
- 238000005345 coagulation Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000007634 remodeling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a directional pyramid coding method of an image. By changing a data structure of a traditional set partitioning in hierarchical tree (SPIHT) algorithm, a directional pyramid of coding of the algorithm is possible to realize. Because a directional pyramid SPIHT algorithm is characterized by being high in redundancy and practical, the directional pyramid SPIHT algorithm is put forward and is preferable, and therefore rebuilding effects of the whole, especially rebuilding effects of image edges and other features are better improved. The algorithm can effectively code and compress images applied to the field of machine visions.
Description
Technical field
The present invention relates to compression of images field, specifically, relate to a kind of direction pyramid compression of image and the method for coding.
Background technology
1. signal conversion and image pyramid
Signal is the carrier of information.The effective expression how realizing signal is a key problem in signal transacting field.From the viewpoint of mathematics, image is the two-dimensional matrix of a gray-scale value.In piece image, usually it is seen that the texture the be connected region similar to gray level, they combine formation object.If dimension of object is very large or contrast very strong (entirety of image), usually only need to adopt lower resolution (Fig. 1 (a), (b)); If the small-sized or contrast of object not high (details of image), then need to adopt higher resolution view (Fig. 1 (c), (d)).If dimension of object varies, or contrast has and has by force weak situation to exist simultaneously, and carrying out research with some resolution to them will have advantage.Certainly the glamour place of this namely multi-resolution hierarchy.
When the sampling rate of signal meets Nyquist requirement, normalization frequency band must be limited between [-π, π].Now with perfect low pass and ideal highpass filter L (ω) and H (ω), it can be resolved into (for positive frequency part) frequency band [0 respectively, pi/2] low frequency part and frequency band at [pi/2, π] HFS, the general picture of difference reflected signal and details, as shown in Figure 2.Because not overlapping between frequency band, after process, two-way output signal is orthogonal.And due to two kinds export bandwidth all reduce by half, therefore sampling rate can reduce by half and the unlikely loss causing information.Here it is can introduce after the filtering " two extract " reason (in Fig. 2 ↓ 2 symbols represent " two extract " operation), so-called two to extract be exactly exported once every one by list entries, forms the new sequence of contraction in length half.Similar process can repeat down the low frequency part after each decomposition, that is: each fraction stem-butts cutting off this grade of input signal resolves into the rough approximation (general picture) of a low frequency and the detail section of a high frequency.And the output sampling rate of every grade can reduce by half again, so just original signal x (n) is carried out Multiresolution Decomposition.
The one carrying out interpretation of images with multiresolution effectively but the simple structure of concept is exactly image pyramid.Image pyramid is at first for machine vision and compression of images, and the pyramid of piece image is exactly a series of image collections progressively reduced with the resolution of pyramid configuration.As shown in Figure 3, pyramidal bottom is that the high resolving power of pending image represents, top is the approximate of low resolution.When moving to pyramidal upper strata, size and resolution reduce.One deck (i.e. (J-1) layer) above of pyramid J layer is of a size of 1/4 of J layer.Usual pyramidal low-resolution image is for analyzing the overall content of large structure or image, and high-definition picture is for analyzing the characteristic of single body.Like this by coarse to meticulous analysis strategy particularly suitable in pattern-recognition.
Pyramid structure be signal multiple dimensioned/important method of multiresolution analysis.Gaussian pyramid (Guassian Pyramid) and laplacian pyramid (Laplacian Pyramid) structure, be mainly used in the Coding Compression Algorithm of digital picture.Gaussian pyramid is obtained by the continuous down sample of original image, is equivalent to constantly low-pass filtering at frequency domain.Different layers gaussian pyramid launches and subtracts each other under same yardstick, is equivalent to high-pass filtering, can generates laplacian pyramid like this in the amount frequency domain subtracted each other.Remove from input signal in principle low pass level and smooth after signal just can obtain the detail signal of this yardstick (i.e. this layer), but due to the sampling rate of smooth signal lower than input signal, two signals can not directly subtract each other.In order to obtain detail signal, needing to carry out interpolation to the output signal of previous step, interpolation M-1 pixel, and carrying out filtering, the image of the resolution such as generation and input signal, then subtract each other with input signal, obtain detail signal.Owing to having carried out interpolation arithmetic to smooth signal, interpolation filter has just determined the similarity degree between predicted value and input value.In order to reconstruct input signal, as long as smooth signal is obtained approximate signal through same low-pass filter process, then being added with detail signal, just can realizing the Perfect Reconstruction of original image.
2. wavelet transformation
The two-dimensional orthogonal wavelets conversion of image is the another kind of important image technique relevant to multiresolution analysis, is also a kind of important behaviour form of image pyramid, and it can obtain image in level, detailed information vertically and in diagonal.The range of application of wavelet transformation is comparatively wide, mainly comprises the aspects such as the compressed encoding of image, Postprocessing technique, image denoising, characteristics extraction.The multiresolution analysis that separable two-dimensional orthogonal wavelets conversion is formed is a most widely used class two-dimensional wavelet transformation.
From the angle of digital filter, Fig. 4 gives the process that 2-d wavelet coefficient decomposes.Wherein L and H represents one dimension low pass and Hi-pass filter respectively, and subscript x and y represents respectively and carries out filtering with column direction in the row direction to matrix.The feature of separable situation exactly can along x and y both direction successively two steps deal with (first filtering is carried out to x direction and filtering is being carried out to y direction), owing to carrying out bandpass filtering treatment, the bandwidth of bandwidth ratio original image is little, subband can carry out the sampling without information loss, and therefore every coagulation all will through twice two extractions.Wherein LL is through the low pass of x and y both direction, the low-frequency component of corresponding original discrete picture on next yardstick (namely descending one deck); High pass in the low pass of LH on x direction, y direction, the corresponding low-frequency component of original image horizontal direction and the radio-frequency component of vertical direction; Accordingly, what HL represented is the radio-frequency component in x direction and the low-frequency component in y direction; The radio-frequency component of x and y both direction that what HH represented is.Fig. 5 gives the frequency spectrum designation of two-dimensional wavelet transformation, initial input matrix is regarded as a two-dimensional discrete image, four parts obtained after once decomposing export respectively through different wave filters, represent general picture (i.e. low frequency) information of original image and the detailed information in vertical, level, diagonal line three directions.After a wavelet transformation, total output data quantity is identical with input data volume, just different according to frequency information, is classified by each component, is convenient to signal transacting.If the general picture part after once decomposing is proceeded wavelet decomposition, just can obtain the multiscale analysis of discrete picture, i.e. its details on different scale and general picture.
Equally, through the process of the single subband of interpolation, filtering and superposition, add the detailed information on different scale, the general picture composition on any thin yardstick can be reconstructed, until final Perfect Reconstruction goes out original image.
3 direction pyramids
Discrete Orthogonal Wavelet Transform is because of its good time frequency analysis characteristic, become method for expressing conventional in multiple dimensioned signal and graphical analysis, but the shortcoming that it exists is owing to causing lacking translation invariance to translation parameters uniform sampling, when less change in displacement occurs input signal, wavelet sub-band coefficient energy can have greatly changed.Simultaneously for rotation and the dimensional variation of two dimensional image, wavelet conversion coefficient energy is also unstable.Another shortcoming that wavelet transformation exists is exactly that its set direction is limited, can only be broken down into level at each metric space, vertically and direction, three, diagonal angle, be difficult to meet the requirement of image to continuous direction.
Direction pyramid can be counted as can the laplacian pyramid of choice direction, by Eero Simoncelli in invention in 1993.It is a kind of multiple dimensioned multidirectional image representational framework linearly, is applied to the field such as image procossing and computer vision.Picture breakdown can be become different scale, multidirectional a series of subband by it, not only can keep translation and rotational invariance, and direction is controlled, thus provides the directional information of more horn of plenty relative to wavelet analysis.
Fig. 6 gives the system chart of one deck direction pyramid level discharge rating of image, and it is all that polarity can be divided that direction pyramid decomposes the subband obtained.Wherein H
0(ω) be Hi-pass filter, L
0(ω), L
1(ω) be the low-pass filter of different scale, B
i(ω), i=0 ..., K is different directions bandpass filter; ↓ 2 represent the process of down-sampling and up-sampling respectively with ↑ 2; X (ω) and
be respectively original image and reconstructed image; Empty circles represents the nested of this system.As shown in Figure 6, first this algorithm is broken down into high pass and low pass two subbands, and signal passes through complementary high-pass and low-pass filter and do not carry out lower sample calculation.Low frequency sub-band is decomposed into one group of band further and is led to the image of subband and the sub-band images of (more) low pass subsequently.Through bandpass filter B
i(ω) band that process obtains leads to subband and has different directivity.Using these as basis function subband, the directional subband of any direction just can be obtained by the linear combination adjusting their frequency response.The logical subband of band does not carry out lower sample calculation.In practical application, can design the basis function directional subband of number K, the direction of these basis functions is respectively
i=0 ..., K-1, such as, if K=4, then 4 directions of basis function are respectively 0 °, 45 °, 90 °, 135 °.Decompose again after (more) low pass subband after decomposition carries out down-sampling, repeat above process, realize Multiresolution Decomposition algorithm.Equally, through the process of each subband of interpolation, filtering and superposition, the detailed information on different scale can be recovered, original image can be gone out by Perfect Reconstruction.
Fig. 7 gives the frequency spectrum designation that direction pyramid three yardsticks (i.e. three layers) four direction decomposes.Wherein frequency axis is [-π, π].In a frequency domain, each subband that direction pyramid decomposes is all that polarity can be divided.First step pre-service is decomposed, and high-frequency sub-band corresponds to four angles of spatial frequency domain, and low frequency sub-band corresponds to great circle region; The decomposable process of second step, the basis function band in one of them direction corresponding, shadow region leads to subband; Final low frequency sub-band after the roundlet region in bosom corresponds to and decomposes on the 3rd yardstick.
Fig. 8 shows for an input (e), represents (a)-(d) by the direction pyramid calculating this image, and reconstructs the process of (f) by (a)-(d).Figure (a)-(c) is the subband that the direction pyramid in three yardstick four directions (0 °, 45 °, 90 °, 135 °) decomposes, and (d) does not show for small throughput, high frequency content.
The main deficiency of direction pyramid is its mistake completeness, and its complete degree of mistake is
wherein K is the number of directional subband.This is equivalent to, if the image of input one secondary 6x6, pyramidal direction number is 4, does not limit the yardstick number (i.e. the number of plies) of decomposition, and the direction pyramid of output containing 192 pixels, will far exceed 36 pixels of original input image.Although cross counting yield and storage cost that completeness can limit pyramid algorith, provide conveniently also to many image processing methods.
4 based on the method for encoding images of embedded wavelet
Due to the multiresolution character of wavelet transformation, the coefficient that image obtains after wavelet transformation has good distribution character in spatial domain and frequency domain, and therefore the various Image Compression based on wavelet transformation achieves very ten-strike.At present more effective wavelet transformation compression method has two kinds: a kind of be 1993 by J.M.Shapiro according to the similarity of the wavelet transformation of a sub-picture between not at the same level, propose embedded zerotree wavelet coding method (EZW:Embedded Zerotree Wavelet); Another kind be A.Said in 1996 according to the basic thought of EZW algorithm, propose a kind of new implementation method, i.e. SPIHT algorithm (SPIHT:Set Partitioning in Hierarchical Trees).
EZW coding thinking is the hypothesis based on still there being very strong correlation between wavelet coefficient not at the same level, and this correlativity shows with father and son's Relationship of Coefficients of wavelet tree.In fact, if the wavelet coefficient being positioned at lower frequency layer is less than a certain threshold value, be then positioned at the possibility that the equidirectional and wavelet coefficient that is locus of upper frequency layer be less than this threshold value very big.If a coefficient on lower frequency layer is less than certain threshold value, and have some coefficients to be greater than this threshold value in Progeny Lines Derived manifold on next upper frequency layer and higher frequency layer corresponding to this coefficient, just this coefficient on lower frequency layer is defined as zero tree, then threshold value reduces by half, again to image scanning, go down and so forth, constantly generate zero tree.In whole successive approximation to quantification process, by the present threshold value that constantly reduces by half, multiple scanning and symbolic coding, until meet target bit rate needs.
EZW is a kind of Embedded Zerotree Wavelet scheme be structured on wavelet transform base.Bit stream to be encoded can sort by its scrambler according to difference of importance, terminates coding at any time according to target bit rate or degree of distortion size requirements; Equally, decoding is terminated at any time for given stream decoder, and the reconstruction image of the target bit rate of corresponding code stream truncated position can be obtained.
The characteristic distributions of wavelet coefficient is more larger toward low frequency sub-band coefficient value, comprises quantity of information more, more less toward high-frequency sub-band coefficient value, comprises quantity of information fewer.EZW algorithm takes full advantage of coefficient characteristic distributions after wavelet decomposition, first passes the significant bits of the coefficient compared with low frequency, then transmits the significant bits of higher-frequency coefficient.
First EZW is defined Zero tree structure: the image coefficient C after an optional wavelet transformation
i, j, for given thresholding T, if had | C
i, j|>=T, then claim wavelet coefficient C
i, jbe significant coefficient, otherwise be defined as inessential coefficient.If wavelet coefficient is inessential about given thresholding T on a thick yardstick, all wavelet coefficients in same locus corresponding with this coefficient on thinner yardstick are also inessential about thresholding T, then claim these wavelet coefficients to define zero tree, and specify that the wavelet coefficient on thick yardstick is father or zerotree root.Wavelet coefficient on thinner yardstick on relevant position is descendants.
SPIHT (Set Partitioning in Hierarchical Trees) algorithm also realizes Wavelet Image Compression by building zero tree, utilizes the theory of progressive transmission to encode.Progressive transmission theory is by descending for the absolute value of numerical value arrangement, is then first transmitted by most important numerical value, and during reduction, the Quality of recovery of image will improve gradually.The output of SPIHT is series of bits stream, and SPIHT can stop at any point transmitting and not affect the correctness of decoding.Criterion and the EZW of the coefficient importance of SPIHT use are similar.It has threshold value T
l, the maximal value of representing matrix element, but its adopt be position coding, the higher bit position of important element of first encoding, its coding parameter n
l=[log
2t
l.
SPIHT uses the structural generation Sum decomposition coefficient sets being called spatial orientation tree, this structure make use of the spatial relationship of wavelet coefficient between image pyramid different layers: experience shows, the subband of the every one deck of pyramid all presents spatial simlanty, any special feature, as linear edge or homogeneous area, can see at the same position of all layers.
The data structure (i.e. spatial orientation tree) that SPIHT uses as shown in Figure 9.In figure, LL1 and HL2 mark represents the LL subband of ground floor and the HL subband of the second layer respectively, and other marks by that analogy.Draw in figure except the second layer (high pass) and ground floor (low pass) two-layer, every one deck is divided into again four subbands (LL, LH, HL, HH), subband LL1 is divided into again four coefficient sets, and wherein shadow region part is that group being positioned at the upper left corner.Often each (except the positive square region at the use solid black round dot place in the upper left corner) organized in 4 coefficients becomes the tree root of spatial orientation tree.How different layers that arrow illustrates these numbers are associated together, arrow points be the child of this subband.In image, the position of the child of the coefficient of position (i, j) is (2i, 2j), (2i+1,2j), (2i, 2j+1), (2i+1,2j+1).Because the data near the upper left corner represent important low frequency amount, so should priority encoding.
Summary of the invention
The object of the invention is to, give transmission for the redundancy that the direction pyramid data structure of image is high and store the difficulty brought, studying a kind of method and carry out efficient coding storage.
The performing step of the pyramidal coding method of image direction of the present invention is as follows:
The first step, becomes the form of the direction pyramid of any given direction number and yardstick number by picture breakdown, image pyramid according to direction number to yardstick number stratified storage;
Second step, can a direction-based function of any given direction pyramid decomposed in the first step, coding that in the third step can be preferential to the subband corresponding to this direction;
3rd step, adopts spiht algorithm to carry out the pyramidal low frequency of image direction and band frequency component is encoded, but adopts different threshold values to low frequency with band frequency amount.
The invention has the beneficial effects as follows:
1., by changing the traditional data structure of traditional spiht algorithm, make this algorithm coding direction pyramid become possibility.
2. for the data structure that this redundance of direction pyramid is high, SPIHT is optimized, makes the coding of directional derivative and overall reconstruction effect have larger lifting.
3. can select a priority encoding direction, reduce the number of bits expressing redundant information, thus reach better data compression effects.
Accompanying drawing explanation
Fig. 1 is natural image and localized variation histogram thereof;
Fig. 2 is that the ideal of frequency band divides;
Fig. 3 is the structure of image pyramid;
Fig. 4 is one-level 2-d wavelet decomposition process figure;
Fig. 5 is that multilevel wavelet decomposes spectrogram;
Fig. 6 is the direction pyramid system chart having K direction;
Fig. 7 is one 3 layers, the spectrogram of the ideal orientation decomposition in 4 directions;
Fig. 8 is pyramid decomposition and the reconstruct example of image; Original image is (e), represents (a)-(d), by (a)-(d) reconstructed image (f) by the direction pyramid calculating this image.
Fig. 9 is SPIHT data structure schematic diagram;
Figure 10 is the pyramidal data structure schematic diagram of codified image direction of the present invention;
Figure 11 is the schematic diagram of the data structure with priority encoding direction of the present invention;
Figure 12 is the original image (a) that the present invention tests spiht algorithm employing, the image (b) of reconstruct, and encode and quantize rear direction pyramid (c), first row is the band flux of ground floor, and second row is the band frequency component of the second layer;
Figure 13 is the original image (a) that the present invention tests, SPIHT quality reconstruction (b), use direction pyramid spiht algorithm quality reconstruction (c), use privileged direction pyramid spiht algorithm quality reconstruction (d), wherein the left side is the image of reconstruct, the right side is the direction pyramid of reconstruct, and first row is the band flux of ground floor, and second row is the band flux of the second layer;
Figure 14 is the effectiveness comparison of algorithms of different reconstructed image;
Figure 15 is the effectiveness comparison of use three kinds of different pieces of information structural remodeling images, overall quality reconstruction (a) and to different layers, the reconstruct (b) of the directional derivative of different directions.
Embodiment
In order to the content and advantage that make technical solution of the present invention clearly understand, below in conjunction with accompanying drawing, be described in further detail for the pyramidal coded system of image direction and method of the present invention.
Coded system for direction pyramid picture breakdown of the present invention and method, the pyramidal coded system of a kind of image direction based on SPIHT and method, mainly for the blank of SPIHT at image direction pyramid coded portion, and the defect crossing redundant representation aspect of process image, based on image direction pyramidal data structure feature, a kind of new data structure and a kind of pyramidal coded system of image direction formed in conjunction with this data structure are proposed.
As can be seen from the data structure diagram (Fig. 9) of SPIHT, the tree of shadow region (i.e. top LL subband) can only along element 2, three directions of 3,4 correspondences are stretched, thus cannot meet the requirement of the direction pyramid to coding any number.Need to modify to this data structure.
First this algorithm removes contributes less HFS to integral image effect and feature extraction, only coding low frequency and band frequency part.Figure 10 is the pyramidal data structure schematic diagram of codified image direction of the present invention, this data structure breaches the data structure that different layers coefficient is all placed on the zones of different of a positive square by traditional SPIHT coding, is placed into respectively in the positive square varied in size by the coefficient in different layers, different directions.And conveniently represent, define the low frequency component (in corresponding diagram 7, circle L is inner) that " ground floor " is image pyramid data structure in Fig. 10; " second layer " is the minimum band frequency component of image pyramid frequency, and in corresponding diagram 7, the annulus at B1 place is inner; In " third layer " corresponding diagram 7, the annulus at B2 place is inner, by that analogy.To every one deck, each party to coefficient encode one by one.Low frequency amount is positioned at ground floor, the important information containing image, needs first to encode.And then the second layer of encoding, third layer, by that analogy.Lower one deck that every one deck is corresponding is all children of this layer, and child's coefficient is stored in another positive square.Such as, in ground floor (low frequency part), upper left coefficient child in the two directions represents with the positive square that second layer mid point is filled.In shown child, the child (namely the grandson of low frequency left upper part coefficient) of lower right coefficients represents with the positive square that third layer center line is filled.In practical application, can design the basis function directional subband of number K, the direction of these basis functions is respectively
i=0 ..., K-1, such as, if K=4, then 4 directions of basis function are respectively 0 °, 45 °, 90 °, 135 °.This data structure can be encoded the direction band frequency component of any amount, therefore can meet coding staff to pyramidal needs.
Figure 11 is the schematic diagram of the data structure with priority encoding direction of the present invention.This data structure changes the data relation of original band frequency component, first select the pyramidal second layer of conduct of the directional subband B (being direction 1 in this example) of a priority encoding, then be with the higher level of frequency component and B as third layer remaining, by that analogy.Experiment represents, selecting privileged direction in advance, can saving the bit for representing inessential information before coding.
In addition, the value of the direction band of the image decomposed for direction pyramid part frequently can a lot of feature less of low frequency component, and the threshold value of being decomposed by low frequency is separated with being with the threshold value of frequency division solution, uses two threshold value T
l, T
hrepresent respectively, T
l, T
hrepresent the maximal value that low frequency amount and band are measured frequently respectively.Thus determine two coding parameters
with
Finally, this algorithm generates a bit bit stream (i.e. a series of bit) and represents this image, if the bit stream received is interrupted in arbitrfary point, can decode error-free and reconstructed image.
The ordered list that this algorithm is introduced is identical with spiht algorithm with aggregated label.First select the pyramidal second layer of conduct of the directional subband B of a priority encoding during coding, be then with the higher level of frequency component and B as third layer remaining.
The image direction pyramid encryption algorithm that we claim the data structure using Figure 10 is direction pyramid spiht algorithm, is privileged direction pyramid spiht algorithm the image direction pyramid encryption algorithm of the data structure using Figure 11.
Because direction pyramid spiht algorithm and privileged direction pyramid spiht algorithm only have the difference of data structure, the false code of specific implementation is all identical.Their false code is as follows:
(1) initialization
Input initial two threshold value T
l, T
h, represent the maximal value that low frequency amount and band are measured frequently respectively.Thus determine two coding parameter n
land n
h;
Initial coordinate integrates as LIP={ (r, c) | (r, c) ∈ H}, LIS={D
r, c| (r, c) ∈ H}.
(2) sequence scanning
1) LIP queue is scanned
Importance is judged to each list item (r, c) of LIP queue;
If (r, c) is important
The sign bit of ' r ' and (r, c) is exported to Sn;
(r, c) is deleted from LIP queue, adds the afterbody of LSP queue to.
If (r, c) is inessential
' 0 ' is exported to Sn.
2) LIS queue is scanned
' if D ' type list item, i.e. D
r, c, judge importance;
If D
r, cimportant
' 1 ' is exported to Sn;
To each (rO, cO) ∈ O
r, r, judge importance
If important, export the sign bit of ' 1 ' and (rO, cO) to Sn, and by O
r, cadd to LIS afterbody;
If inessential, then (rO, cO) is added to the afterbody of LIP.
Judge L
r, cwhether be empty set
If not empty, then by L
r, cadd the afterbody of LIS to;
If empty set, then by D
r, cdelete from LIS.
If D
r, cinessential
' 0 ' is exported to Sn.
' if L ' type list item, i.e. L
r, c, judge importance;
If L
r, cimportant
To Sn output ' 1 ' and by D
rO, cOadd the afterbody of LIS to, by L
r, cdelete from LIS;
If L
r, cinessential
' 0 ' is exported to Sn.
(3) fine scanning
The absolute value of coefficient in LSP is converted to binary representation and exports the n-th most important position to Rn.
(4) threshold exponent is upgraded
By threshold exponent n
lreduce to n
l-1, n
hreduce to n
h-1, turn back to step (2) and carry out next stage coded scanning.
Actual effect:
Because direction pyramid was complete, for preventing misreading, we as enforcement and evaluation index, but limit the bit number of its bit stream without compressibility.First we use standard spiht algorithm to utilize 10
7bit is encoded to test pattern Girl (256 × 256) Figure 12 (a) image (Figure 12 (b)) and the corresponding direction pyramid (Figure 12 (c)) of the reconstruct obtained.
Bit after coding is assigned as: RnList:25930 bit; SnList:975672 bit.Wherein total bit number is than 10
7more be because the separator of bit stream causes.The recovery effects can seeing image is pretty good, but the bit expended is too many.
Test pattern Lena (64 × 64) (Figure 13 (a)) of using, T as calculated
l=816, T
h=105.To three kinds of algorithms: spiht algorithm, direction pyramid spiht algorithm and privileged direction pyramid spiht algorithm are tested respectively, utilize different algorithms all to use 5 × 10
4bit is encoded, and the reconstructed image obtained and direction pyramid are as shown in Figure 13 (b)-(d), and wherein the privileged direction of algorithm II is 0 °.Because 0 ° is the direction of priority encoding, so compared with other directions, use algorithm II has had larger lifting to the reconstruction effect of 0 ° than using algorithm I, and the reconstruction effect of integral image there has also been and improves significantly.
(overhead bit is from 5 × 10 to use different overhead bits to compare three kinds of algorithms respectively to Lena (64 × 64)
3bit is until 8 × 10
4bit), use the preferential coding staff of algorithm II to being 0 °.Concrete PSNR curve is shown in Figure 14.Can see, algorithm I, the relative SPIHT of algorithm II is for the reconstruction effect of direction pyramid, and the lifting especially for the reconstruction effect of directional derivative is significant.In addition, because 0 ° is privileged direction, algorithm II is preferentially encoded, so algorithm II will apparently higher than other data structures of test to the quality reconstruction in 0 ° of direction when can find low bit rate.
Claims (2)
1. the pyramidal coding method of image direction, is characterized in that, comprise the following steps:
A. picture breakdown is become the pyramidal form of any given direction number and yardstick number, to the basis function directional subband of number K, the direction of these basis functions is respectively
B. image pyramid is according to direction number and yardstick number stratified storage;
C. from the K steps A basis function direction, establish the basis function direction of a priority encoding;
D. adopt privileged direction pyramid spiht algorithm, different threshold values is adopted to low frequency and band frequency component;
E. utilize bit stream to encode, generate low frequency amount and band amount frequently that a bit bit stream represents this image;
Described privileged direction pyramid spiht algorithm introduces two threshold value T
l, T
h; Three ordered lists are used to deposit important information: significant coefficient table LSP; Inessential coefficient table LIP; Inessential subset table LIS; And the set symbol introduced below: the coordinate set H of all tree roots is pyramidal low frequency and the one deck closest to low-frequency range; Tree is the descendants of tree root and tree must have child; The set O of all children of node (r, c)
r, c; The set D of all descendants of node (r, c)
r, c, be called that D class is set; The set L of all non-direct descendants of node (r, c)
r, c, be called that L class is set; If have at least one coefficient to be important in above-mentioned tree set, then this tree is claimed to be important; The embedding bit stream of SPIHT is divided into sequence bit stream Sn and meticulous bit stream Rn;
The step of described privileged direction pyramid spiht algorithm is as follows:
The first step, setting threshold value T
l, T
h; LIP is made to be the coefficient of all tree root coordinate sets; LIS is D class tree; LSP is empty set;
Second step, detects the importance of all coefficients in LIP, if important, then export 1 and sign bit, and coefficient is moved into LSP; If inessential, then export 0;
3rd step, according to the importance of all trees in the type checking LIS of tree, specific as follows: D class to be set: if important, then export 1, and its child of encoding; If child is important, then export 1, then export the symbol that a bit represents this coefficient, and this child is joined in LSP; If child is inessential, then export 0, be added to the end of LIP; If this child has descendants, then this D class tree is changed to L class tree and move on to the last of LIS, otherwise delete from LIS; If inessential, then export 0; L class is set: if important, then export 1, and the D class of each child tree is added to LIS end, and from LIS, remove original L class tree; If inessential, then export 0;
4th step, changes threshold value T
l, T
h, determine whether to meet halt condition, if do not met, transfer to second step.
2. method according to claim 1, becomes described direction pyramid some band frequency components of high frequency content, low frequency amount and different layers picture breakdown in steps A; The every one deck generating direction pyramid needs lower column filter: Hi-pass filter, the low-pass filter of different scale, the bandpass filter of different scale, different directions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310024432.1A CN103065336B (en) | 2013-01-23 | 2013-01-23 | Directional pyramid coding method of image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310024432.1A CN103065336B (en) | 2013-01-23 | 2013-01-23 | Directional pyramid coding method of image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103065336A CN103065336A (en) | 2013-04-24 |
| CN103065336B true CN103065336B (en) | 2015-05-27 |
Family
ID=48107953
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310024432.1A Active CN103065336B (en) | 2013-01-23 | 2013-01-23 | Directional pyramid coding method of image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103065336B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109308723A (en) * | 2017-07-26 | 2019-02-05 | 合肥图鸭信息科技有限公司 | A kind of method for compressing image and device |
| CN109308724A (en) * | 2017-07-26 | 2019-02-05 | 合肥图鸭信息科技有限公司 | A kind of method for compressing image and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101594532A (en) * | 2009-06-23 | 2009-12-02 | 上海电力学院 | Region growing compression method based on quad-tree partition |
| CN102637302A (en) * | 2011-10-24 | 2012-08-15 | 北京航空航天大学 | Image coding method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8385670B2 (en) * | 2008-08-20 | 2013-02-26 | Microsoft Corporation | Image restoration by vector quantization utilizing visual patterns |
-
2013
- 2013-01-23 CN CN201310024432.1A patent/CN103065336B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101594532A (en) * | 2009-06-23 | 2009-12-02 | 上海电力学院 | Region growing compression method based on quad-tree partition |
| CN102637302A (en) * | 2011-10-24 | 2012-08-15 | 北京航空航天大学 | Image coding method |
Non-Patent Citations (2)
| Title |
|---|
| "基于DCT分层结构的遥感图像分级多描述编码算法";王相海等;《遥感学报》;20111231;第998-1007页 * |
| "基于提升方向波变换域的SAR图像压缩";白静等;《红外与毫米波学报》;20090831;第28卷(第4期);第311-315页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103065336A (en) | 2013-04-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110087092B (en) | Low-bit-rate video coding and decoding method based on image reconstruction convolutional neural network | |
| JP4212587B2 (en) | Method and apparatus for zero-tree encoding of wavelet data and computer system | |
| DE19534943B4 (en) | Device for compression using embedded small waves | |
| CN105761290B (en) | A kind of adaptive multiple dimensioned splits' positions perception method of sampling | |
| CN102186069B (en) | Remote sensing image data compression method capable of maintaining measurement performance | |
| CN110248190A (en) | A kind of compressed sensing based multilayer residual error coefficient image encoding method | |
| Kumar et al. | A review: DWT-DCT technique and arithmetic-Huffman coding based image compression | |
| Yadav et al. | Study and analysis of wavelet based image compression techniques | |
| CN110097490A (en) | The insertion of area-of-interest reversible water mark and extracting method based on IWT-Schur | |
| CN108093264A (en) | Core image compression, decompressing method and the system perceived based on splits' positions | |
| CN113962882A (en) | JPEG image compression artifact eliminating method based on controllable pyramid wavelet network | |
| CN103065336B (en) | Directional pyramid coding method of image | |
| CN106530196A (en) | JPEG2000 based multi-scale Laplacian pyramid watermarking method | |
| CN1159922C (en) | Method for encoding binary shape signal | |
| CN1284120C (en) | Synthetic aperture radar complex numeric image data real time automatic compression method | |
| Bin et al. | An improved SPIHT wavelet transform in the underwater acoustic image compression | |
| CN110827399A (en) | Three-dimensional model data compression method and system | |
| Jiang et al. | A hybrid image compression algorithm based on human visual system | |
| Radad et al. | A Hybrid Discrete Wavelet Transform with Vector Quantization for Efficient Medical Image Compression | |
| CN101267557A (en) | A method for image compression based on compound vector quantification | |
| Nagamani et al. | EZW and SPIHT image compression techniques for high resolution satellite imageries | |
| Feng et al. | A Fractal Image Compression Method Based on Multi-Wavelet | |
| Revathi et al. | An efficient trapezoidal compression algorithm using wavelet transformation for medical image | |
| CN114449248B (en) | Spherical image compression method based on spherical wavelet transformation | |
| Padmashree et al. | Statistical analysis of objective measures using fractal image compression for medical 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 | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |