CN107027028A - Random offset based on JND quantifies the method and system of multiple description coded decoding - Google Patents
Random offset based on JND quantifies the method and system of multiple description coded decoding Download PDFInfo
- Publication number
- CN107027028A CN107027028A CN201710192541.2A CN201710192541A CN107027028A CN 107027028 A CN107027028 A CN 107027028A CN 201710192541 A CN201710192541 A CN 201710192541A CN 107027028 A CN107027028 A CN 107027028A
- Authority
- CN
- China
- Prior art keywords
- subset
- yield
- value
- description
- transform
- 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.)
- Granted
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
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/124—Quantisation
-
- 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
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/12—Selection from among a plurality of transforms or standards, e.g. selection between discrete cosine transform [DCT] and sub-band transform or selection between H.263 and H.264
- H04N19/122—Selection of transform size, e.g. 8x8 or 2x4x8 DCT; Selection of sub-band transforms of varying structure or type
-
- 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
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/13—Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC]
Abstract
The present invention relates to a kind of random offset based on JND it is multiple description coded, decoding method and system, wherein multi-description coding method includes obtaining image information, image information is divided into two subsets, respectively as the first subset and yield in the second subset, the description of generation first simultaneously and the second description, described two descriptions are simultaneously including the first subset and yield in the second subset;Described to first, based on many description methods, yield in the second subset is predicted using the first subset, and entropy code is carried out according to the predicted value of the first subset and yield in the second subset, output more than first describes code stream;Described to second, based on many description methods, the first subset is predicted using yield in the second subset, and entropy code is carried out according to the predicted value of yield in the second subset and the first subset, output more than second describes code stream;Many description methods include successively carrying out described two subsets the lapped transform based on time-domain, dct transform, the threshold filter based on JND and random offset quantization processing.
Description
Technical field
The present invention relates to communication technical field, specifically a kind of random offset based on JND quantifies multiple description coded, decoding
Method and system.
Background technology
Internet and cordless communication network are unreliable transmission channel:On the one hand be because internet have it is very strong different
Structure, user can be whenever and wherever possible by wired or be wirelessly connected into network, and different user is in CPU computing energy
All there is very big difference in power and peripheral hardware performance, the subnet that user Internet access passes through in physical medium, bandwidth resources and
Also all varied in terms of transmission delay, or even same user is also not quite similar in the network transmission situation of different time sections.Mutually
The characteristics of isomerism and channel shared by multiple users of networking, determine its Best-effort method of service.Therefore, interconnect
The transmission of net cannot ensure the reliability of data receiver, that is to say, that packet can be produced due to various channel problems
Packet loss and transmission delay etc..And on the other hand, the high bit-error that wireless channel has, multi-path jamming, decline etc. property
Determine that the environment being wirelessly transferred is more severe than internet, be likely to result in the whole figure loss of transmission image or the whole frame of video is lost
Lose, so as to cause transmission channel entirely ineffective.Therefore, this bottle for also becoming realtime graphic and video transmission technologies development
Neck.Therefore, design image/video efficient compression and the encoding scheme with robust performance, that is, image/video difference
Wrong control technology (Error Control), be the research topic of a unusual meaningful and practicality, as message sink coding end
Multiple Description Coding in error-control technique.
Multiple description coded appearance be initially for solving practical problems the need for, then researchers have carried out phase to it
Ground theory analysis and innovation are answered, it is last to return to practical application again.Earliest, in the 1970s, in order to be provided on telephone network
Continual telephone service, carries out odd even sampled point by the signal from a call and separates to form two paths of signals, and at two
The channel of separation, at that time the problem by Bell laboratories be referred to as channel separation (ChalmelSPhtting) problem.1979
In the theoretical seminars of the Shannon of year September, Gersho, Ozarow, Witsenhausen, Wolf, Wyner and Ziv et al. are formal
Propose many description problems.After the theoretical research in one period, VaishamPayan etc. proposes first and practical retouched more
State encoding scheme (MDSQ, Multiple Deseription Scalar Quantization).Later many novel practicalities
Type multi-description coding method occurs:Such as based on sampling it is multiple description coded, based on quantization it is multiple description coded, based on correlating transforms
Multiple description coded and based on non-equal protection multiple description coded scheme.Domestic many scholars are also to multiple description coded scheme
In depth studied.
The multiple description coded channel condition that is mainly used in is than in relatively rugged environment.Multiple description coded conduct one kind can be easy
The technology of transmission robustness is improved on wrong channel, has attracted increasing scholar's research.Assume that information source and the stay of two nights it
Between there are many channels, it is impossible to all channels are simultaneously in the error of frame of video, and multiple description coded is in this hypothesis
On the basis of put forward.Multiple bit streams (referred to as describing) are produced in coding side information source, each bit stream has equally excellent
First level, is transmitted on multiple separate channels.In decoding end, each description can be decoded independently, and rebuild and used
The video sequence of family acceptable quality;With the increase for receiving description quantity, rebuild video sequence quality and also improve therewith.Such as
Really some description is lost in the channel, and receiving terminal is without waiting for the transmission again for losing description in this case, and can be with
According to the description received with losing the correlation between description, to estimate the part description of loss, so avoid
Delay caused by retransmitting, it is ensured that the real-time of information transfer.In addition, the description quantity received is more, reconstruction image matter
Amount is better, and the reliability of information transfer is higher.
But conventional Image Coding Algorithms are all to consider how to improve code efficiency and picture quality from objective angle,
The influence of this subjective criterion of human-eye visual characteristic was not considered, and this is worthless, in most cases human eye after all
The final recipient of image, thus we can with and be highly desirable to using human visual system come Optimized Coding Based algorithm.
Vision system is that the mankind inherit from the jellyfish before 700,000,000 years, there is it, and the mankind can just perceive light and multicoloured
Color, preferably could be interacted with the external world.This vision system is referred to as human visual system.From ancient times to the present, the mankind are from psychology
Learn and the angle of physiology has done very many researchs to human visual system.Human eye can tell figure in a short period of time
Key message as in, ignores secondary information therein.
JND (Just noticeable difference) model is to set up perception by exploring all kinds of visual signatures to miss
The threshold value of difference, distinguishes that people be can perceiving and not perceived signal, and then removes visually-perceptible redundancy.When image becomes
When change value is less than threshold value, human eye will not perceive this change.
The content of the invention
For the deficiencies in the prior art, the present invention proposes a kind of random offset based on JND and quantifies describe more
The method of coding and decoding, can preferably be distributed coding resource, coded image is more conformed to the subjective vision of human eye
Impression, so as to improve code efficiency on the premise of ensureing to solve the problems, such as packet loss.
The present invention uses following technical scheme:
A kind of random offset based on JND quantifies multi-description coding method, including:
Image information is obtained, image information is divided into two subsets, it is simultaneously raw respectively as the first subset and yield in the second subset
Into the first description and the second description, described two descriptions are simultaneously including the first subset and yield in the second subset;
Described to first, based on many description methods, yield in the second subset is predicted using the first subset, and according to the first subset and the
The predicted value of two subsets carries out entropy code, and output more than first describes code stream;
Described to second, based on many description methods, the first subset is predicted using yield in the second subset, and according to yield in the second subset and the
The predicted value of one subset carries out entropy code, and output more than second describes code stream;
Many description methods include successively carrying out described two subsets the lapped transform based on time-domain, dct transform,
Threshold filter and random offset quantization processing based on JND.
It is specially to assume that quantization steps of the x in the first description is q that the random offset, which quantifies processing,0, it is retouched second
Predicted value in stating isPrediction redundancy is ei, predict the quantization step q of redundancy1More than q0。
Further, the dct transform is two-dimensional dct:
K, l=0,1 ..., N-1
Wherein
Further, the threshold filter model based on JND includes spatial contrast sensitivity function, the adaptive letter of brightness
Number and the product of texture shielding effect model three:
TJND=TBasic×Flum×Fcontrast
1. space CSF characteristic models TBasicDraw:
First, DCT normalization coefficients φ is calculatedm:
Secondly, the corresponding spatial frequency w of DCT sub-band coefficients at position (i, j) place is calculatedi,j:
Wherein between 3 and 6, P is the height of piece image to R values (unit is pixel).
Then, the deflection of DCT coefficient is calculated
Finally, space CSF characteristic models are calculated:
It is the basic JND threshold value of the DCT coefficient of (i, j), its calculation formula that space CSF characteristic models, which embody call number,
For:
Wherein, s represents to gather effect, and value is 0.25;Human eye gap tilt effect is represented, r, which learns from else's experience, to be tested
Value 0.6;A, b, c distinguish value 1.33,0.11,0.18;I, j=1:N.
2. background luminance adaptive weighted factor FlumCalculation formula is:
WhereinFor the mean flow rate of regional area, it can be byCalculate, DC is DCT direct currents system
Number, N is block size.
3. texture shielding effect weighted factor Fcontrast
When calculating the factor, first to image block and by block sort, block is divided into three classes:Smooth area, marginal zone and
Texture area.To an image block, if comprising less edge pixel, being regarded as smooth block.On the other hand, if comprising
More edge pixel, it is meant that contain higher energy in image block, be then considered as texture block, Canny operators are used here
To image block classification, its calculation formula is:
Wherein ∑edgelAll edge pixel sums in certain block calculated by Canny operators are represented, N represents block size,
Wherein different regions is weighted, weighting ψ=1 of smooth region marginal zone;Because human eye is to texture region
Sensitive relatively small, therefore weight coefficient ψ=2.25 of low frequency coefficient, and weighting ψ=1.25 of high frequency coefficient.
Further, the method for the description of output more than first code stream includes:Described to first, using based on the overlapping of time-domain
Convert before being carried out respectively to the first subset, yield in the second subset to filtering, DCT changes are carried out successively to the forward direction filter value of the first subset
Change, the threshold filter based on JND and random offset quantization are handled, and yield in the second subset is predicted according to the DCT reconstructed coefficients of the first subset
Forward direction filter value, the forward direction filter forecasting redundancy to yield in the second subset carries out dct transform and quantification treatment successively, to described first
Entropy code is carried out to the quantized value of filter forecasting redundancy before the quantized value and yield in the second subset of subset, output more than first describes code stream.
The yield in the second subset obtained according to the forward direction filter forecasting redundancy of the yield in the second subset in the lapped transform of time-domain
To the redundancy between filter value before forward direction filter value, with the prediction of yield in the second subset.
Further, the method for the description of output more than second code stream includes:Described to second, using based on the overlapping of time-domain
Convert before being carried out respectively to the first subset, yield in the second subset to filtering, DCT changes are carried out successively to the forward direction filter value of yield in the second subset
Change, the threshold filter based on JND and random offset quantization are handled, and the first subset is predicted according to the DCT reconstructed coefficients of yield in the second subset
Forward direction filter value, the forward direction filter forecasting redundancy to the first subset carries out dct transform and quantification treatment successively, to described second
Entropy code is carried out to the quantized value of filter forecasting redundancy before the quantized value of subset and the first subset, output more than second describes code stream.
The first subset obtained according to the forward direction filter forecasting redundancy of first subset in the lapped transform of time-domain
Forward direction filter value, to the redundancy between filter value before the prediction with the first subset.
Further, the forward direction filtering is using forward-direction filter P:P=W diag { I, V } W, wherein I beList
Bit matrix, V isInvertible matrix, W is butterfly matrix:
Wherein J isAnti- unit matrix.
Further, the quantification treatment stage in being described first, the quantization step of the first subset is less than before yield in the second subset
To the quantization step of filter forecasting redundancy;Quantification treatment stage in being described second, the quantization step of yield in the second subset is more than the
To the quantization step of filter forecasting redundancy before one subset.
Further, according to spatial contrast sensitivity function, brightness auto-adaptive function and texture shielding effect model three
Product, carry out the threshold filter processing based on JND.
Present invention also offers describing solution, coding method a kind of time-domain lapped transform based on JND more, including:
Receive and describe code stream more described first and carry out entropy decoding, based on many description inverse transform methods, output first is reconstructed
Image information;
Receive and describe code stream more described second and carry out entropy decoding, based on many description inverse transform methods, output second is reconstructed
Image information;
Many description inverse transform methods include carrying out inverse quantization processing, idct transform and base to described two subsets successively
In the anti-lapped transform of time-domain.
Further, the idct transform is calculated as follows:
M, n=0,1 ..., N-1
Wherein
Further, the method for the image information of the reconstruct of output first includes:According to the entropy of more than first description code streams
Decoded result, carries out inverse quantization, idct transform, according to the to the first subset, the forward direction filter forecasting redundancy of yield in the second subset respectively
The idct transform value of one subset predicts the idct transform value of yield in the second subset, and the forward direction of the predicted value and the yield in the second subset is filtered
Ripple predicts the idct transform results added of redundancy, obtains the DCT coefficient reconstructed value of yield in the second subset;By the IDCT of first subset
The DCT coefficient reconstructed value of transformed value and yield in the second subset carries out the backward filtering process based on time-domain lapped transform, output first
The image information of reconstruct.
Further, the method for the image information of the reconstruct of output second includes:According to the entropy of more than second description code streams
Decoded result, carries out inverse quantization, idct transform, according to the to yield in the second subset, the forward direction filter forecasting redundancy of the first subset respectively
The idct transform value of two subsets predicts the idct transform value of the first subset, and the forward direction of the predicted value and first subset is filtered
Ripple predicts the idct transform results added of redundancy, obtains the DCT coefficient reconstructed value of the first subset;By the IDCT of the yield in the second subset
The DCT coefficient reconstructed value of transformed value and the first subset carries out the backward filtering process based on time-domain lapped transform, output second
The image information of reconstruct.
Further, the backward filtering is using backward filter T:T=P-1=W diag { I, V-1}W.Wherein I isUnit matrix, V isInvertible matrix, W is butterfly matrix:
Wherein J isAnti- unit matrix.
Further, the inverse quantization processing stage in being described first, the inverse quantization step-length of the first subset is less than the second son
The inverse quantization step-length of collection prediction redundancy;Inverse quantization processing stage in being described second, the inverse quantization step-length of yield in the second subset is more than
First subset predicts the inverse quantization step-length of redundancy.
Present invention also offers the system that coding and decoding is described a kind of time-domain lapped transform based on JND more, including:
Encoder, obtains image information, image information is divided into two subsets, respectively as the first subset and the second son
Collection, while the description of generation first and the second description, described two descriptions are simultaneously including the first subset and yield in the second subset;Based on many
Description method, yield in the second subset is predicted to the first description using the first subset, and according to the first subset and the predicted value of yield in the second subset
Entropy code is carried out, output more than first describes code stream;Described to second, the first subset is predicted using yield in the second subset, and according to second
The predicted value of subset and the first subset carries out entropy code, and output more than second describes code stream;
Transmission network, for being transmitted to many descriptor code streams;
Decoder, receives more than the first descriptions code stream and second and describes code stream and carry out entropy decoding, anti-based on many descriptions
The image information of transform method, the image information of the reconstruct of output first and the second reconstruct;
Many description methods include successively carrying out described two subsets the lapped transform based on time-domain, dct transform,
Threshold filter and random offset quantization processing based on JND.
Many description inverse transform methods include carrying out inverse quantization processing, idct transform and base to described two subsets successively
In the anti-lapped transform of time-domain.
Beneficial effects of the present invention:The present invention can ensure to solve the problems, such as to improve coding on the basis of image transmitting packet loss
Efficiency, using can the lapped transform based on time-domain, forward-direction filter and backward filtering can be optimized according to different applications
Device, eliminates blocking artifact, the model based on JND, adds the openness of DCT coefficient, so as to improve code efficiency.
Brief description of the drawings
Fig. 1 is method schematic of the invention;
Fig. 2 is the lapped transform fundamental block diagram of the invention based on time-domain;
Fig. 3 is the dct transform image of an embodiment;
The prediction block diagram that Fig. 4 describes for the present invention two;
Fig. 5 is the schematic diagram that random offset of the present invention quantifies.
Embodiment:
The invention will be further described with embodiment below in conjunction with the accompanying drawings:
It is noted that described further below is all exemplary, it is intended to provide further instruction to the application.Unless another
Indicate, all technologies used herein and scientific terminology are with usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singulative
It is also intended to include plural form, additionally, it should be understood that, when in this manual using term "comprising" and/or " bag
Include " when, it indicates existing characteristics, step, operation, device, component and/or combinations thereof.
A kind of exemplary embodiments of the present invention are a kind of time-domain lapped transform multi-description coding methods based on JND, bag
Include:
As shown in figure 1, obtaining image information, image information is divided into two subsets (S0, S1), while generating 0 He of description
Description 1, two descriptions are simultaneously including subset S0 and subset S1;
Described to first, based on many description methods, yield in the second subset is predicted using the first subset, and according to the first subset and the
The predicted value of two subsets carries out entropy code, and output more than first describes code stream;
Described to second, based on many description methods, the first subset is predicted using yield in the second subset, and according to yield in the second subset and the
The predicted value of one subset carries out entropy code, and output more than second describes code stream;
Technical scheme in the present embodiment first to one of description (assuming that description 0) is clearly and completely introduced.
Step one:Read image and divide an image into two subsets (S0, S1)
Step 2:Lapped transform based on time-domain is carried out to subset S0 and subset S1, i.e., it is preceding to filtering (P), removal side
Blocking effect;
Step 3:Dct transform is carried out after forward direction filtering, so that image energy is converted in the dispersed distribution of spatial domain
The Relatively centralized distribution of transform domain;
Step 4:JND pretreatments are carried out to one of subset (S0).
Step 5:The subset (S0) is quantified (Q) with less quantization (q0) step-length;
Step 6:The forward direction filter value of another subset (S1) is predicted with the DCT coefficient reconstructed value of the subset (S0), and it is right
Predict that redundancy carries out dct transform, quantified, higher value is taken to the quantization step (q1) that prediction redundancy is carried out;
Step 7:Entropy code, many description code streams of output description 0.
Wherein quantized segment quantifies method using random offset:As shown in figure 5, being the example of one two description, it is assumed that x exists
Quantization step in description 0 is q0, it is in the predicted value described in 1Prediction redundancy is ei, predict redundancy larger quantities
Change step-length q1Quantify.
To description 1 processing ibid, different place is, since step 4, is that one of subset (S1) is entered
Row JND is pre-processed;Then the subset (S1) is quantified (Q) with less quantization (q0) step-length;With the DCT of the subset (S1)
Coefficient reconstruction value predicts the forward direction filter value of another subset (S0), and carries out dct transform to prediction redundancy, quantifies, to prediction
The quantization step (q1) that redundancy is carried out takes higher value;It is last to carry out entropy code, many description code streams of output description 1 again.
One more embodiment of the present invention is a kind of time-domain lapped transform multiple description encoding method based on JND, above-mentioned
On the basis of embodiment,
Step 8:Entropy decoding is carried out to many descriptor code streams of description 0,
Step 9:According to description 0 entropy decoding result, respectively the forward direction filter forecasting redundancy to subset S0, subset S1 enter
Row inverse quantization (IQ);
Step 10:Inverse DCT converts (IDCT);
Step 11:Predict subset S1 idct transform value according to subset S0 idct transform value, and by the predicted value with
The idct transform results added of the forward direction filter forecasting redundancy of the subset S1, obtains subset S1 DCT coefficient reconstructed value;By institute
State subset S0 idct transform value and subset S1 DCT coefficient reconstructed value is carried out at the backward filtering based on time-domain lapped transform
Reason, the image information of output description 1.
To description 1 processing ibid, different place is, since step 11, according to subset S1 IDCT become
Value prediction subset S0 idct transform value is changed, and the IDCT of the predicted value and the forward direction filter forecasting redundancy of the subset S0 is become
Results added is changed, subset S0 DCT coefficient reconstructed value is obtained;By the idct transform value of the subset S1 and subset S0 DCT systems
Number reconstructed value carries out the backward filtering process based on time-domain lapped transform, the image information of output description 1.
Wherein the lapped transform based on time-domain is as shown in Fig. 2 its advantage is before being optimized according to different applications
To wave filter (step 2) and backward filter (step 11).In coding side, dct transform and forward-direction filter are applied on side
Boundary, boundary is first with L × L forward direction prefilter P, the then each piece of DCT using L points.TDLT basic letter
Number include two pieces, and adjacent block junction it is overlapping be one piece.In decoding end, DCT inverse transformations and backward filter T are employed
To block boundary.P and T are:P=W diag { I, V } W T=P-1=W diag { I, V-1}W.Wherein I isUnit matrix, V
It isInvertible matrix, W is butterfly matrix:
Wherein J isAnti- unit matrix.
Dct transform, that is, discrete cosine transform are specifically a kind of conversion related to Fourier transformation, and it can make image
Energy is converted to the Relatively centralized distribution of transform domain in the dispersed distribution of spatial domain, to reach the purpose for removing spatial redundancy.
Image, Video coding mainly use two-dimensional dct:
K, l=0,1 ..., N-1
Wherein
The weighted sum for 64 basic images being expressed as any 8 × 8 block of pixels in the present embodiment shown in Fig. 3, its weights is
For the DCT coefficient of correspondence position.
IDCT inverse transformations (IDCT) are calculated as follows:
M, n=0,1 ..., N-1
Using JND model to test image (Lena Boat Peppers Couple Goldhill in the present embodiment
Baboon DCT coefficient) is filtered processing, to the DCT coefficient zero setting processing less than JND threshold value;The JND model is space pair
Than the product of sensitivity function, brightness auto-adaptive function and texture shielding effect model three:
TJND=TBasic×Flum×Fcontrast (4)
1. space CSF characteristic models TBasic
First, DCT normalization coefficients φ is calculatedm:
Secondly, the corresponding spatial frequency w of DCT sub-band coefficients at position (i, j) place is calculatedi,j:
Between 3 and 6, P is the height (unit is pixel) of piece image to the usual values of wherein R, in addition when having calculated (8),
Radian is converted into angle.
Then, the deflection of DCT coefficient is calculated
Finally, space CSF characteristic models are calculated:
It is the basic JND threshold value of the DCT coefficient of (i, j), its calculation formula that space CSF characteristic models, which embody call number,
For:
Wherein, s represents to gather effect, and value is 0.25;Human eye gap tilt effect is represented, r, which learns from else's experience, to be tested
Value 0.6;A, b, c distinguish value 1.33,0.11,0.18;I, j=1:N.
2. background luminance adaptive weighted factor Flum
Human nervous system is inversely proportional to the sensitivity of signal and the intensity of background signal, and the intensity of background signal is got over
Greatly, the susceptibility of the mankind is lower, and separating capacity is just smaller, and its calculation formula is:
WhereinFor the mean flow rate of regional area, it can be byCalculate, DC is DCT direct currents system
Number, N is block size.
3. texture shielding effect weighted factor Fcontrast
When calculating the factor, first to image block and by block sort, block is divided into three classes:Smooth area, marginal zone and
Texture area.To an image block, if comprising less edge pixel, being regarded as smooth block.On the other hand, if comprising
More edge pixel, it is meant that contain higher energy in image block, be then considered as texture block, Canny operators are used here
To image block classification, its calculation formula is:
ρedgel=∑edgel/N2 (11)
Wherein ∑edgelAll edge pixel sums in certain block calculated by Canny operators are represented, N represents block size,
Block sort method can be written as to (13) formula according to (12):
Great amount of images is trained, it is as a result best when α takes 0.1, β to take 0.2.
Secondly, after block sort terminates, texture shielding effect weighted factor is calculated, its computational methods is:
Wherein different regions is weighted, weighting ψ=1 of smooth region marginal zone;Because human eye is to texture region
Sensitive relatively small, therefore weight coefficient ψ=2.25 of low frequency coefficient, and weighting ψ=1.25 of high frequency coefficient.
Table 1 is to 6 width test images (Lena Boat Peppers Couple Goldhill using JND model
Baboon DCT coefficient) is filtered the result of processing, such as table 1 as can be seen that the coefficient after JND is handled occurs a lot
Zero, signal becomes more sparse.This is also indicated that in the case where not influenceing the basis of subjective quality, adds the openness of DCT coefficient;
Table 1
Predicted portions in the present embodiment are as shown in Figure 4.In description 0, subset S0 is quantified with less quantization step, its
Inverse quantization inverse DCT value is used for predicting subset S1, and then the prediction redundancy to subset S1 carries out DCT and quantification treatment, amount now
Change step-length and take higher value.In description 1, subset S1 is quantified with less quantization step, and its inverse quantization inverse DCT value is used for predicting son
Collect S0, then the prediction redundancy to subset S0 carries out DCT and quantification treatment, and quantization step now takes higher value.Predicted portions
It is predicted with inverse DCT value, that is to say, that predicted portions herein are carried out in pixel domain.
Another embodiment of the present invention is the system that coding and decoding is described a kind of time-domain lapped transform based on JND more,
Include on the basis of above-described embodiment:
Encoder, obtains image information, image information is divided into two subsets, respectively as subset S0 and subset S1, together
Shi Shengcheng describes 0 and description 1, and described two descriptions are simultaneously including subset S0 and subset S1;Based on many description methods, to description
0 using subset S0 prediction subset S1, and carry out entropy code, output 0 code stream of description according to subset S0 and subset S1 predicted value;It is right
Description 1, predicts subset S0, and carry out entropy code, 1 yard of output description according to subset S1 and subset S0 predicted value using subset S1
Stream;
Transmission network, for being transmitted to many descriptor code streams;
Decoder, receives 0 code stream of the description and 1 code stream of description and carries out entropy decoding, based on many description inverse transform methods,
The reconstructing image information of output description 0 and the reconstructing image information of description 1;
Many description methods include successively carrying out described two subsets the lapped transform based on time-domain, dct transform,
Threshold filter and random offset quantization processing based on JND.
Many description inverse transform methods include carrying out inverse quantization processing, idct transform and base to described two subsets successively
In the anti-lapped transform of time-domain.
Transmission network therein can be finite element network or wireless network.
The preferred embodiment of the application is the foregoing is only, the application is not limited to, for the skill of this area
For art personnel, the application can have various modifications and variations.It is all within spirit herein and principle, made any repair
Change, equivalent substitution, improvement etc., should be included within the protection domain of the application.
Claims (10)
1. a kind of random offset based on JND quantifies multi-description coding method, it is characterised in that including:
Image information is obtained, image information is divided into two subsets, respectively as the first subset and yield in the second subset, while generation the
One description and the second description, described two descriptions are simultaneously including the first subset and yield in the second subset;
Described to first, based on many description methods, yield in the second subset is predicted using the first subset, and according to the first subset and the second son
The predicted value of collection carries out entropy code, and output more than first describes code stream;
Described to second, based on many description methods, the first subset is predicted using yield in the second subset, and according to yield in the second subset and the first son
The predicted value of collection carries out entropy code, and output more than second describes code stream;
Many description methods include carrying out the lapped transform based on time-domain, dct transform to described two subsets successively, are based on
JND threshold filter and random offset quantization processing;
It is specially to assume that quantization steps of the x in the first description is q that the random offset, which quantifies processing,0, it is in second describes
Predicted value isPrediction redundancy is ei, predict the quantization step q of redundancy1More than q0。
2. according to the method described in claim 1, it is characterised in that the method that output more than first describes code stream includes:To first
Description, to filtering before being carried out respectively to the first subset, yield in the second subset using the lapped transform based on time-domain, to the first subset
Forward direction filter value carries out dct transform, the threshold filter based on JND and random offset quantization processing successively, according to the first subset
DCT reconstructed coefficients predict the forward direction filter value of yield in the second subset, and the forward direction filter forecasting redundancy to yield in the second subset carries out DCT changes successively
Change and quantification treatment, to carrying out entropy volume to the quantized value of filter forecasting redundancy before the quantized value and yield in the second subset of first subset
Code, output more than first describes code stream.
3. method according to claim 2, it is characterised in that the method that output more than second describes code stream includes:To second
Description, to filtering before being carried out respectively to the first subset, yield in the second subset using the lapped transform based on time-domain, to yield in the second subset
Forward direction filter value carries out dct transform, the threshold filter based on JND and random offset quantization processing successively, according to yield in the second subset
DCT reconstructed coefficients predict the forward direction filter value of the first subset, and the forward direction filter forecasting redundancy to the first subset carries out DCT changes successively
Change and quantification treatment, entropy volume is carried out to the quantized value of filter forecasting redundancy before the quantized value and the first subset to the yield in the second subset
Code, output more than second describes code stream.
4. method according to claim 3, it is characterised in that the quantification treatment stage in being described first, the first subset
Quantization step be less than yield in the second subset before to filter forecasting redundancy quantization step;The quantification treatment stage in being described second,
The quantization step of yield in the second subset is more than the quantization step to filter forecasting redundancy before the first subset.
5. according to the method described in claim 1, it is characterised in that according to spatial contrast sensitivity function, the adaptive letter of brightness
Number and the product of texture shielding effect model three, carry out the threshold filter processing based on JND.
6. a kind of many description solutions based on claim 1, coding method, it is characterised in that including:
Receive and describe code stream more described first and carry out entropy decoding, based on many description inverse transform methods, the figure of the reconstruct of output first
As information;
Receive and describe code stream more described second and carry out entropy decoding, based on many description inverse transform methods, the figure of the reconstruct of output second
As information;
Many description inverse transform methods include successively carrying out described two subsets inverse quantization processing, idct transform and based on when
Between domain anti-lapped transform.
7. method according to claim 6, it is characterised in that the method bag for the image information that the output first is reconstructed
Include:According to the entropy decoding result of more than first description code streams, the first subset, the forward direction filter forecasting redundancy of yield in the second subset are entered respectively
Row inverse quantization, idct transform, the idct transform value of yield in the second subset is predicted according to the idct transform value of the first subset, and this is predicted
The idct transform results added of value and the forward direction filter forecasting redundancy of the yield in the second subset, obtains the DCT coefficient weight of yield in the second subset
Built-in value;The DCT coefficient reconstructed value of the idct transform value of first subset and yield in the second subset is carried out based on the overlapping change of time-domain
The backward filtering process changed, the image information of the reconstruct of output first.
8. method according to claim 7, it is characterised in that the method bag for the image information that the output second is reconstructed
Include:According to the entropy decoding result of more than second description code streams, yield in the second subset, the forward direction filter forecasting redundancy of the first subset are entered respectively
Row inverse quantization, idct transform, the idct transform value of the first subset is predicted according to the idct transform value of yield in the second subset, and this is predicted
The idct transform results added of value and the forward direction filter forecasting redundancy of first subset, obtains the DCT coefficient weight of the first subset
Built-in value;The DCT coefficient reconstructed value of the idct transform value of the yield in the second subset and the first subset is carried out based on the overlapping change of time-domain
The backward filtering process changed, the image information of the reconstruct of output second.
9. method according to claim 8, it is characterised in that the inverse quantization processing stage in being described first, the first son
The inverse quantization step-length of collection is less than the inverse quantization step-length that yield in the second subset predicts redundancy;Inverse quantization processing stage in being described second,
The inverse quantization step-length of yield in the second subset is more than the inverse quantization step-length that the first subset predicts redundancy.
10. the system of coding and decoding is described a kind of random offset based on JND more, it is characterised in that including:
Encoder, obtains image information, image information is divided into two subsets, respectively as the first subset and yield in the second subset, together
Shi Shengcheng first is described and the second description, and described two descriptions are simultaneously including the first subset and yield in the second subset;Based on many descriptions
Method, predicts yield in the second subset, and carry out according to the predicted value of the first subset and yield in the second subset to the first description using the first subset
Entropy code, output more than first describes code stream;Described to second, the first subset is predicted using yield in the second subset, and according to yield in the second subset
Entropy code is carried out with the predicted value of the first subset, output more than second describes code stream;
Transmission network, for being transmitted to many descriptor code streams;
Decoder, receives more than the first descriptions code stream and second and describes code stream and carry out entropy decoding, based on many description inverse transformations
The image information of method, the image information of the reconstruct of output first and the second reconstruct;
Many description methods include carrying out the lapped transform based on time-domain, dct transform to described two subsets successively, are based on
JND threshold filter and random offset quantization processing.
Many description inverse transform methods include successively carrying out described two subsets inverse quantization processing, idct transform and based on when
Between domain anti-lapped transform;
It is specially to assume that quantization steps of the x in the first description is q that the random offset, which quantifies processing,0, it is in second describes
Predicted value isPrediction redundancy is ei, predict the quantization step q of redundancy1More than q0。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710192541.2A CN107027028B (en) | 2017-03-28 | 2017-03-28 | Random offset based on JND quantifies multiple description coded, decoded method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710192541.2A CN107027028B (en) | 2017-03-28 | 2017-03-28 | Random offset based on JND quantifies multiple description coded, decoded method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107027028A true CN107027028A (en) | 2017-08-08 |
CN107027028B CN107027028B (en) | 2019-05-28 |
Family
ID=59525354
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710192541.2A Expired - Fee Related CN107027028B (en) | 2017-03-28 | 2017-03-28 | Random offset based on JND quantifies multiple description coded, decoded method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107027028B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107566831A (en) * | 2017-09-18 | 2018-01-09 | 山东师范大学 | Multiple description coded adaptive coding/decoding method and system are quantified based on random offset |
CN108600750A (en) * | 2018-04-10 | 2018-09-28 | 山东师范大学 | Multiple description coded, coding/decoding method based on KSVD and system |
CN109889848A (en) * | 2019-02-22 | 2019-06-14 | 山东师范大学 | Based on the multiple description coded of convolution self-encoding encoder, coding/decoding method and system |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1672421A (en) * | 2002-07-31 | 2005-09-21 | 皇家飞利浦电子股份有限公司 | Method and apparatus for performing multiple description motion compensation using hybrid predictive codes |
CN101340261A (en) * | 2007-07-05 | 2009-01-07 | 华为技术有限公司 | Multiple description encoding, method, apparatus and system for multiple description encoding |
CN103501441A (en) * | 2013-09-11 | 2014-01-08 | 北京交通大学长三角研究院 | Multiple-description video coding method based on human visual system |
CN104219530A (en) * | 2014-09-28 | 2014-12-17 | 北京工商大学 | HEVC multiple-description coding scheme based on prediction mode copying |
-
2017
- 2017-03-28 CN CN201710192541.2A patent/CN107027028B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1672421A (en) * | 2002-07-31 | 2005-09-21 | 皇家飞利浦电子股份有限公司 | Method and apparatus for performing multiple description motion compensation using hybrid predictive codes |
CN101340261A (en) * | 2007-07-05 | 2009-01-07 | 华为技术有限公司 | Multiple description encoding, method, apparatus and system for multiple description encoding |
CN103501441A (en) * | 2013-09-11 | 2014-01-08 | 北京交通大学长三角研究院 | Multiple-description video coding method based on human visual system |
CN104219530A (en) * | 2014-09-28 | 2014-12-17 | 北京工商大学 | HEVC multiple-description coding scheme based on prediction mode copying |
Non-Patent Citations (1)
Title |
---|
HUIHUI BAI 等: "Multiple Description Video Coding Based on Human Visual System Characteristics", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107566831A (en) * | 2017-09-18 | 2018-01-09 | 山东师范大学 | Multiple description coded adaptive coding/decoding method and system are quantified based on random offset |
CN107566831B (en) * | 2017-09-18 | 2020-11-20 | 山东师范大学 | Self-adaptive coding and decoding method and system based on random offset quantization multiple description coding |
CN108600750A (en) * | 2018-04-10 | 2018-09-28 | 山东师范大学 | Multiple description coded, coding/decoding method based on KSVD and system |
CN109889848A (en) * | 2019-02-22 | 2019-06-14 | 山东师范大学 | Based on the multiple description coded of convolution self-encoding encoder, coding/decoding method and system |
Also Published As
Publication number | Publication date |
---|---|
CN107027028B (en) | 2019-05-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
ZainEldin et al. | Image compression algorithms in wireless multimedia sensor networks: A survey | |
KR20190117651A (en) | Image processing and video compression methods | |
CN110798690B (en) | Video decoding method, and method, device and equipment for training loop filtering model | |
WO2020238439A1 (en) | Video quality-of-service enhancement method under restricted bandwidth of wireless ad hoc network | |
CN102123278A (en) | Signal source encoding method based on distributed compressive sensing technology | |
CN104199627B (en) | Gradable video encoding system based on multiple dimensioned online dictionary learning | |
CN106961607B (en) | Time-domain lapped transform based on JND is multiple description coded, decoded method and system | |
CN107027028B (en) | Random offset based on JND quantifies multiple description coded, decoded method and system | |
CN106385584A (en) | Spatial correlation-based distributed video compressive sensing adaptive sampling and coding method | |
CN101854555B (en) | Video coding system based on prediction residual self-adaptation regulation | |
ZainEldin et al. | A modified listless strip based SPIHT for wireless multimedia sensor networks | |
Hagag et al. | HyperCast: hyperspectral satellite image broadcasting with band ordering optimization | |
CN106899810A (en) | A kind of mine video image fusion method and device | |
Doutsi et al. | Retinal-inspired filtering for dynamic image coding | |
Khosravi et al. | Modified data aggregation for aerial ViSAR sensor networks in transform domain | |
Liu et al. | Multimedia image compression method based on biorthogonal wavelet and edge intelligent analysis | |
CN107566831A (en) | Multiple description coded adaptive coding/decoding method and system are quantified based on random offset | |
Shin et al. | RL-SPIHT: reinforcement learning-based adaptive selection of compression ratios for 1-D SPIHT algorithm | |
Zhang et al. | Can lower resolution be better? | |
CN108600751A (en) | Polygon information-distribution type Video coding based on JND, decoded method and system | |
Kaur et al. | Algorithm for fuzzy based compression of gray JPEG images for big data storage | |
Masmoudi et al. | A bio-inspired image coder with temporal scalability | |
Bao et al. | Image Compression for Wireless Sensor Network: A Model Segmentation-Based Compressive Autoencoder | |
CN103167282A (en) | Online updating method for image compression dynamic regulation | |
Lv et al. | Multiscale compressed sensing method for roi coding |
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 |
Granted publication date: 20190528 |
|
CF01 | Termination of patent right due to non-payment of annual fee |