CN103955908A - Digital image halftone method based on multi-scale perception error measure approximation global optimization - Google Patents
Digital image halftone method based on multi-scale perception error measure approximation global optimization Download PDFInfo
- Publication number
- CN103955908A CN103955908A CN201410152246.0A CN201410152246A CN103955908A CN 103955908 A CN103955908 A CN 103955908A CN 201410152246 A CN201410152246 A CN 201410152246A CN 103955908 A CN103955908 A CN 103955908A
- Authority
- CN
- China
- Prior art keywords
- perceptual error
- yardstick
- global optimization
- multiple dimensioned
- function
- 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
- 238000005457 optimization Methods 0.000 title claims abstract description 64
- 238000000034 method Methods 0.000 title claims abstract description 46
- 230000008447 perception Effects 0.000 title abstract description 10
- 239000000203 mixture Substances 0.000 claims description 12
- 230000007704 transition Effects 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 230000004044 response Effects 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 3
- 230000007423 decrease Effects 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000005286 illumination Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 abstract description 69
- 238000012545 processing Methods 0.000 abstract description 3
- 238000007639 printing Methods 0.000 abstract description 2
- 230000009466 transformation Effects 0.000 abstract 1
- 238000011156 evaluation Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 238000002156 mixing Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 2
- 238000001303 quality assessment method Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003706 image smoothing Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a digital image halftone method based on multi-scale perception error measure approximation global optimization, and belongs to the technical field of digital image pre-printing processing. A multi-scale perception error measure function model framework is established through two-dimensional discrete wavelet transformation, and an intra-scale clustering and cross-scale continuity model is established; an antithesis perception error measure function of an original perception error measure function is established, and the maximum upper bound of the antithesis perception error measure function is calculated, so that approximation global optimization of the original perception error measure function is achieved, and maximum posterior probability configuration of halftone image two-value pixels is achieved based on a perception error measure approximation global optimization strategy; an antithesis perception error measure global optimum upper bound is calculated in an iterative mode through a reparameterization algorithm, two-value pixel optimal configuration in different regions after iteration is conducted each time is defined, and then global optimization approximation coefficients are provided and proved; digital image halftone performance in a dynamic environment is objectively evaluated, and algorithm complexity is analyzed.
Description
Technical field
The present invention relates to a kind of multiple dimensioned perceptual error and estimate the digital picture halftoning method of approximate global optimization, the digital picture that belongs to laser plate-making prints pretreatment technology.
Background technology
Digital picture shadow tone is continuous toned image is developed on the two-value equipment such as laser platemaker, digital printer, laser printer and in human visual system, produce the gordian technique of continuous toned image illusion.In producing, live people, be widely used.The current small desk ink-jet from family, office use, laser printer, laser platemaker are to large-scale publication and printing system, and digital halftone technology can be described as ubiquitous.
For multiple dimensioned digital picture shadow tone, it is dynamic and random estimating due to low scale error, require multiple dimensioned error to estimate blending algorithm and there is higher robustness, under different initialization condition, require multiple dimensioned error to estimate optimized algorithm and converge on consistance result.Like this, make multiple dimensioned error estimate the robustness of blending algorithm and the consistance of optimum results and become multiple dimensioned digital picture shadow tone under dynamic environment and be different from the key restrain of single yardstick shadow tone under static environment.Although existing multiple dimensioned halftoning method has solved the multiresolution imaging problem of half tone image on different two-value equipment, but still do not solve multiple dimensioned error and estimate interaction problems in the yardstick of information and between yardstick.
Summary of the invention
The object of the present invention is to provide a kind of multiple dimensioned perceptual error to estimate the digital picture halftoning method of approximate global optimization; The method is set up the multiple dimensioned model of continuous toned image by two-dimensional discrete wavelet conversion, adopt the response of Nasanen human visual system CSF to set up multiple dimensioned perceptual error and estimate objective function model framework; Adopt digraph to characterize multiple dimensioned perceptual error and estimate objective function model framework, adopt two component Gaussian mixture model to set up the relevant perceptual error measure function of yardstick with HMM; Adopt Lagrangian transform method to prove that antithesis perceptual error measure function is convex function; Formulate the approximate global optimization strategy of the relevant perceptual error measure function of yardstick, obtain the configuration of the approximate global optimum of half tone image binarized pixel; Adopt Graph Theory definition global optimization approximation coefficient and prove that global optimization approximate extents is definite theoretical; The performance of objective evaluation digital picture shadow tone under dynamic environment, and algorithm complex is analyzed; And then obtain optimum half tone image.
The concrete steps of the method for the invention are as follows:
(1) judge whether continuous toned image is 2 of standard
n× 2
nimage;
(2) construct multiple dimensioned perceptual error and estimate objective function model framework: include the standard continuous toned image information in frequency field and direction territory in multiple dimensioned perceptual error and estimate in object module framework, utilize two-dimensional discrete wavelet conversion to set up multiple dimensioned perceptual error and estimate objective function model framework;
(3) set up the relevant perceptual error measure function of yardstick: cluster modeling in the yardstick that employing two component Gaussian mixture model are estimated perceptual error, adopt HMM to quaternary tree across the modeling of yardstick continuation, the error that the relevant perceptual error measure function of this yardstick can be included in yardstick and between yardstick is simultaneously estimated the information of interdepending, to ensure that multiple dimensioned perceptual error estimates the robustness of blending algorithm;
(4) the relevant perceptual error of yardstick is estimated approximate global optimization strategy: its antithesis perceptual error measure function of convex duality model conversation the Theory Construction that utilizes the relevant perceptual error measure function of the archeus described in step (3), calculate antithesis perceptual error by Reparameterization residual plot algorithm iteration and estimate the global optimum upper bound, realize maximum a posteriori probability configuration and the approximate global optimum of the relevant perceptual error measure function of archeus of half tone image binarized pixel, to improve the consistance of multiple dimensioned perceptual error measure function optimum results;
(5) global optimization approximate extents is determined theoretical: utilize antithesis perceptual error described in step (4) to estimate the global optimum upper bound and define the binarized pixel allocation optimum in zones of different after each iteration, and then propose and prove global optimization approximation coefficient; By determining the approximation coefficient of the relevant perceptual error measure function global optimization of yardstick, multiple dimensioned perceptual error can be estimated to optimum results consistance and be controlled in definite known range;
(6) algorithm validity experimental verification under exemplary dynamic environment: the multiple dimensioned perceptual error proposing according to step (1) ~ (5) is estimated approximate global optimization approach framework, set up shadow tone computing platform, the in the situation that of illumination and background dynamics variation, the described relevant perceptual error of yardstick of applying step (3) is estimated modeling and the described approximate global optimization policy calculation half tone image binarized pixel allocation optimum of step (4), the validity of the halftoning algorithm proposing with checking, utilize texture entropy, the method for objectively evaluating of structural similarity and angle second moment is relatively evaluated the performance of digital picture shadow tone under dynamic environment with LSMB method, and algorithm complex is analyzed.
In step of the present invention (2), construct multiple dimensioned perceptual error and estimate objective function model framework and specifically comprise the steps:
1. adopt the multiple dimensioned model of Haar two-dimensional discrete wavelet conversion Criterion continuous toned image in the L * a*b* color space, according to multiple dimensioned model definition wavelet field spatial point
the multiple dimensioned perceptual error at place is estimated information
for corresponding wavelet coefficient is at the Euclidean distance of the L*a*b* color space;
2. the space, frequency and the response of directivity characteristics CSF that adopt Nasanen human vision system model to determine multiple dimensioned perceptual error to estimate
, according to step 1. in multiple dimensioned perceptual error estimate information
respond with CSF
calculating multiple dimensioned perceptual error estimates with the convolution of CSF response and obtains at yardstick
direction
spatial point
locate multiple dimensioned perceptual error and estimate objective function
;
3. by general perceives error measure function
be defined as step
the multiple dimensioned perceptual error of middle diverse location, yardstick and direction is estimated objective function
all side and,
.
In step of the present invention (3), set up the relevant perceptual error measure function of yardstick and specifically comprise the steps:
1. adopt digraph
characterize multiple dimensioned perceptual error and estimate objective function model, wherein
for node set, it is corresponding that each node and perceptual error are estimated,
for source point,
for meeting point,
for line set;
2. adopt cluster modeling in the yardstick that two component Gaussian mixture model estimate perceptual error: on yardstick and subband, add up independent at wavelet field wavelet coefficient, obtain wavelet coefficient set by wavelet coefficient probability distribution
joint probability density function:
, the probability density function of each wavelet coefficient
utilize
two component Gaussian mixture model
, wherein,
represent wavelet coefficient
corresponding hidden state variable
value
time probability mass function,
represent given hidden state variable
value
time wavelet coefficient
conditional probability density function, suppose that this conditional probability density function obeys zero-mean Gaussian distribution,
, wherein,
represent the variance of zero-mean Gaussian distribution,
, and indirectly reflect the size of wavelet coefficient amplitude with this side's extent; In the multi-scale image model of wavelet field, it is the wavelet coefficient of each yardstick, each subband
according to probability mass function
an associated hidden state variable
, this hidden variable has two state values, is respectively 0 and 1, and the height of reflection wavelet coefficient amplitude on this basis, is realized the interior dependence statistical modeling of yardstick of wavelet coefficient;
3. adopt HMM to quaternary tree across the modeling of yardstick continuation: utilize in wavelet field the father-sub-dependence of wavelet coefficient quaternary tree probability graph configuration model wavelet coefficient in subband, across yardstick continuation; Represent a given specific wavelet coefficient across yardstick continuation, this coefficient itself may have identical hidden state tag with its father very much; Connect hidden state variable that wavelet coefficient is corresponding to across the modeling of yardstick continuation with oriented Markov probability figure; Introduce the hidden state transition probability along wavelet field quaternary tree graph model structure
represent wavelet coefficient
the father of a upper yardstick
hidden state
for large or hour,
hidden state
for large or little probability, set up the hidden state of wavelet coefficient across yardstick dependence model; Hidden state when between father-son is identical,
time, expectation state transition probability
larger, set up thus across yardstick continuation model; On this basis, set up the single subband hidden Markov of wavelet field tree-model:
, wherein,
represent single wavelet sub-band and
,
with
for the value of the hidden state variable of wavelet coefficient and
,
the hidden state that represents father's node is
time child's node hidden state be
probability;
by step 2. in two component Gaussian mixture model and 3. in HMM model construction be hidden Markov tree
, wherein
for node probability mass function,
for node
arrive
state transition probability,
for variance;
5. by digraph
class label
the relevant perceptual error measure function of yardstick be defined as
, wherein
with
represent respectively smoothness constraint and unusual constraint,
for smoothness constraint influence coefficient; Adopt the negative log-likelihood of continuous toned image L*a*b* color space brightness amplitude to represent smoothness constraint
, wherein state
, build unusual constraint with exponential function
, utilize smoothness constraint and unusual constraint to set up the relevant perceptual error measure function of yardstick.
In step of the present invention (4), the relevant perceptual error of yardstick is estimated approximate global optimization strategy and is specifically comprised the steps:
1. adopt Lagrangian transform method to realize the relevant perceptual error measure function of archeus and transform to antithesis perceptual error measure function, and prove that antithesis perceptual error measure function is convex function;
2. adopt digraph
reparameterization method realizes the approximate global optimization of the relevant perceptual error measure function of yardstick, meeting under capacity and conservation constraint prerequisite, on the basis of residual capacity, increases and decreases constant flow, carries out residual plot Reparameterization;
3. adopt Graph Theory to prove terminal line and node line cubage correction theory: to pass through digraph
after Reparameterization
the relevant perceptual error measure function of yardstick
there is same item label
method, realizes the iteration optimization that perceptual error is estimated;
4. while reaching iterations or setting accuracy, if source point
with node
be connected,
label is 1; If meeting point
with node
be connected,
label is 0, formulates the approximate global optimization strategy of the relevant perceptual error measure function of yardstick, obtains the configuration of the approximate global optimum of half tone image binarized pixel.
Described in step of the present invention (5), global optimization approximate extents determines that theory specifically comprises the steps:
1. adopt Graph Theory definition global optimization approximation coefficient and prove approximate extents, order
for digraph after Reparameterization
approximate global optimization class tag configurations,
for digraph
global optimum's class tag configurations, definition
cfor global optimization approximation coefficient, set coefficient
; By approximation coefficient
can determine that perceptual error estimates global optimization approximate extents
;
2. define digraph
global optimum's class tag configurations is
node set, and define respectively this node set inside, outside and borderline node set;
3. according to the definition of different label area node sets, determine each region perceptual error measure function size after Reparameterization, set up that approximate error on each node set is estimated and global optimum's contacting between estimating, by respectively estimating component at digraph
the number of times of middle appearance determines that perceptual error estimates global optimization approximate extents
.
In step of the present invention (6), under exemplary dynamic environment, algorithm validity experimental verification specifically comprises the steps:
1. the relevant perceptual error of yardstick that application proposes estimates modeling and approximate global optimization is theoretical calculates the allocation optimum of half tone image binarized pixel, to verify the validity of halftoning algorithm of proposition;
2. utilize the method for objectively evaluating of texture entropy, structural similarity and angle second moment under dynamic environment, relatively to evaluate the performance of digital picture shadow tone with LSMB method, and algorithm complex is analyzed.
Beneficial effect of the present invention:
(1) having solved existing method cannot take into account multiple dimensioned error and estimate interaction problems in the yardstick of information and between yardstick;
(2) utilize multiple dimensioned error of the present invention to estimate the digital halftone method of global optimization, program is simple, and processing speed is very fast, can obtain the laser plate-making image of high-quality.
Brief description of the drawings
Fig. 1 is process flow diagram of the present invention;
Fig. 2 is quaternary tree probability graph structural drawing;
The result that Fig. 3 LSMB shadow tone obtains;
The result that Fig. 4 the method for the invention obtains.
Embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in further details, but protection scope of the present invention is not limited to described content.
Embodiment 1
The digital picture halftoning method that described in the present embodiment, multiple dimensioned perceptual error is estimated approximate global optimization as shown in Figure 1, specifically comprises the steps:
(1) judge whether continuous toned image is 2 of standard
n× 2
nimage;
(2) construct multiple dimensioned perceptual error and estimate objective function model framework: specifically comprise the steps:
1. adopt the multiple dimensioned model of Haar two-dimensional discrete wavelet conversion Criterion continuous toned image in the L * a*b* color space, according to multiple dimensioned model definition wavelet field spatial point
the multiple dimensioned perceptual error at place is estimated information
for corresponding wavelet coefficient is at the Euclidean distance of the L*a*b* color space;
2. the space, frequency and the response of directivity characteristics CSF that adopt Nasanen human vision system model to determine multiple dimensioned perceptual error to estimate
, according to step 1. in multiple dimensioned perceptual error estimate information
respond with CSF
calculating multiple dimensioned perceptual error estimates with the convolution of CSF response and obtains at yardstick
direction
spatial point
locate multiple dimensioned perceptual error and estimate objective function
;
3. by general perceives error measure function
be defined as step
the multiple dimensioned perceptual error of middle diverse location, yardstick and direction is estimated objective function
all side and,
.
(3) set up the relevant perceptual error measure function of yardstick: specifically comprise the steps:
1. adopt digraph
characterize multiple dimensioned perceptual error and estimate objective function model, wherein
for node set, it is corresponding that each node and perceptual error are estimated,
for source point,
for meeting point,
for line set;
2. adopt cluster modeling in the yardstick that two component Gaussian mixture model estimate perceptual error: on yardstick and subband, add up independent at wavelet field wavelet coefficient, obtain wavelet coefficient set by wavelet coefficient probability distribution
joint probability density function:
, the probability density function of each wavelet coefficient
utilize
two component Gaussian mixture model
, wherein,
represent wavelet coefficient
corresponding hidden state variable
value
time probability mass function;
represent given hidden state variable
value
time wavelet coefficient
conditional probability density function, suppose that this conditional probability density function obeys zero-mean Gaussian distribution,
, wherein,
represent the variance of zero-mean Gaussian distribution,
, and indirectly reflect the size of wavelet coefficient amplitude with this side's extent; In the multi-scale image model of wavelet field, it is the wavelet coefficient of each yardstick, each subband
according to probability mass function
an associated hidden state variable
, this hidden variable has two state values, is respectively 0 and 1, and the height of reflection wavelet coefficient amplitude on this basis, is realized the interior dependence statistical modeling of yardstick of wavelet coefficient;
3. adopt HMM to quaternary tree across the modeling of yardstick continuation: utilize in wavelet field the father-sub-dependence of wavelet coefficient quaternary tree probability graph configuration model wavelet coefficient in subband, in the wavelet coefficient quaternary tree probability graph structure (as shown in Figure 2) in wavelet field in a subband, black node represents independent wavelet coefficient
, the white ode table on each black node left side is shown the hidden state variable of each wavelet coefficient association
, four hidden state variables of a corresponding next lower yardstick of hidden state variable, are connected by straight line between them, and straight line represents the father-sub-dependence of wavelet coefficient, across yardstick continuation; Represent a given specific wavelet coefficient across yardstick continuation, this coefficient itself may have identical hidden state tag with its father very much; Connect the hidden state variable that wavelet coefficient is corresponding with oriented Markov probability figure, to across the modeling of yardstick continuation; Introduce the hidden state transition probability along wavelet field quaternary tree graph model structure
, represent wavelet coefficient
the father of a upper yardstick
hidden state
for large or hour,
hidden state
for large or little probability, set up the hidden state of wavelet coefficient across yardstick dependence model; Hidden state when between father-son is identical,
time, expectation state transition probability
larger, set up thus across yardstick continuation model; On this basis, set up the single subband hidden Markov of wavelet field tree-model:
, wherein,
represent single wavelet sub-band and
,
with
for the value of the hidden state variable of wavelet coefficient and
,
the hidden state that represents father's node is
time child's node hidden state be
probability;
in will be 2. two component Gaussian mixture model and 3. in HMM model construction be that hidden Markov is set
, wherein
for node probability mass function,
for node
arrive
state transition probability,
for variance;
5. by digraph
class label
the relevant perceptual error measure function of yardstick be defined as
, wherein
with
represent respectively smoothness constraint and unusual constraint,
for smoothness constraint influence coefficient; Adopt the negative log-likelihood of continuous toned image L*a*b* color space brightness amplitude to represent smoothness constraint
, wherein state
, build unusual constraint with exponential function
, utilize smoothness constraint and unusual constraint to set up the relevant perceptual error measure function of yardstick.
(4) the relevant perceptual error of yardstick is estimated approximate global optimization strategy: specifically comprise the steps:
1. adopt Lagrangian transform method to realize the relevant perceptual error measure function of archeus and estimate and 1. adopt Lagrangian transform method to realize the relevant perceptual error measure function of archeus to the conversion of antithesis perceptual error measure function to antithesis perceptual error, and prove that antithesis perceptual error measure function is convex function;
2. adopt digraph
reparameterization method realizes the approximate global optimization of the relevant perceptual error measure function of yardstick, meeting under capacity and conservation constraint prerequisite, on the basis of residual capacity, increases and decreases constant flow, carries out residual plot Reparameterization;
3. adopt Graph Theory to prove terminal line and node line cubage correction theory: to pass through digraph
after Reparameterization
the relevant perceptual error measure function of yardstick
there is same item label
method, realizes the iteration optimization that perceptual error is estimated;
4. while reaching iterations or setting accuracy, if source point
with node
be connected,
label is 1; If meeting point
with node
be connected,
label is 0, formulates the approximate global optimization strategy of the relevant perceptual error measure function of yardstick, obtains the configuration of the approximate global optimum of half tone image binarized pixel.
Described in step (5), global optimization approximate extents determines that theory specifically comprises the steps:
1. adopt Graph Theory definition global optimization approximation coefficient and prove approximate extents, order
for digraph after Reparameterization
approximate global optimization class tag configurations,
for digraph
global optimum's class tag configurations, definition
cfor global optimization approximation coefficient, set coefficient
; By approximation coefficient
can determine that perceptual error estimates global optimization approximate extents
;
2. define digraph
global optimum's class tag configurations is
node set, and define respectively this node set inside, outside and borderline node set;
3. according to the definition of different label area node sets, determine each region perceptual error measure function size after Reparameterization, set up that approximate error on each node set is estimated and global optimum's contacting between estimating, by respectively estimating component at digraph
the number of times of middle appearance determines that perceptual error estimates global optimization approximate extents
.
In described step (6), under exemplary dynamic environment, algorithm validity experimental verification specifically comprises the steps:
1. the relevant perceptual error of yardstick that application proposes estimates modeling and approximate global optimization is theoretical calculates the allocation optimum of half tone image binarized pixel, to verify the validity of halftoning algorithm of proposition;
2. utilize the method for objectively evaluating of texture entropy, structural similarity and angle second moment under dynamic environment, relatively to evaluate the performance of digital picture shadow tone with LSMB method, and algorithm complex is analyzed.
The result that method described in the present embodiment is obtained is analyzed, as shown in Fig. 3 ~ 4, and the shadow tone quality that this patent keeps degree to obtain with the different half tune methods of smooth region artificial texture inhibition level two aspect rational evaluations from edge.Evaluating objective quality system is decomposed into two parts: the low frequency part evaluation of HFS evaluation (2) the image smoothing part at (1) image detail and edge.
Adopt gradient magnitude and (G) with texture entropy (Entropy Feature,
ent) evaluate edge penalty situation and details in high fdrequency component and keep level; The low frequency component of image has been concentrated most energy of image, and human eye vision is the degree of uniformity of intensity profile to the perception of this part; Application angle second moment (Angle Second Moment Feature,
asm) consider the smooth degree of low frequency residual plot; G is larger, and in image, marginal element is more, illustrates that halftone image edge penalty is larger; Otherwise halftone image edge keeps situation better; Texture entropy is less, illustrates that the quantity of information comprising in high frequency residual plot is fewer, and in half tone image, overall loss of detail is less, and edge keeps better; Angle second moment is larger, illustrates that intensity profile is more even, and the low-frequency image of shadow tone is more smooth.
From " laterally " of table 1, same scan image to be carried out, after different halftoning method processing, can finding, this patent method quality assessment data are better than Lsmb method; The evaluation method that has proved this patent proposition can be described shadow tone noise level rightly, provided half tone image at edge, the residual noise at details place changes and the uniform properties of smooth region, for objective description half-tone picture image quality provides effective foundation.
Table 1 half tone image Objective Quality Assessment result
Table 2 algorithm analysis result
Claims (6)
1. multiple dimensioned perceptual error is estimated a digital picture halftoning method for approximate global optimization, it is characterized in that: specifically comprise the steps:
(1) judge whether continuous toned image is 2 of standard
n× 2
nimage;
(2) construct multiple dimensioned perceptual error and estimate objective function model framework: include the standard continuous toned image information in frequency field and direction territory in multiple dimensioned perceptual error and estimate in object module framework, utilize two-dimensional discrete wavelet conversion to set up multiple dimensioned perceptual error and estimate objective function model framework;
(3) set up the relevant perceptual error measure function of yardstick: adopt cluster modeling in the yardstick that two component Gaussian mixture model estimate perceptual error, adopt HMM to quaternary tree across the modeling of yardstick continuation;
(4) the relevant perceptual error of yardstick is estimated approximate global optimization strategy: its antithesis perceptual error measure function of convex duality model conversation the Theory Construction that utilizes the relevant perceptual error measure function of the archeus described in step (3), calculate antithesis perceptual error by Reparameterization residual plot algorithm iteration and estimate the global optimum upper bound, realize maximum a posteriori probability configuration and the approximate global optimum of the relevant perceptual error measure function of archeus of half tone image binarized pixel;
(5) global optimization approximate extents is determined theoretical: utilize antithesis perceptual error described in step (4) to estimate the global optimum upper bound and define the binarized pixel allocation optimum in zones of different after each iteration, and then propose and prove global optimization approximation coefficient; By determining the approximation coefficient of the relevant perceptual error measure function global optimization of yardstick, multiple dimensioned perceptual error can be estimated to optimum results consistance and be controlled in definite known range;
(6) algorithm validity experimental verification under exemplary dynamic environment: the multiple dimensioned perceptual error proposing according to step (1) ~ (5) is estimated approximate global optimization approach framework, set up shadow tone computing platform, the in the situation that of illumination and background dynamics variation, the described relevant perceptual error of yardstick of applying step (3) is estimated modeling and the described approximate global optimization policy calculation half tone image binarized pixel allocation optimum of step (4), the validity of the halftoning algorithm proposing with checking, utilize texture entropy, the method for objectively evaluating of structural similarity and angle second moment is relatively evaluated the performance of digital picture shadow tone under dynamic environment with LSMB method, and algorithm complex is analyzed.
2. multiple dimensioned perceptual error according to claim 1 is estimated the digital picture halftoning method of approximate global optimization, it is characterized in that: described in step (2), construct multiple dimensioned perceptual error and estimate objective function model framework and specifically comprise the steps:
1. adopt the multiple dimensioned model of Haar two-dimensional discrete wavelet conversion Criterion continuous toned image in the L*a*b* color space, according to multiple dimensioned model definition wavelet field spatial point
the multiple dimensioned perceptual error at place is estimated information
for corresponding wavelet coefficient is at the Euclidean distance of the L*a*b* color space;
2. the space, frequency and the response of directivity characteristics CSF that adopt Nasanen human vision system model to determine multiple dimensioned perceptual error to estimate
, according to step 1. in multiple dimensioned perceptual error estimate information
respond with CSF
calculating multiple dimensioned perceptual error estimates with the convolution of CSF response and obtains at yardstick
direction
spatial point
locate multiple dimensioned perceptual error and estimate objective function
;
3. by general perceives error measure function
be defined as step
the multiple dimensioned perceptual error of middle diverse location, yardstick and direction is estimated objective function
all side and,
.
3. multiple dimensioned perceptual error according to claim 1 is estimated the digital picture halftoning method of approximate global optimization, it is characterized in that: described in step (3), set up the relevant perceptual error measure function of yardstick and specifically comprise the steps:
1. adopt digraph
characterize multiple dimensioned perceptual error and estimate objective function model, wherein
for node set, it is corresponding that each node and perceptual error are estimated,
for source point,
for meeting point,
for line set;
2. adopt cluster modeling in the yardstick that two component Gaussian mixture model estimate perceptual error: on yardstick and subband, add up independent at wavelet field wavelet coefficient, obtain wavelet coefficient set by wavelet coefficient probability distribution
joint probability density function:
, the probability density function of each wavelet coefficient
utilize
two component Gaussian mixture model
, wherein,
represent wavelet coefficient
corresponding hidden state variable
value
time probability mass function,
represent given hidden state variable
value
time wavelet coefficient
conditional probability density function, suppose that this conditional probability density function obeys zero-mean Gaussian distribution,
, wherein,
represent the variance of zero-mean Gaussian distribution,
, and indirectly reflect the size of wavelet coefficient amplitude with this side's extent; In the multi-scale image model of wavelet field, it is the wavelet coefficient of each yardstick, each subband
according to probability mass function
an associated hidden state variable
, this hidden variable has two state values, is respectively 0 and 1, and the height of reflection wavelet coefficient amplitude on this basis, is realized the interior dependence statistical modeling of yardstick of wavelet coefficient;
3. adopt HMM to quaternary tree across the modeling of yardstick continuation: application is along the hidden state transition probability of wavelet field quaternary tree graph model structure
represent wavelet coefficient
the father of a upper yardstick
hidden state
for large or hour,
hidden state
for large or little probability, set up the hidden state of wavelet coefficient across yardstick dependence model; Hidden state when between father-son is identical,
time, expectation state transition probability
for greatly, set up thus across yardstick continuation model; On this basis, set up the single subband hidden Markov of wavelet field tree-model:
, wherein,
represent single wavelet sub-band and
,
with
for the value of the hidden state variable of wavelet coefficient and
,
the hidden state that represents father's node is
time child's node hidden state be
probability;
by step 2. in two component Gaussian mixture model and 3. in HMM model construction be hidden Markov tree
, wherein
for node probability mass function,
for node
arrive
state transition probability,
for variance;
5. by digraph
class label
the relevant perceptual error measure function of yardstick be defined as
, wherein
with
represent respectively smoothness constraint and unusual constraint,
for smoothness constraint influence coefficient; Adopt the negative log-likelihood of continuous toned image L*a*b* color space brightness amplitude to represent smoothness constraint
, wherein state
, build unusual constraint with exponential function
, utilize smoothness constraint and unusual constraint to set up the relevant perceptual error measure function of yardstick.
4. multiple dimensioned perceptual error according to claim 1 is estimated the digital picture halftoning method of approximate global optimization, it is characterized in that: the relevant perceptual error of yardstick described in step (4) is estimated approximate global optimization strategy and specifically comprised the steps:
1. adopt Lagrangian transform method to realize the relevant perceptual error measure function of archeus and transform to antithesis perceptual error measure function, and prove that antithesis perceptual error measure function is convex function;
2. adopt digraph
reparameterization method realizes the approximate global optimization of the relevant perceptual error measure function of yardstick, meeting under capacity and conservation constraint prerequisite, on the basis of residual capacity, increases and decreases constant flow, carries out residual plot Reparameterization;
3. adopt Graph Theory to prove terminal line and node line cubage correction theory: to pass through digraph
after Reparameterization
the relevant perceptual error measure function of yardstick
there is same item label
method, realizes the iteration optimization that perceptual error is estimated;
4. while reaching iterations or setting accuracy, if source point
with node
be connected,
label is 1; If meeting point
with node
be connected,
label is 0, formulates the approximate global optimization strategy of the relevant perceptual error measure function of yardstick, obtains the configuration of the approximate global optimum of half tone image binarized pixel.
5. multiple dimensioned perceptual error according to claim 1 is estimated the digital picture halftoning method of approximate global optimization, it is characterized in that: described in step (5), global optimization approximate extents determines that theory specifically comprises the steps:
1. adopt Graph Theory definition global optimization approximation coefficient and prove approximate extents, order
for digraph after Reparameterization
approximate global optimization class tag configurations,
for digraph
global optimum's class tag configurations, definition
cfor global optimization approximation coefficient, set coefficient
; By approximation coefficient
can determine that perceptual error estimates global optimization approximate extents
;
2. define digraph
global optimum's class tag configurations is
node set, and define respectively this node set inside, outside and borderline node set;
3. according to the definition of different label area node sets, determine each region perceptual error measure function size after Reparameterization, set up that approximate error on each node set is estimated and global optimum's contacting between estimating, by respectively estimating component at digraph
the number of times of middle appearance determines that perceptual error estimates global optimization approximate extents
.
6. multiple dimensioned perceptual error according to claim 1 is estimated the digital picture halftoning method of approximate global optimization, it is characterized in that: described in step (6), under exemplary dynamic environment, algorithm validity experimental verification specifically comprises the steps:
1. the relevant perceptual error of yardstick that application proposes estimates modeling and approximate global optimization is theoretical calculates the allocation optimum of half tone image binarized pixel, to verify the validity of halftoning algorithm of proposition;
2. utilize the method for objectively evaluating of texture entropy, structural similarity and angle second moment under dynamic environment, relatively to evaluate the performance of digital picture shadow tone with LSMB method, and algorithm complex is analyzed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410152246.0A CN103955908B (en) | 2014-04-16 | 2014-04-16 | Digital image halftone method based on multi-scale perception error measure approximation global optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410152246.0A CN103955908B (en) | 2014-04-16 | 2014-04-16 | Digital image halftone method based on multi-scale perception error measure approximation global optimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103955908A true CN103955908A (en) | 2014-07-30 |
| CN103955908B CN103955908B (en) | 2017-01-11 |
Family
ID=51333177
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410152246.0A Expired - Fee Related CN103955908B (en) | 2014-04-16 | 2014-04-16 | Digital image halftone method based on multi-scale perception error measure approximation global optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103955908B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921886A (en) * | 2018-06-11 | 2018-11-30 | 昆明理工大学 | A kind of texture information fusion Multi-scale model forest digital picture halftoning method |
| CN115004256A (en) * | 2020-03-03 | 2022-09-02 | 赫尔实验室有限公司 | Perceptual adjustment based on contrast and entropy using optimization based on probability signal temporal logic |
| CN112199689B (en) * | 2020-09-07 | 2024-05-03 | 宁波创源文化发展股份有限公司 | Halftone information hiding and identifying method based on mobile terminal |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2402009A (en) * | 2003-05-20 | 2004-11-24 | Software 2000 Ltd | Halftone bit mask generation for reduction of contouring |
| CN101600039A (en) * | 2008-06-05 | 2009-12-09 | 佳世达科技股份有限公司 | The method of half tone image conversion method, Method of printing and generation halftone shield |
-
2014
- 2014-04-16 CN CN201410152246.0A patent/CN103955908B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2402009A (en) * | 2003-05-20 | 2004-11-24 | Software 2000 Ltd | Halftone bit mask generation for reduction of contouring |
| CN101600039A (en) * | 2008-06-05 | 2009-12-09 | 佳世达科技股份有限公司 | The method of half tone image conversion method, Method of printing and generation halftone shield |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921886A (en) * | 2018-06-11 | 2018-11-30 | 昆明理工大学 | A kind of texture information fusion Multi-scale model forest digital picture halftoning method |
| CN108921886B (en) * | 2018-06-11 | 2021-09-14 | 昆明理工大学 | Texture information fusion multi-scale structured forest digital image halftone method |
| CN115004256A (en) * | 2020-03-03 | 2022-09-02 | 赫尔实验室有限公司 | Perceptual adjustment based on contrast and entropy using optimization based on probability signal temporal logic |
| CN112199689B (en) * | 2020-09-07 | 2024-05-03 | 宁波创源文化发展股份有限公司 | Halftone information hiding and identifying method based on mobile terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103955908B (en) | 2017-01-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108171762B (en) | Deep learning compressed sensing same-class image rapid reconstruction system and method | |
| Liu et al. | Learning converged propagations with deep prior ensemble for image enhancement | |
| CN113658051B (en) | Image defogging method and system based on cyclic generation countermeasure network | |
| CN110992275B (en) | Refined single image rain removing method based on generation of countermeasure network | |
| Yang et al. | Fast low-rank Bayesian matrix completion with hierarchical Gaussian prior models | |
| CN101882304B (en) | Self-adaptive de-noising and characteristic enhancing method of SAR (Synthetic Aperture Radar) image | |
| US7536064B2 (en) | Image comparison by metric embeddings | |
| Wang et al. | Robust locally linear analysis with applications to image denoising and blind inpainting | |
| JP2018169672A (en) | Method of creating teacher image, computer and program | |
| CN106204482B (en) | Based on the mixed noise minimizing technology that weighting is sparse | |
| CN111179189B (en) | Image processing method and device based on generation of countermeasure network GAN, electronic equipment and storage medium | |
| Dai et al. | Hidden talents of the variational autoencoder | |
| CN108932536A (en) | Human face posture method for reconstructing based on deep neural network | |
| CN108197669B (en) | Feature training method and device of convolutional neural network | |
| CN111047543A (en) | Image enhancement method, device and storage medium | |
| CN104021536B (en) | A kind of adaptive SAR image and Multispectral Image Fusion Methods | |
| CN102622729B (en) | Spatial self-adaptive block-matching image denoising method based on fuzzy set theory | |
| CN102567973A (en) | Image denoising method based on improved shape self-adaptive window | |
| Goyal et al. | Clustered-dot halftoning with direct binary search | |
| CN107133923A (en) | A kind of blurred picture non-blind deblurring method based on self-adaption gradient sparse model | |
| CN106663315A (en) | Method for denoising noisy image | |
| CN103955908A (en) | Digital image halftone method based on multi-scale perception error measure approximation global optimization | |
| Jiang et al. | Layer-wise deep neural network pruning via iteratively reweighted optimization | |
| CN104680549B (en) | SAR image change detection based on higher order neighborhood TMF models | |
| CN103793897B (en) | A kind of digital picture halftoning method based on wavelet multi-scale information fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170111 |