CN108921886A - A kind of texture information fusion Multi-scale model forest digital picture halftoning method - Google Patents
A kind of texture information fusion Multi-scale model forest digital picture halftoning method Download PDFInfo
- Publication number
- CN108921886A CN108921886A CN201810594811.7A CN201810594811A CN108921886A CN 108921886 A CN108921886 A CN 108921886A CN 201810594811 A CN201810594811 A CN 201810594811A CN 108921886 A CN108921886 A CN 108921886A
- Authority
- CN
- China
- Prior art keywords
- image
- texture
- model
- forest
- threshold
- 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
- 238000000034 method Methods 0.000 title claims abstract description 59
- 230000004927 fusion Effects 0.000 title abstract description 3
- 238000005192 partition Methods 0.000 claims abstract description 22
- 230000000007 visual effect Effects 0.000 claims abstract description 19
- 238000003708 edge detection Methods 0.000 claims abstract description 16
- 230000003044 adaptive effect Effects 0.000 claims abstract description 7
- 238000009792 diffusion process Methods 0.000 claims description 45
- 238000003066 decision tree Methods 0.000 claims description 12
- 238000001914 filtration Methods 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000004438 eyesight Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000013139 quantization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 claims description 2
- 231100000656 threshold model Toxicity 0.000 description 27
- 230000000694 effects Effects 0.000 description 14
- 241000282537 Mandrillus sphinx Species 0.000 description 11
- 230000003595 spectral effect Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 239000000523 sample Substances 0.000 description 5
- 238000001228 spectrum Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 238000004445 quantitative analysis Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000004451 qualitative analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- PLXMOAALOJOTIY-FPTXNFDTSA-N Aesculin Natural products OC[C@@H]1[C@@H](O)[C@H](O)[C@@H](O)[C@H](O)[C@H]1Oc2cc3C=CC(=O)Oc3cc2O PLXMOAALOJOTIY-FPTXNFDTSA-N 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000035807 sensation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
- G06T7/44—Analysis of texture based on statistical description of texture using image operators, e.g. filters, edge density metrics or local histograms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20192—Edge enhancement; Edge preservation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of texture information fusion Multi-scale model forest digital picture halftoning method, including continuous toned image is converted into multi-grey image；Multi-grey image plus channel operation are made it have into three channels, multi-scale morphology is carried out to the multi-grey image after adding channel using the edge detection method of structuring forest；In conjunction with human visual system to the difference of original image local grain information sensing degree, the texture partition model of multi-grey image is established；The marginal information of multi-scale morphology image is extracted by adaptive threshold；The threshold model that texture structure is combined with noise modulated is established in conjunction with picture edge characteristic by texture partition model；Digital halftone is carried out to multi-grey image using threshold model.The half tone image visual deformation that the present invention obtains is reduced, and clarity and contrast are high, and the edge and details position of image are clear, complete, and whole visual effect is good, and have good blue noise characteristic and good versatility.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of half tone image visual deformation reduction, clarity
And contrast is high, the edge and details position of image are clear, complete, and whole visual effect is good, and have good blue noise
The texture information of characteristic and good versatility merges Multi-scale model forest digital picture halftoning method.
Background technique
Digital halftone technology is exactly the two-values such as laser printer and laser platemaker seen in similar daily life
Or in polychrome two-value equipment, the vision low-pass characteristic of human eye is simulated to copy the variation of the contrast of former multi-grey image, it will be former continuous
The technology for the two-value half tone image changed the line map as being converted into suitable eye-observation.Digital halftone technology not only affects laser carving
Carve and Laser Printed Image output quality, and with data information hide and digital watermark technology be also it is closely related, together
When with the compression of medical image, image storage with the fields such as to transmit also inseparable.
Currently, digital halftone technology is generally divided into " error-diffusion method ", " dithering " and " iteration optimization method " three
Major class.Numerous researchers both domestic and external in recent years to the improvement of halftoning method all by center of gravity be placed on error diffusion method and
With iterative optimization method.For digital halftoning method overall study situation, the digital halftone side based on iteration optimization
Although method be in the quality of the half tone image of generation it is best, efficiency is too low, and operating cost is higher, it is difficult to meet life
The practical property demand produced.And error diffusion method is the amount by then will obtain to a certain processes pixel in gray level image
Change error and passes to the untreated pixel of the neighborhood of pixels according to certain weight coefficient to carry out error diffusion.Error
The effect for the bianry image that method of diffusion method obtains has a big promotion, and high-efficient, and operating cost is low, but generate
Half tone image quality has two o'clock problem to overcome completely not yet, i.e.,：" in the smooth gradual change area in the half tone image of generation
Undesired texture completely eliminates not yet ", and " phenomena such as fuzzy, discontinuous spy is easy to appear in the marginal portion of image
It is not for the image method with complex edge ".
Classical error diffusion method surely belongs to Floyd-Steinberg error diffusion method（FLOYED R W,
STEINBERG L.An adaptive algorithm for spatial grey scale.Society for
Information Display,1976,17(2):75-77）（Hereinafter referred to as F-S error diffusion method）Although F-S error expands
The half tone image general effect for dissipating method generation is preferable, and efficiency is also high；But this method does not go to consider image well
Edge details characteristic and the half tone image of generation " worm effect " are more obvious.In addition, in view of the above deficiencies, Shiau-
Fan（JENG-NAN SHIAU ,FAN Z. Set of easily implementable coefficients in error
diffusion with reduced worm artifacts. Proceedings of SPIE-The International
Society for Optical Engineering ,1996,2658:222-225）It is proposed will be in F-S error diffusion method
135 ° of support legs of error filtering core weighting coefficient rotate clockwise 45 °, are overlapped with horizontal line and symmetry is presented（Following letter
Claim S-F error diffusion method）.So that the error return range of S-F error diffusion method is bigger, it can be in broader angle model
It works in enclosing, plays the role of reducing beam-shaping effect, worm effect can be effectively reduced.But S-F error diffused sheet
Method without well consider image edge details characteristic so that half tone image generated is inadequate at edge and detail section
Clearly.In addition, also proposed in the prior art, such as " noise modulated error diffusion method ", " edge B.W.Hwang enhances error
Method of diffusion "（Hwang B W,Kang T H,Lee T S.Distortion-Free of General Information
with Edge Enhanced Error Diffusion Halftoning.Computational Science and ITS
Applicati- ons,International Conference(DBLP2004).Italy,2004）The methods of, although one
Determining, " blur margin clear " and " undesired texture excessively obvious " are solved the problems, such as in half tone image in degree, but for
For edge and the more complicated image of texture structure, still there can be some tiny edges and some important details exist
The case where disappearing during halftone, and both methods does not go fully to consider human visual system to texture in image
The sensitivity in region.
Summary of the invention
The purpose of the present invention is to provide a kind of reduction of half tone image visual deformation, clarity and contrast are high, image
Edge and details position it is clear, complete, whole visual effect is good, and has good blue noise characteristic and good general
Property texture information merge Multi-scale model forest digital picture halftoning method.
What the object of the invention was realized in, including image conversion, edge detection, establish texture partition model, extract side
Edge information establishes threshold model, halftone step, specifically includes：
A, image is converted：Continuous toned image is converted into the pixel of multi-grey image；
B, edge detection：The multi-grey image that step A is obtained carries out plus channel operation, makes it have three channels, then sharp
With the edge detection method of structuring forest, to tool, there are three the multi-grey images in channel to carry out multi-scale morphology；
C, texture partition model is established：In conjunction with human visual system to the difference of original image local grain information sensing degree, build
The texture partition model for the multi-grey image that vertical step A obtains；
D, marginal information is extracted：The final marginal information of image of step B multi-scale morphology is extracted by adaptive threshold,
Obtain picture edge characteristic；
E, threshold model is established：It is built by the texture partition model that step C is established in conjunction with the picture edge characteristic that D step obtains
The threshold model that vertical texture structure is combined with noise modulated；
F, halftone：The threshold model established using E step carries out digital halftone to the multi-grey image of step A, defeated
The pixel binary value of half tone image is corresponded to outb(m, n)。
The present invention uses the rapid edge-detection method based on structuring forest（Dollar P ,Zitnick C L.Fast
Edge Detection Using Structured Forests. IEEE Transactions on Pattern
Analysis & Machine Intelligence,2015,37(8):1558-1570）Multiple dimensioned edge is carried out to original image
Information extraction, and adaptive threshold fuzziness is carried out to extracted marginal information and obtains fine binary edge map；Then sharp
Texture partition model is established with difference of the human visual system to original image local grain information sensing degree, finally utilizes S-F
The error filtering core addition white noise generator of error diffusion method is in the way of raster scanning, to smooth in multi-grey image
Area carries out digital halftone, and establishes one in conjunction with its texture feedback information in " slight texture area " and " severe texture area "
The threshold model that texture structure is combined with noise modulated, then with threshold model to " the slight texture area " in multi-grey image
" severe texture area " carries out digital halftone, ultimately generates digital halftone image.The qualitative analysis of experimental result shows
The half tone image visual deformation that the present invention obtains is reduced, and halftoning reproduces image and the degree of closeness of original image is higher, can
It is effectively retained the edge and details position of image, clarity increases and contrast improves, more shadow details can be shown,
Visual effect is preferable, has good blue noise characteristic and versatility.The quantitative analysis of experimental result shows the present invention and existing
After four kinds of methods in technology carry out halftone to gray level image respectively, PSNR value increases 1.92~5.53dB, MSSIM value
Increase 0.19~0.96.
Detailed description of the invention
Fig. 1 is frame diagram of the invention；
Fig. 2 is flow diagram of the invention；
Fig. 3 is experimental example clock original image；
Fig. 4 is the half tone image that experimental example clock is generated through F-S error diffusion method；
Fig. 5 is the half tone image that experimental example clock is generated through S-F error diffusion method；
Fig. 6 is the half tone image that experimental example clock is generated through noise modulated error diffusion method；
Fig. 7 is the half tone image that experimental example clock enhances that error diffusion method is generated through the edge B.W.Hwang；
Fig. 8 is the half tone image that clock is generated through the present invention in experimental example；
Fig. 9 is experimental example goldhill original image；
Figure 10 is the half tone image that experimental example goldhill is generated through F-S error diffusion method；
Figure 11 is the half tone image that experimental example goldhill is generated through S-F error diffusion method；
Figure 12 is the half tone image that experimental example goldhill is generated through noise modulated error diffusion method；
Figure 13 is the half tone image that experimental example goldhill enhances that error diffusion method is generated through the edge B.W.Hwang；
Figure 14 is the half tone image that goldhill is generated through the present invention in experimental example；
Figure 15 is experimental example mandrill original image；
Figure 16 is the half tone image that experimental example mandrill is generated through F-S error diffusion method；
Figure 17 is the half tone image that experimental example mandrill is generated through S-F error diffusion method；
Figure 18 is the half tone image that experimental example mandrill is generated through noise modulated error diffusion method；
Figure 19 is the half tone image that experimental example mandrill enhances that error diffusion method is generated through the edge B.W.Hwang；
Figure 20 is the half tone image that mandrill is generated through the present invention in experimental example；
Figure 21 is comparison of the 40 width standard pictures about PSNR parameter in experimental example；
Figure 22 is comparison of the 40 width standard pictures about MSSIM parameter in experimental example；
Figure 23 is the half tone image that 1/32 uniform gray image block is generated through S-F error diffusion method in embodiment；
Figure 24 is half color that 1/32 uniform gray image block enhances error diffusion method generation through the edge B.W.Hwang in embodiment
It changes the line map picture；
Figure 25 is the half tone image that 1/32 uniform gray image block is generated through the present invention in embodiment；
Figure 26 is that 1/32 uniform gray image block exports the corresponding RAPSD of result in embodiment.
Specific embodiment
The present invention will be further described below with reference to the accompanying drawings and embodiments, but is not subject in any way to the present invention
Limitation, according to the teachings of the present invention made any change or replacement, all belong to the scope of protection of the present invention.
The present invention includes image conversion, edge detection, establishes texture partition model, extract marginal information, establish threshold value mould
Type, halftone step, specifically include：
A, image is converted：Continuous toned image is converted into the pixel of multi-grey image；
B, edge detection：The multi-grey image that step A is obtained carries out plus channel operation, makes it have three channels, then sharp
With the edge detection method of structuring forest, to tool, there are three the multi-grey images in channel to carry out multi-scale morphology；
C, texture partition model is established：In conjunction with human visual system to the difference of original image local grain information sensing degree, build
The texture partition model for the multi-grey image that vertical step A obtains；
D, marginal information is extracted：The final marginal information of image of step B multi-scale morphology is extracted by adaptive threshold,
Obtain picture edge characteristic；
E, threshold model is established：It is built by the texture partition model that step C is established in conjunction with the picture edge characteristic that D step obtains
The threshold model that vertical texture structure is combined with noise modulated；
F, halftone：The threshold model established using E step carries out digital halftone to the multi-grey image of step A, defeated
The pixel binary value of half tone image is corresponded to outb(m, n)。
The edge detection method of structuring forest carries out multi-scale morphology to multi-grey image in the step B, is
To multi-grey image and on multi-grey image respectively with 2 for sample rate down-sampled image and up-sampling image construction three
The image of a scale detects edge respectively, and the image detection result of three scales above-mentioned is then carried out average value solution.
In the step B edge detection method of structuring forest extract first plus channel operation after multi-grey image it
Then feature vector carries out ballot to feature vector by structuring forest classified device and determines its classification, end-results forest
The two-value decision probability of final classification is obtained by following formula：
,
Wherein：NFor decision tree number independent in structuring forest,For the classification two of decision tree in structuring forest
Value determines.
The decision tree of structuring forest in the step Bf t (x)By recursive mode to the left or to inferior division tree come point
Class sample, until reaching leaf node, decision treef t (x)In each nodejAll letter is divided with the binary of a following formula
NumberIt is associated：
,
Wherein：If, then nodejIt sends to the leftx, otherwise to the right, process terminates at leaf node.
Established in the step C multi-grey image texture partition model include it is following step by step：
C1, average gray value is calculated：Calculate the average ash of 3 × 3 image local areas in the multi-grey image that step A is obtained
Angle value：
,
C2, computation vision deviation：The local visual deviation of current pixel is calculated as follows：
,
Wherein：For the weight with directional correlation；
C3, partition model divide：
IfWhen, the regional area of present image is smooth area；
IfWhen, the regional area of present image is slight texture area；
IfWhen, the regional area of present image is severe texture area.
In the C2 stepBoth horizontally and vertically weight be 0.1465, diagonal weight is
0.1035。
White noise random generator is added using the error filtering core in S-F error diffusion method in the E stepN r Come
Establish smooth area threshold model；And slight texture area and severe texture area are then first respectively corresponded and calculate slight texture feedback letter
Then breath, severe texture feedback information establish slight texture feedback threshold using the error filtering core in S-F error diffusion method
Model and severe texture feedback threshold model.
The smooth area threshold model is as follows：
,
Wherein：u(m,n) it is quantization error,Q[ ] is threshold quantizer,TFor threshold value andT=128。
Described slight and severe texture area feedback threshold model by it is following step by step：
E1, slight texture feedback information is calculated as follows：
,
Wherein：M(m,n) estimate for the spatiality of current pixel,αFor slight texture feedback information impact factor andα=4.3；
E2：Severe texture feedback information is calculated as follows：
,
Wherein：T(m,n) it is the picture edge characteristic that height texture area is obtained through D step,β=0.45；
E3, feedback threshold model：
Slight texture area threshold model：；
Severe texture area threshold model：；
Wherein：TFor threshold value andT=128。
The smooth area in image is handled in the way of raster scanning using smooth area threshold model in the F-step, is utilized
Slight texture area in slight texture area threshold model processing image, and using in severe texture area threshold model processing image
Severe texture area.
Experimental example：
1, experiment condition
In order to verify effectiveness of the invention, experiment uses and is configured to 2.56GHz CPU, 4G memory, 64 win7 notebooks
Computer, used software are Matlab2014a, based on realizing compliance test result of the invention with upper mounting plate.
2, experimentation of the present invention
2.1 are converted into continuous toned image the pixel of multi-grey image.
2.2 carry out multi-grey image to add channel operation, and forming it into tool, there are three the multi-grey images in channel, more
Down-sampled image and up-sampling image are formed for sample rate with 2 respectively on gray level image, constitutes three scales with multi-grey image
Image, then extract plus channel operation after three scales multi-grey image feature vector, pass through structuring forest classified
Device carries out ballot to feature vector and determines its classification, and end-results forest is determined general by the two-value that following formula obtains final classification
Rate：
,
Wherein：NIt is that the classification two-value of decision tree in structuring forest determines for decision tree number independent in structuring forest；
The decision tree of structuring forestf t (x)By recursive mode to the left or to inferior division tree come classification samples, directly
To arrival leaf node, decision treef t (x)In each nodejAll function is divided with the binary of a following formulaIt is associated：
,
Wherein：If, then nodejIt sends to the leftx, otherwise to the right, process terminates at leaf node.
2.3 combine human visual systems to the difference of original image local grain information sensing degree, by it is following step by step
Establish the texture partition model of multi-grey image：
2.3.1 the average gray value of 3 × 3 image local areas in multi-grey image is calculated：
；
2.3.2 the local visual deviation of current pixel is calculated as follows：
,
Wherein：For the weight with directional correlation,Both horizontally and vertically weight be 0.1465, diagonally
Line directional weighting is 0.1035；
2.3.3 partition model divides：
IfWhen, the regional area of present image is smooth area；
IfWhen, the regional area of present image is slight texture area；
IfWhen, the regional area of present image is severe texture area.
The 2.4 final marginal informations of image for extracting multi-scale morphology in above-mentioned 2.2 by adaptive threshold, obtain
To picture edge characteristic；
2.5 by the above-mentioned 2.3 texture partition models established, and the picture edge characteristic obtained in conjunction with above-mentioned 2.4 is missed using S-F
Error filtering core in poor method of diffusion adds white noise random generatorN r To establish smooth area threshold model：
,
Wherein：u(m,n) it is quantization error,Q[ ] is threshold quantizer,TFor threshold value andT=128；
And slight texture area and severe texture area are then first respectively corresponded and calculate slight texture feedback information, severe texture feedback
Then information establishes slight and severe texture feedback by following using the error filtering core in S-F error diffusion method step by step
Threshold model：
2.5.1 slight texture feedback information is calculated as follows：
,
Wherein：M(m,n) estimate for the spatiality of current pixel,αFor slight texture feedback information impact factor andα=4.3；
2.5.2 severe texture feedback information is calculated as follows：
,
Wherein：T(m,n) it is the picture edge characteristic that height texture area is obtained through D step,β=0.45；
2.5.3 feedback threshold model：
Slight texture area threshold model：；
Severe texture area threshold model：；
Wherein：TFor threshold value andT=128。
The 2.6 smooth area threshold models established using above-mentioned 2.5 are in the way of raster scanning to flat in multi-grey image
Skating area carries out digital halftone, carries out number to the slight texture area in multi-grey image using slight texture area threshold model
Halftone, and digital halftone is carried out to the severe texture area in multi-grey image using severe texture area threshold model
Change, be finally synthesizing the digital halftone image of above-mentioned smooth area, slight texture area and severe texture area, and export synthesis after
The pixel binary value of digital halftone imageb(m, n)。
3, qualitative analysis
In order to illustrate actual effect of the invention, the image and image texture that existing flat region textured area again has been respectively adopted are believed
The tri- width image of clock, goldhill, mandrill of complexity from simple to complex is ceased as experimental image, visible sensation distance
It is 24 inches.Five kinds of methods of Experimental comparison：F-S error diffusion method, the improved S-F error diffusion method, noise of classics
Modulation error method of diffusion, the edge B.W.Hwang enhancing error diffusion method and the present invention.The experiment effect of clock image point
Not such as Fig. 3 ~ 8, the experiment effect of goldhill image is respectively such as Fig. 9 ~ 14, and the experiment effect of mandrill image is respectively such as Figure 15
~20。
By taking first group of image as an example, all images in Fig. 3 ~ 8 are examined, it is seen that the boxed area in Fig. 4,
Namely on dial（That is the gray scale flat region of image）Number is smudgy and digital edge details between dial plate
It is excessively fuzzy；Although Fig. 5 is greatly reduced compared to " worm " effect in Fig. 4, but picture quality is not still especially desirable；Figure
6 because introduce noise, and image blurring is more serious, and loss in detail is serious；Texture structure in Fig. 7 is relatively clear, edge
Also more prominent；However, Fig. 8 is compared to for other four width images, the reproduction effects at texture and edge are the clearest, prominent in image
Out, artificial property texture is minimum and has good visual perception effect.
4, quantitative analysis
The Y-PSNR (PSNR) and the image based on structural similarity that table 1, table 2, table 3 list above three groups of images respectively
Quality evaluation parameter (MSSIM).MSSIM is to keep a measurement of degree to refer to about structural information between original image and half-tone picture
Mark, value are bigger, then it represents that half tone image quality reproduction is better.Observation table 1 ~ 3 can be seen that the present invention compared to other four kinds of sides
Method all has higher value.
The parameter value of 1 five kinds of methods of table（Clock figure）
The parameter value of 2 five kinds of methods of table（Goldhill figure）
The parameter value of 3 five kinds of methods of table（Mandrill figure）
5, versatility is verified
In order to further verify versatility of the invention, to existing flat region in standard image data library and 40 of textured area
Standard picture is tested with above five kinds of methods respectively, and calculates S-F error diffusion method, the increasing of the edge B.W.Hwang
The ratio of strong error diffusion method and PSNR of the invention and MSSIM, as a result respectively as shown in figure 21 and figure.
Figure 21 and Figure 22 is observed, shows that the present invention is higher than S-F on the two quantitative assessing index of PSNR and MSSIM and misses
Poor method of diffusion and the edge B.W.Hwang enhance error diffusion method, and therefore, the present invention has certain versatility.
6, average power spectral density
Halftone process will reproduce continuous tune not for the purpose of all frequency contents for accurately reproducing continuous toned image
The overall picture and details of image.The overall picture of continuous toned image is mainly determined by low frequency and intermediate frequency component, and details then depends on height
Frequency ingredient.In Fourier, the power spectral density of the half tone image obtained by error diffusion method can be by for spectrum
Estimation divides.Research has shown that the value of spectrum estimation be equal to its desired value, it be byKThe average of a cyclic graph is formed, sample
This variance is shown below：
It is two-dimensional function, probes intoThree-dimensional figure the anisotropic character of half tone image can be determined
Spectrum domain can be resolved into parameter more quantitative in the method description spectrum of annulus by the observation of amount ground and description, pass through ring width, radial frequency
Rate and frequency samples these three types attribute are come the characteristics of describing annulus.Radial average power spectral density is defined by Ulchney
（RAPSD）, expression formula is as follows：
Therefore, the present invention is analyzed to the effect of two-value halftone process reconstruct continuous toned image in order to preferably estimate from frequency domain
Fruit calculates the obtained halftoning of uniform gray image that the highest three kinds of methods of the above number of quantitative analysis are 1/32 to gray level
Export the radial average power spectral density (RAPSD) of result.The halftoning output for the gray level image that gray level is 1/32 such as Figure 23 ~
Shown in 25；Corresponding radial direction average power spectral density is as shown in figure 26.
It is found that the distribution at the obtained half tone image midpoint of the present invention is relative to other two methods from Figure 23 ~ 25
For the most uniformly.As can be seen from Figure 26, the corresponding radial average power spectral density of present invention gained half tone image exists
Low frequency part and high frequency section are respectively provided with the characteristic of precipitous cutoff frequency response characteristic and flat stable, therefore institute of the present invention
It obtains image and is closer to blue noise characteristic.
7, experiment analysis results
7.1 present invention and after remaining four kinds of method carries out halftoning respectively to gray level image, PSNR value increases 1.92~
5.53dB, MSSIM value increase 0.19~0.96.Illustrate that the resulting half tone image visual deformation of the present invention is reduced, vision effect
Fruit is good, and halftoning reproduces image and the degree of closeness of original image is higher.
7.2 clarity increase and contrast improves, especially at the box mark in clock figure, in Mandrill figure
Edge junction also observes that at the lines of face and in goldhill figure between house and house.
7.3 can show more shadow details, as the left side of Mandrill shows at the hair of black block.
7.4 methods have good versatility, by carrying out halftoning respectively to 40 standard pictures in image library
Change, the PSNR value and MSSIM value of the obtained half tone image of the present invention are all highest.
7.5 have good blue noise characteristic, pass through the uniform gray image progress halftone for being 1/32 to gray level
Afterwards it is found that the black and white two o'clock in half tone image is uniformly distributed.From corresponding radial average power spectral density figure it is found that the present invention
The characteristic of precipitous cutoff frequency response characteristic and flat stable is respectively provided in low frequency part and high frequency section.
Claims (10)
1. a kind of texture information merges Multi-scale model forest digital picture halftoning method, it is characterised in that turn including image
It changes, edge detection, establishes texture partition model, extract marginal information, establish threshold model, halftone step, specifically include：
A, image is converted：Continuous toned image is converted into the pixel of multi-grey image；
B, edge detection：The multi-grey image that step A is obtained carries out plus channel operation, makes it have three channels, then sharp
With the edge detection method of structuring forest, to tool, there are three the multi-grey images in channel to carry out multi-scale morphology；
C, texture partition model is established：In conjunction with human visual system to the difference of original image local grain information sensing degree, build
The texture partition model for the multi-grey image that vertical step A obtains；
D, marginal information is extracted：The final marginal information of image of step B multi-scale morphology is extracted by adaptive threshold,
Obtain picture edge characteristic；
E, threshold model is established：It is built by the texture partition model that step C is established in conjunction with the picture edge characteristic that D step obtains
The threshold model that vertical texture structure is combined with noise modulated；
F, halftone：The threshold model established using E step carries out digital halftone to the multi-grey image of step A, defeated
The pixel binary value of half tone image is corresponded to outb(m, n)。
2. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 1
It is that the edge detection method of structuring forest in the step B carries out multi-scale morphology to multi-grey image, is to more
Gray level image and the down-sampled image and three rulers of up-sampling image construction for being sample rate respectively on multi-grey image with 2
The image of degree detects edge respectively, and the image detection result of three scales above-mentioned is then carried out average value solution.
3. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 2
It is that the edge detection method of structuring forest in the step B extracts the feature of the multi-grey image after adding channel operation first
Then vector carries out ballot to feature vector by structuring forest classified device and determines its classification, end-results forest passes through
Following formula obtains the two-value decision probability of final classification：
,
Wherein：NFor decision tree number independent in structuring forest,For the classification two of decision tree in structuring forest
Value determines.
4. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 3
It is the decision tree of structuring forest in the step Bf t (x)By recursive mode to the left or to inferior division tree come sample of classifying
This, until reaching leaf node, decision treef t (x)In each nodejAll function is divided with the binary of a following formulaIt is associated：
,
Wherein：If, then nodejIt sends to the leftx, otherwise to the right, process terminates at leaf node.
5. according to claim 1, the texture information of 2,3 or 4 merges Multi-scale model forest digital picture halftoning method,
It is characterized in that the texture partition model that multi-grey image is established in the step C include it is following step by step：
C1, average gray value is calculated：Calculate the average ash of 3 × 3 image local areas in the multi-grey image that step A is obtained
Angle value：
,
C2, computation vision deviation：The local visual deviation of current pixel is calculated as follows：
,
Wherein：For the weight with directional correlation；
C3, partition model divide：
IfWhen, the regional area of present image is smooth area；
IfWhen, the regional area of present image is slight texture area；
IfWhen, the regional area of present image is severe texture area.
6. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 5
It is in the C2 stepBoth horizontally and vertically weight be 0.1465, diagonal weight be 0.1035.
7. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 5
It is the error filtering core addition white noise random generator in the E step using S-F error diffusion methodN r It is flat to establish
Skating area threshold model；And slight texture area and severe texture area are then first respectively corresponded and calculate slight texture feedback information, again
Spend texture feedback information, then using S-F error diffusion method error filtering core establish slight texture feedback threshold model and
Severe texture feedback threshold model.
8. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 7
It is that the smooth area threshold model is as follows：
,
Wherein：u(m,n) it is quantization error,Q[ ] is threshold quantizer,TFor threshold value andT=128。
9. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 8
Be described slight and severe texture area feedback threshold model by it is following step by step：
E1, slight texture feedback information is calculated as follows：
,
Wherein：M(m,n) estimate for the spatiality of current pixel,αFor slight texture feedback information impact factor andα=4.3；
E2：Severe texture feedback information is calculated as follows：
,
Wherein：T(m,n) it is the picture edge characteristic that height texture area is obtained through D step,β=0.45；
E3, feedback threshold model：
Slight texture area threshold model：；
Severe texture area threshold model：；
Wherein：TFor threshold value andT=128。
10. texture information merges Multi-scale model forest digital picture halftoning method, feature according to claim 9
It is in the F-step to handle smooth area in image in the way of raster scanning using smooth area threshold model, using slight
Texture area threshold model handles the slight texture area in image, and utilizes the weight in severe texture area threshold model processing image
Spend texture area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810594811.7A CN108921886B (en) | 2018-06-11 | 2018-06-11 | Texture information fusion multi-scale structured forest digital image halftone method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810594811.7A CN108921886B (en) | 2018-06-11 | 2018-06-11 | Texture information fusion multi-scale structured forest digital image halftone method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108921886A true CN108921886A (en) | 2018-11-30 |
| CN108921886B CN108921886B (en) | 2021-09-14 |
Family
ID=64419121
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810594811.7A Active CN108921886B (en) | 2018-06-11 | 2018-06-11 | Texture information fusion multi-scale structured forest digital image halftone method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921886B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116934618A (en) * | 2023-07-13 | 2023-10-24 | 江南大学 | Image halftone method, system and medium based on improved residual error network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090148003A1 (en) * | 2007-12-05 | 2009-06-11 | Canon Kabushiki Kaisha | Block-based noise detection and reduction method with pixel level classification granularity |
| CN103793897A (en) * | 2014-01-15 | 2014-05-14 | 昆明理工大学 | Digital image halftone method based on small-wavelet-domain multi-scale information fusion |
| CN103955908A (en) * | 2014-04-16 | 2014-07-30 | 昆明理工大学 | Digital image halftone method based on multi-scale perception error measure approximation global optimization |
| EP2812849A1 (en) * | 2012-02-09 | 2014-12-17 | Hewlett-Packard Development Company, L.P. | Forensic verification from halftone images |
| CN106506901A (en) * | 2016-09-18 | 2017-03-15 | 昆明理工大学 | A kind of hybrid digital picture halftoning method of significance visual attention model |
| CN106651816A (en) * | 2016-12-30 | 2017-05-10 | 首都师范大学 | Halftone processing method and system for digital image |
| CN107133936A (en) * | 2017-05-09 | 2017-09-05 | 上海理工大学 | Digital halftoning method |
| CN107209865A (en) * | 2015-02-27 | 2017-09-26 | 惠普发展公司，有限责任合伙企业 | Recurrence halftone and ash value are replaced |
-
2018
- 2018-06-11 CN CN201810594811.7A patent/CN108921886B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090148003A1 (en) * | 2007-12-05 | 2009-06-11 | Canon Kabushiki Kaisha | Block-based noise detection and reduction method with pixel level classification granularity |
| EP2812849A1 (en) * | 2012-02-09 | 2014-12-17 | Hewlett-Packard Development Company, L.P. | Forensic verification from halftone images |
| CN103793897A (en) * | 2014-01-15 | 2014-05-14 | 昆明理工大学 | Digital image halftone method based on small-wavelet-domain multi-scale information fusion |
| CN103955908A (en) * | 2014-04-16 | 2014-07-30 | 昆明理工大学 | Digital image halftone method based on multi-scale perception error measure approximation global optimization |
| CN107209865A (en) * | 2015-02-27 | 2017-09-26 | 惠普发展公司，有限责任合伙企业 | Recurrence halftone and ash value are replaced |
| CN106506901A (en) * | 2016-09-18 | 2017-03-15 | 昆明理工大学 | A kind of hybrid digital picture halftoning method of significance visual attention model |
| CN106651816A (en) * | 2016-12-30 | 2017-05-10 | 首都师范大学 | Halftone processing method and system for digital image |
| CN107133936A (en) * | 2017-05-09 | 2017-09-05 | 上海理工大学 | Digital halftoning method |
Non-Patent Citations (3)
| Title |
|---|
| FLING Y H等: "《Optimizing the error diffusion filter for blue noise halftoning with multiscale error diffusion》", 《IEEE TRANSACTIONS ON IMAGE PROCESSING A PUBLICATION OF THE IEEE SIGNAL PROCESSING SOCIETY》 * |
| 何自芬等: "《基于K-means聚类的数字半色调算法》", 《计算机应用研究》 * |
| 吴启科等: "《基于系数优化的图像边缘增强误差扩散方法》", 《价值工程》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116934618A (en) * | 2023-07-13 | 2023-10-24 | 江南大学 | Image halftone method, system and medium based on improved residual error network |
| CN116934618B (en) * | 2023-07-13 | 2024-06-11 | 江南大学 | Image halftone method, system and medium based on improved residual error network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108921886B (en) | 2021-09-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | Tensor oriented no-reference light field image quality assessment | |
| Kandhway et al. | An optimal adaptive thresholding based sub-histogram equalization for brightness preserving image contrast enhancement | |
| US5201013A (en) | Dither processing method | |
| CN101682686B (en) | Apparatus and method for image processing | |
| Kolås et al. | Spatio-temporal Retinex-inspired envelope with stochastic sampling: a framework for spatial color algorithms | |
| Saarinen | Color image segmentation by a watershed algorithm and region adjacency graph processing | |
| GB2581374A (en) | 3D Face reconstruction system and method | |
| JPH1153525A (en) | Facial organ detector and medium | |
| Aykroyd | Bayesian estimation for homogeneous and inhomogeneous Gaussian random fields | |
| CN110211016A (en) | A kind of watermark embedding method based on convolution feature | |
| CN108921886A (en) | A kind of texture information fusion Multi-scale model forest digital picture halftoning method | |
| CN108810314A (en) | Unordered error based on the enhancing of multi-grey image edge spreads digital halftoning method | |
| CN116703685A (en) | Image dewatering printing method, device, equipment and medium | |
| CN111340741A (en) | Particle swarm optimization gray level image enhancement method based on quaternion and L1 norm | |
| Kandhway et al. | Modified clipping based image enhancement scheme using difference of histogram bins | |
| Gooran et al. | 3D surface structures and 3D halftoning | |
| CN110211090A (en) | A method of for assessment design composograph quality | |
| CN115909129A (en) | Face forgery detection method based on frequency domain feature double-flow network | |
| Parihar | Histogram modification and DCT based contrast enhancement | |
| Guo et al. | Efficient halftoning based on multiple look-up tables | |
| Azam et al. | A state-of-the-art review on wavelet based image resolution enhancement techniques: performance evaluation criteria and issues | |
| Kobayashi et al. | Halftoning technique using genetic algorithms | |
| Liu et al. | Fast color image quantization with error diffusion and morphological operations | |
| JPH08214159A (en) | Image processing method and image processing unit | |
| CN108737686A (en) | Edge enhancing error based on image coefficient change spreads digital halftoning method |
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 |