CN101976442A - Method for extracting fractal profile for representing fabric texture and Sobel operator filtering detail mixed characteristic vector - Google Patents

Abstract

The invention belongs to the field of digital image processing and mode recognition, in particular relates to a method for extracting a fractal profile for representing fabric texture and a Sobel operator filtering detail mixed characteristic vector, which comprises the following steps of: firstly, longitudinally and laterally projecting an original fabric image respectively and synchronously, combining two time sequences obtained by projection into one sequence, and estimating the fractal dimension of the sequence as a profile characteristic on the basis; subsequently, respectively carrying out lateral and vertical Sobel operator filtering on the original fabric image, and extracting four extreme value gray scale statistics as detail characteristics based on the basic cycle period of the fabric texture and the traversal method principle; and finally, combing the one fractal profile characteristic and the four Sobel operator filtering detail characteristics to form a mixed characteristic vector. The mixed characteristic vector has high complementarity among all characteristics, combines the profile information and the detail information of the texture, also combines the lateral information and the longitudinal information of the texture and can describe the characteristics of the fabric texture comprehensively in detail.

Description

A kind ofly be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method

Technical field

The invention belongs to Digital Image Processing and area of pattern recognition, particularly a kind ofly be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method.

Background technology

Can realize cloth textured parameter estimation, Texture classification, appearance of fabrics evaluation, Defect Detection or the like purpose by cloth textured characterization technique.Any cloth textured important information that all comprises two aspects, i.e. profile information and detailed information.Profile information provides overall rough structure and gray scale impression for human eye or machine vision, and detailed information then provides local meticulous structure and gray scale impression.Therefore, be comprehensively and characterize texture structure meticulously, reflect texture feature to greatest extent, when feature extraction, just must take into account the general picture and the detailed information of texture.For the ease of statement, the application plans those features that mainly reflect profile information and is called the general picture feature, and those features that mainly reflect detailed information are called minutia.Obviously, general picture feature and minutia emphasize particularly on different fields, and have great complementarity.The present invention is intended to discuss the cloth textured characterizing method based on fractal general picture feature and Sobel operator filtering minutia.

Than euclidean geometry, fractal geometry can provide better method when describing or generating natural things with self-similarity or class natural things, thereby extensively are used in the simulation of pattern-recognition, image and emulation or the like numerous areas.Self-similarity is one of central concept of fractal theory, and the notion of it and dimension is closely related.The object that fractal geometry are described has the self similarity on the statistical significance, and it is one of characteristic parameter of the most normal use when carrying out graphical analysis with fractal theory that self-similarity characterizes FRACTAL DIMENSION with FRACTAL DIMENSION.Fractal characteristic particularly fractal dimension can be portrayed coarse texture degree and complexity preferably, thus in practices such as Texture classification, identification as tolerance feature be rational.Wherein box counting dimension simple owing to notion, calculate the easy the most general a kind of fractal dimension of use that becomes.

For ease of explanation invention main points, be necessary box counting dimension and Sobel operator Principle of Filtering are done simple the introduction.If

Be non-NULL bounded aggregate arbitrarily, with N (δ, F) expression cover collection F required diameter be to the maximum δ the minimal number of collection, then the box dimension definition of F is

$D_B\left(F\right)=\underset{\mathrm{\δ}\→0}{\lim }\frac{\log N(\mathrm{\δ},F)}{-\log \mathrm{\δ}}$

Note, used δ-covering is still a general collection class in the definition, in this patent collection F refer in particular to into textile image when the vertical, horizontal projection, by the gradation of image one dimension time series that each is capable, each row pixel grey scale adds up and get the average gained, it also is the curve of each each row pixel grey scale Change in Mean of row of a presentation video, (δ, F) the required length of side of expression covering F is the minimum number of squares of δ to N, notes by abridging into N (δ).D _B(F) brief note is D.

During seasonal effect in time series meter box dimension of actual estimation, because this sequence is a curve, horizontal ordinate is the position of each point in the sequence, and ordinate is the sequential value of each point correspondence, need remove to cover fully this curve and add up N (δ) with the grid that is of a size of δ * δ.From the definition of box counting dimension as can be known, logN (δ) ∞ Dlog (1/ δ), this shows, and some is straight line to (1og (1/ δ), logN (δ)) asymptotic line in δ → 0 o'clock, and its slope is D.It is right that thereby change δ size can obtain a plurality of above-mentioned points, goes out respective straight by least square fitting then.The slope of this straight line is the box counting dimension of being asked.

Consider that woven fabric is by vertically being interwoven mutually through weft yarn, its image is a kind of typical texture image, and therefore available fractal characteristic characterizes cloth textured.People such as Conci (1998) adopt difference meter box method to extract cloth textured FRACTAL DIMENSION and standard deviation is used for characterizing cloth textured as characteristic parameter and detect fabric defects.People such as Xu Zengbo (2000) carry out at cloth textured image on the basis of Wold model analysis, with ask in the FRACTAL DIMENSION process whole fractal characteristic curve as characterizing cloth textured feature, carried out the fabric defects detection.People such as Wen (2002) adopt this fractal parameter of Hurst coefficient of estimating textile image based on the Fourier domain maximal possibility estimation operator of fractal Brown motion, detect fault with this as characterizing cloth textured characteristic parameter.People such as Yang Yan (2007) have extracted an overall FRACTAL DIMENSION feature and have realized objective evaluation to the fabric crape effect from the crepe fabric image.Go on foot the red firm people of grade (2007) in order to overcome the limitation of single fractal characteristic, a kind of many fractal characteristic vector extracting method have been proposed, the more relevant research on the defect detection effect of this method has had significantly improvement, but because a plurality of fractal characteristic vectors that extracted all are general picture features of reflection global information, thereby be unsuitable for detecting a lot of local faults.In the patent of " based on square and fractal texture classifying method " (2006), the researcher is the second moment of computed image at first, produce the moment characteristics image, again original image piece and moment characteristics image are estimated its fractal dimension, fractal dimension with original image piece and six moment characteristics images forms proper vector at last, carries out cloth textured classification as the input of support vector machine.

The Sobel operator is one of operator in the Flame Image Process, mainly as rim detection.Technically, it is a discreteness difference operator, is used for the approximate value of gradient of arithmograph image brightness function.Use this operator in any point of image, will produce corresponding gradient vector.The Sobel operator has two, a detection level edge, another detection of vertical edge.The Sobel operator image space utilize two 3 * 3 direction template in other words in convolution kernel and the image each point carry out the neighborhood convolution and finish rim detection, one of them strengthens the horizontal direction edge of image by the almost vertical direction gradient this both direction template, and another then strengthens the vertical direction edge of image by the level of approximation direction gradient.Sobel level and vertical edge strengthen template and are respectively

$T_x=\left(\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right)$

With $T_y=\left(\begin{array}{ccc}1& 0& -1\\ 2& 0& -2\\ 1& 0& -1\end{array}\right)$

At cloth textured Sobel operator representational field, do not see relevant domestic article or patent report as yet.In order to detect fabric crapand flaw, the Jasper of the U.S. and Potapalli adopt Sobel horizontal filtering operator that textile image has been carried out Filtering Processing, but only extracted the filtered edge image sectional view of Sobel, do not carry out more deep texture phenetic analysis, and only relate to a Sobel filter operator.In order to detect fabric defects, Lane has proposed the texture characterizing method that a kind of Sobel of employing operator and mathematical morphology combine in the American National patent of its application, its method is, at first original image is carried out Sobel level and vertical filtering, then two filtering images are merged, then this image binaryzation is handled, then carried out the mathematical morphology of binary image and handle, at last with the feature of frontier point as the sign texture.This report does not have to consider with the foundation of basic length of the cycle of texture cycle as feature extraction, the choosing method of binary-state threshold is not described, the single feature of being extracted also only relates to the frontier point pixel quantity, and clearly do not define the implication of frontier point, do not consider the distribution situation of frontier point in image.

The cloth textured characterizing method that above-mentioned existing document or patent relate to all is the extraction that is confined to global characteristics to the sign of cloth textured information, fail to take into account cloth textured general picture and detailed information, thereby can not be comprehensively and characterize cloth textured essential characteristic meticulously.In addition, above-mentioned Sobel operator texture characterizing method principal feature is that texture image must be chosen certain threshold value to realize the binaryzation of image after Sobel operator Filtering Processing.Two major defects are arranged like this: the one, choose the thing that optimal threshold is a difficulty at different texture; The 2nd, image is after binary conversion treatment, and a large amount of grayscale transition information are lost, only remaining complete black and complete two white value informations, and pending texture image has 256 gray levels usually.Sobel operator filtering Feature Extraction is underused fabric and is had regular round-robin characteristics through weft yarn.Therefore, the comparatively loaded down with trivial details and feature that extract on this basis of above-mentioned disposal route can not realize the more abundant properer sign of texture.And above-mentioned document or patent are in cloth textured fractal characterization, have following shortcoming: 1, fractal characteristic directly extracts on the two dimensional image basis, the global information that calculated amount is big by 2, fractal characteristic that extracted only can be portrayed texture is underused cloth textured intrinsic longitude and latitude orientation characteristics to improve the stability of feature in the time of can not carefully profoundly characterizing cloth textured detailed information 3, feature extraction.

Summary of the invention

The invention belongs to Digital Image Processing and area of pattern recognition, particularly a kind ofly be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method.At first the source textile texture image is carried out the vertical and horizontal projection respectively synchronously, and two time serieses that projection obtains are united become a sequence, adopt meter box method to estimate that the fractal dimension of above-mentioned sequence is as the general picture feature on this basis; Then the source textile image is carried out level and the filtering of vertical Sobel operator respectively, extract four extreme value gray-scale statistical amounts as minutia on this basis and according to cloth textured basic cycle period and traversal method principle; At last an above-mentioned fractal general picture feature and four Sobel filtering minutias are formed the composite character vector.The complementarity that has height in this composite character vector between each feature is taken into account the profile information and the detailed information of texture, also takes into account the horizontal information of texture and vertical information, can be comprehensively and portray cloth textured characteristics meticulously.

A kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method of the present invention, aspect fractal general picture feature extraction, at first to the synchronization implementation vertical and horizontal projection pre-service respectively of cloth textured image, longitudinal projection refers to ask for respectively the gray scale accumulated value or the average of each row pixel of image herein, transverse projection refers to ask for respectively the gray scale accumulated value of each row pixel of image, obtain two one dimension time serieses thus, with new one dimension time series of the end to end composition of these two sequences, estimate that then this seasonal effect in time series fractal dimension is as general picture feature of the present invention; Aspect the extraction of Sobel operator filtering minutia, at first adopt the one dimension Fast Fourier Transform (FFT) to obtain the horizontal and vertical basic cycle period size of textile image respectively, then textile image is carried out level and vertical Sobel operator Filtering Processing respectively synchronously; On this basis, calculate four extreme value gray-scale statistical amounts of cloth textured proper vector as minutia according to cloth textured basic cycle period and traversal method principle, described four extreme value gray-scale statistical measure features are respectively the very big statistic of transverse edge, the minimum statistic of transverse edge, the very big statistic of longitudinal edge and the minimum statistic of longitudinal edge, and four features of Ti Quing have characterized two extreme details statistical informations of cloth textured transverse edge and two extreme details statistical informations of longitudinal edge respectively thus; At last an above-mentioned fractal general picture feature and four Sobel filtering minutias are formed the composite character vector, cloth textured in order to characterize.

The described leaching process that is used to characterize the cloth textured composite character vector of being made up of fractal general picture feature and sobel operator filtering minutia is as follows:

W carries out vertical and horizontal projection pre-service respectively to original fabrics image rectangular window, promptly ask for the average of each row of W and the average of each row of W respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, the fractal characteristic that calculates this sequence then is designated as Bd as characterizing cloth textured general picture feature.

W carries out Sobel operator horizontal filtering to original fabrics image rectangular window, is designated as W through filtered image _h, at filtered textile image rectangular window W _hIn set up rectangle subwindow W ₂, described rectangle subwindow W ₂Length equal filtered textile image rectangular window W _hLength, described rectangle subwindow W ₂Width to equal cloth textured vertically basic cycle period unit length be the line period unit length, with rectangle subwindow W ₂With each fixing step-length slippage vertically to travel through whole W _hThereby, corresponding several gray-scale statistical values of trying to achieve, note reckling and the maximum wherein is E respectively ₁And E ₂, E ₁Be the minimum statistic of transverse edge, E ₂Be the very big statistic of transverse edge;

W carries out the vertical filtering of Sobel operator to original fabrics image rectangular window, is designated as W through filtered image _v, at filtered textile image rectangular window W _vIn set up rectangle subwindow W ₁, described rectangle subwindow W ₁Length equal cloth textured laterally basic cycle period unit length and promptly be listed as the cycle unit length, described rectangle subwindow W ₁Width equal filtered textile image rectangular window W _vWidth, with rectangle subwindow W ₁With each fixing step-length slippage flatly to travel through whole W _vThereby, corresponding several gray-scale statistical values of trying to achieve, note reckling and the maximum wherein is E respectively ₃And E ₄, E ₃Be the minimum statistic of longitudinal edge, E ₄Be the very big statistic of longitudinal edge;

Finally obtain characterizing cloth textured composite character vector [Bd E ₁E ₂E ₃E ₄].

As optimized technical scheme:

Wherein, aforesaid a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, described fractal characteristic refers to adopt meter box method to estimate the box counting dimension that obtains, and its concrete computing method are referring to the relevant introduction in the background technology.

Aforesaid a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, used δ size sequence was 2～6 pixels when described box counting dimension was estimated.

Aforesaid a kind of cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method that is used to characterize, described statistic can be one of any for the agricultural entropy of celestial being, logarithm energy entropy, p-norm entropy, gray average, standard deviation, the degree of bias, kurtosis or the like.

Aforesaid a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, described fabric is a woven fabric.

Aforesaid a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, described cloth textured image laterally consistent with weft direction, and cloth textured image is vertically consistent with warp thread direction, be intended to bring into play better the effect of Sobel operator level and vertical filtering, thereby characterize cloth textured weft yarn information and warp thread information better.

Aforesaid a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, described fixing step-length is 1～3 pixel.

Aforesaid a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, described rectangle subwindow W ₁Horizontal sliding fixed step size and W ₂Vertical slippage fixed step size needn't be identical.

Consider the rectangle ABCD of textile image W, it is of a size of L ₁* L ₂, promptly laterally (AD) and vertical (AB) length are respectively L ₁And L ₂, then to obtain a length after longitudinal projection handles be L to W ₁The one dimension time series, W obtains a length after transverse projection is handled be L ₂The one dimension time series, length of both end to end compositions is L ₁+ L ₂New time series, counting then under the situation that the box method is 2～6 pixels at box size sequence δ and calculating this seasonal effect in time series fractal dimension is box counting dimension, the gained feature is fractal general picture feature, is designated as Bd.

Promptly the row cycle is P along the horizontal basic cycle to establish W again ₁Individual pixel, the basic cycle longitudinally is that line period is P ₂, the pixel count after subwindow size here and line period thereof and row cycle all refer to round.P ₁The basic cycle period of the arbitrary capable pixel grey scale value set by calculating W obtains P ₂The basic cycle period of the arbitrary row pixel grey scale value set by calculating W obtains, and wherein the calculating of above-mentioned basic cycle realizes by one dimension Fast Fourier Transform (FFT) (FFT).For N point sequence x (n), its FFT transfer pair is defined as

$X\left(k\right)=\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}x\left(n\right){\mathrm{\ω}}_N^{\mathrm{nk}},k=\mathrm{0,1},L,N-1$

$x\left(n\right)=\frac{1}{N}\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}X\left(k\right){\mathrm{\ω}}_N^{-\mathrm{nk}},n=\mathrm{0,1},L,N-1$

Wherein,

Be called twiddle factor.

X (k) sequence that sequence of real numbers x (n) obtains after FFT handles is a sequence of complex numbers, first value respective frequencies of this sequence of complex numbers is 0, does not have practical significance, directly with its removal, remaining sequence is a structural symmetry sequence, and the data that only need get its preceding N/2 when carrying out analysis of spectrum get final product.The mould of X (k) is called amplitude, amplitude square be called power, be designated as W.The pairing frequency of peak power is the dominant frequency of sequence x (n), and the inverse of dominant frequency is the basic cycle of this sequence.If the sample frequency of sequence x (n) is f _s(Hz), then the k point is the pairing actual frequency of X (k)

Generally speaking, regulation f _s=1, therefore,

Thereby the actual cycle that the k point is corresponding

Rectangle A ₁B ₁C ₁D ₁Represent that arbitrary lateral length is P among the W ₁And longitudinal length is L ₂Subwindow, be designated as W ₁, rectangle A ₂B ₂C ₂D ₂Represent that arbitrary lateral length is L among the W ₁And longitudinal length is P ₂Subwindow, be designated as W ₂

Original W is carried out Sobel operator horizontal filtering, remember that filtered image is W _hIn this case, we only pay close attention to W _hAnd W ₂, therefrom the feature of Ti Quing is the horizontal edge information that mainly reflects texture naturally.With W ₂With the step-length slippage vertically of each 1～3 pixel to travel through whole W _h, total L ₂-P ₂+ 1 W ₂Thereby, can correspondingly try to achieve L ₂-P ₂+ 1 gray-scale statistical amount, note reckling and the maximum wherein is E respectively ₁And E ₂, both reflect the horizontal and detailed information of texture for this.

Class is carried out the vertical filtering of Sobel operator to original W, remembers that filtered image is W _vIn this case, we only pay close attention to W _vAnd W ₁, therefrom the feature of Ti Quing is the vertical edge information that mainly reflects texture naturally.With W ₁With the step-length slippage flatly of each 1～3 pixel to travel through whole W _v, total L ₁-P ₁+ 1 W ₁Thereby, can correspondingly try to achieve L ₁-P ₁+ 1 gray-scale statistical amount, note reckling and the maximum wherein is E respectively ₃And E ₄, both reflect the vertical and detailed information of texture for this.

Finally obtain characterizing cloth textured composite character vector [Bd E ₁E ₂E ₃E ₄].

Preferred celestial agricultural entropy, all the other statistic computing formula can be consulted interrelated data.Celestial agricultural entropy is an important tolerance of the degree of uncertainty of signal.For image, its quantity of information is big more, and intensity profile is regular more, and corresponding celestial agricultural entropy is just more little, otherwise celestial agricultural entropy is just big more.Celestial agricultural entropy is defined as follows:

$X\left(s\right)=-{\mathrm{\Σ}}_i{s}_i^2{\log }_2\left(s_i^2\right)$

Wherein arrange 0log as usual ₂(0)=0, s _iBe the image pixel gray-scale value.

Can replace celestial agricultural entropy in above-mentioned with the entropy of other kind or gray-scale statistical characteristics.

Beneficial effect

Of the present inventionly a kind ofly be used to characterize that cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method extracted is used to characterize cloth textured composite character vector:

(1) have the complementarity of height between each feature, take into account the profile information and the detailed information of texture, also take into account the horizontal information of texture and vertical information, can be comprehensively and portray cloth textured characteristics meticulously;

(2) the complementation between profile information and detailed information, fractal characteristic belongs to two kinds of different texture characterizing methods with Sobel filtering feature itself, emphasizes particularly on different fields, thereby also has the characteristic effect complementary effect that causes because of method divergence between them;

(3) thus in the fractal characteristic leaching process, made full use of the cloth textured characteristics that its main information of warp-wise and broadwise orientation concentrates on warp-wise and broadwise that have, on the basis of the vertical or horizontal one dimension projection associating of image sequence, calculate, both keep most Useful Informations, reduced calculated amount again significantly;

(4) calculating of Sobel operator filtering minutia has made full use of cloth textured distinctive cycline rule characteristics, thereby the feature that calculates is stable more and press close to true;

(5) calculating of Sobel operator filtering minutia is not owing to need filtering image is implemented binary conversion treatment, therefore kept how useful transitional information in the filtering image, and be not only simple two value informations, and therefore need not to spend preferred binary-state threshold time, make computation process become simple and efficient.

Correlative study does not in the past then possess above-mentioned five advantages.

Description of drawings

Fig. 1 is that cloth textured composite character vector of the present invention extracts synoptic diagram

Fig. 2 is the textile image of the width of cloth 64*64 pixel size of embodiment 1

Fig. 3 is the image of embodiment 1 former figure behind Sobel operator horizontal filtering

Fig. 4 is the image of embodiment 1 former figure after the vertical filtering of Sobel operator

Fig. 5 is the textile image of the width of cloth 64*64 pixel size of embodiment 2

Fig. 6 is the images of embodiment 2 former figure behind Sobel operator horizontal filtering

Fig. 7 is the images of embodiment 2 former figure after the vertical filtering of Sobel operator

Fig. 8 is the textile image of the width of cloth 64*64 pixel size of embodiment 3

Fig. 9 is the images of embodiment 3 former figure behind Sobel operator horizontal filtering

Figure 10 is the images of embodiment 3 former figure after the vertical filtering of Sobel operator

Figure 11 is the textile image of the width of cloth 64*64 pixel size of embodiment 4

Figure 12 is the images of embodiment 4 former figure behind Sobel operator horizontal filtering

Figure 13 is the images of embodiment 4 former figure after the vertical filtering of Sobel operator

Embodiment

Below in conjunction with embodiment, further set forth the present invention.Should be understood that these embodiment only to be used to the present invention is described and be not used in and limit the scope of the invention.Should be understood that in addition those skilled in the art can make various changes or modifications the present invention after the content of having read the present invention's instruction, these equivalent form of values fall within the application's appended claims institute restricted portion equally.

A kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method of the present invention, when extracting fractal general picture feature, at first the original fabrics image is carried out vertical and horizontal projection pre-service respectively synchronously, promptly ask for the average of each row of image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, the fractal characteristic that calculates this sequence then is as characterizing cloth textured general picture feature; When extracting Sobel operator filtering minutia, at first adopt the one dimension Fast Fourier Transform (FFT) to obtain the horizontal and vertical basic cycle period size of textile image respectively, then textile image is carried out level and vertical Sobel operator Filtering Processing respectively; On this basis, calculate four extreme value gray-scale statistical amounts of cloth textured proper vector as minutia according to cloth textured basic cycle period and traversal method principle, described four extreme value gray-scale statistical measure features are respectively the very big statistic of transverse edge, the minimum statistic of transverse edge, the very big statistic of longitudinal edge and the minimum statistic of longitudinal edge, and four features of Ti Quing have characterized two extreme details statistical informations of cloth textured transverse edge and two extreme details statistical informations of longitudinal edge respectively thus; At last fractal general picture feature and Sobel filtering minutia are formed one the five composite character vector of tieing up, promptly can be used for characterizing cloth textured characteristics.

Described cloth textured fractal general picture characteristic extraction procedure is as follows:

W carries out vertical and horizontal projection pre-service respectively to original fabrics image rectangular window, promptly ask for the average of each row of W and the average of each row of W respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, the fractal characteristic that calculates this sequence then is designated as Bd as characterizing cloth textured general picture feature.

Described four cloth textured extremal statistic Feature Extraction processes are as follows:

W carries out Sobel operator horizontal filtering to original fabrics image rectangular window, is designated as W through filtered image _h, at filtered textile image rectangular window W _hIn set up rectangle subwindow W ₂, described rectangle subwindow W ₂Length equal filtered textile image rectangular window W _hLength, described rectangle subwindow W ₂Width to equal cloth textured vertically basic cycle period unit length be the line period unit length, with rectangle subwindow W ₂With the step-length slippage vertically of each 1～3 pixel to travel through whole W _hThereby, corresponding several gray-scale statistical values of trying to achieve, note reckling and the maximum wherein is E respectively ₁And E ₂, E ₁Be the minimum statistic of transverse edge, E ₂Be the very big statistic of transverse edge;

W carries out the vertical filtering of Sobel operator to original fabrics image rectangular window, is designated as W through filtered image _v, at filtered textile image rectangular window W _vIn set up rectangle subwindow W ₁, described rectangle subwindow W ₁Length equal cloth textured laterally basic cycle period unit length and promptly be listed as the cycle unit length, described rectangle subwindow W ₁Width equal filtered textile image rectangular window W _vWidth, with rectangle subwindow W ₁With the step-length slippage flatly of each 1～3 pixel to travel through whole W _vThereby, corresponding several gray-scale statistical values of trying to achieve, note reckling and the maximum wherein is E respectively ₃And E ₄, E ₃Be the minimum statistic of longitudinal edge, E ₄Be the very big statistic of longitudinal edge;

Finally obtain characterizing cloth textured composite character vector [Bd E ₁E ₂E ₃E ₄].

Wherein, aforesaid a kind of cloth textured Sobel operator filtering minutia extracting method that is used to characterize, described gray-scale statistical amount can be one of any for the agricultural entropy of celestial being, logarithm energy entropy, p-norm entropy, gray average, standard deviation, the degree of bias, kurtosis or the like.

Be further described below in conjunction with accompanying drawing:

(1) gather the cloth textured image of digitizing, be designated as W, shown in the rectangle ABCD among Fig. 1, it is of a size of L ₁* L ₂Pixel, promptly laterally (AD) and vertical (AB) length are respectively L ₁Pixel and L ₂Pixel;

(2) W is carried out vertical and horizontal projection pre-service respectively, promptly ask for the average of each row of W and the average of each row of W respectively, obtain two one dimension time serieses,, be designated as S new time series of the end to end composition of these two sequences;

(3) adopt meter box method to estimate the box counting dimension of S, be designated as Bd, wherein box size sequence δ adopts 2～6 pixels;

(4) extract the arbitrary capable pixel grey scale value set of cloth textured image to be analyzed, Fast Fourier Transform (FFT) is tried to achieve it and promptly is listed as basic cycle P along the horizontal basic cycle by one dimension ₁

(5) extract arbitrary row pixel grey scale value set of cloth textured image to be analyzed, its basic cycle longitudinally basic cycle P is at once tried to achieve in Fast Fourier Transform (FFT) by one dimension ₂

(6) original W is implemented the horizontal operator filtering of Sobel, remember that filtered image is W _h

(7) original W is implemented the filtering of Sobel vertical operator, remember that filtered image is W _v

(8) at W _hIn appoint and to get one as rectangle A ₂B ₂C ₂D ₂Shown subwindow is noted by abridging to being designated as W ₂, this subwindow lateral length is L ₁And longitudinal length is P ₂, then with W ₂With the distance slippage vertically of each 1～3 pixel to travel through whole W _h, total L ₂-P ₂+ 1 W ₂Thereby, can correspondingly try to achieve L according to formula ₂-P ₂+ 1 gray-scale statistical value, preferred celestial agricultural entropy, celestial agricultural entropy is defined as follows:

$X\left(s\right)=-{\mathrm{\Σ}}_i{s}_i^2{\log }_2\left(s_i^2\right)$

Note reckling and the maximum wherein is E respectively ₁And E ₂

(9) at W _vIn appoint and to get one as rectangle A ₁B ₁C ₁D ₁Shown subwindow is noted by abridging to being designated as W ₁, this subwindow lateral length is P ₁And longitudinal length is L ₂, then with W ₁With the distance slippage flatly of each 1～3 pixel to travel through whole W _v, total L ₁-P ₁+ 1 W ₁Thereby, can correspondingly try to achieve L according to formula ₁-P ₁+ 1 gray-scale statistical value, preferred celestial agricultural entropy, celestial agricultural entropy is defined as follows:

$X\left(s\right)=-{\mathrm{\Σ}}_i{s}_i^2{\log }_2\left(s_i^2\right)$

Note reckling and the maximum wherein is E respectively ₃And E ₄

(10) obtain characterizing cloth textured proper vector [Bd E ₁E ₂E ₃E ₄].

Embodiment 1

(1) obtains textile image, as shown in Figure 2.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.55.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=6 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=4 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain filtering image as shown in Figure 3.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain filtering image as shown in Figure 4.

(7) according to Fig. 3 and P ₂And in conjunction with celestial agricultural entropy computing formula

$X\left(s\right)=-{\mathrm{\Σ}}_i{s}_i^2{\log }_2\left(s_i^2\right)$

Adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 1 pixel herein, the result is respectively-9.22 * 10 ⁷With-3.36 * 10 ⁷

(8) according to Fig. 4 and P ₁And in conjunction with celestial agricultural entropy computing formula

$X\left(s\right)=-{\mathrm{\Σ}}_i{s}_i^2{\log }_2\left(s_i^2\right)$

Adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 1 pixel herein, the result is respectively-12.10 * 10 ⁷With-5.80 * 10 ⁷

(9) finally obtain characterizing this cloth textured proper vector [1.55-9.22 * 10 ⁷-3.36 * 10 ⁷-12.10 * 10 ⁷-5.80 * 10 ⁷].

Embodiment 2

(1) obtains textile image, as shown in Figure 5.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.34.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=20 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=11 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain filtering image as shown in Figure 6.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain filtering image as shown in Figure 7.

(7) according to Fig. 6 and P ₂And in conjunction with logarithm energy entropy computing formula

$X\left(s\right)={\mathrm{\Σ}}_i{\log }_2\left(s_i^2\right)$

Adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 1 pixel herein, the result is respectively 3525.8 and 3796.0.

(8) according to Fig. 7 and P ₁And in conjunction with logarithm energy entropy computing formula

$X\left(s\right)={\mathrm{\Σ}}_i{\log }_2\left(s_i^2\right)$

Adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 2 pixels herein, the result is respectively 6008.0 and 6399.3.

(9) finally obtain characterizing this cloth textured proper vector [1.323525.83796.06008.06399.3].

Embodiment 3

(1) obtains textile image, as shown in Figure 8.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.26.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=8 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=15 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain filtering image as shown in Figure 9.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain filtering image as shown in figure 10.

(7) according to Fig. 9 and P ₂And in conjunction with the gray average computing formula

$X\left(s\right)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{s}_i$

Adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 2 pixels herein, the result is respectively 74.47 and 93.54.

(8) according to Figure 10 and P ₁And in conjunction with the gray average computing formula

$X\left(s\right)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{s}_i$

Adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 1 pixel herein, the result is respectively 95.47 and 112.58.

(9) finally obtain characterizing this cloth textured proper vector [1.2674.4793.5495.47112.58].

Embodiment 4

(1) obtains textile image, as shown in figure 11.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.43.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=4 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=4 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain filtering image as shown in figure 12.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain filtering image as shown in figure 13.

(7) according to Figure 12 and P ₂And the standard deviation computing formula of combining image gray scale

$X\left(s\right)=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(s_i-\stackrel{\‾}{s})}^2},$

Wherein, $\stackrel{\‾}{s}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{s}_i$

Adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 1 pixel herein, the result is respectively 56.54 and 110.56.

(8) according to Figure 13 and P ₁And the standard deviation computing formula of combining image gray scale

$X\left(s\right)=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(s_i-\stackrel{\‾}{s})}^2},$

Wherein, $\stackrel{\‾}{s}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{s}_i$

Adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 3 pixels herein, the result is respectively 97.19 and 122.63.

(9) finally obtain characterizing this cloth textured proper vector [1.4356.54110.5697.19122.63].

Embodiment 5

(1) obtain textile image, the image size still is the 64*64 pixel.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.29.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=6 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=4 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain the corresponding horizontal filtering image.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain corresponding vertical filtering image.

(7) according to horizontal filtering image and P ₂And in conjunction with p-norm entropy computing formula

X(s)＝∑ _i|s _i| ^p’

Wherein, p＞1

For example when p=1.5, adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 2 pixels herein, the result is respectively 71330 and 53856.

(8) according to vertical filtering image and P ₁And in conjunction with p-norm entropy computing formula

X(s)＝∑ _i|s _i| ^p

Wherein, p＞1

For example when p=1.5, adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 2 pixels herein, the result is respectively 21679 and 57284.

(9) finally obtain characterizing this cloth textured proper vector [1.29 71,330 53,856 21,679 57284].

Embodiment 6

(1) obtain textile image, the image size still is the 64*64 pixel.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.37.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=10 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=17 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain the corresponding horizontal filtering image.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain corresponding vertical filtering image.

(7) according to horizontal filtering image and P ₂And in conjunction with degree of bias computing formula

$X\left(s\right)=\frac{E(s_i-\stackrel{\‾}{s})}{{\mathrm{\σ}}^3}$

Wherein, Be all s _iAverage, σ is all s _iStandard deviation, the mathematical expectation of E (t) expression variable t,

Adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 2 pixels herein, the result is respectively-0.2305 and 2.9258.

(8) according to vertical filtering image and P ₁And in conjunction with degree of bias computing formula

$X\left(s\right)=\frac{E(s_i-\stackrel{\‾}{s})}{{\mathrm{\σ}}^3}$

Wherein,

Be all s _iAverage, σ is all s _iStandard deviation, the mathematical expectation of E (t) expression variable t,

Adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 3 pixels herein, the result is respectively-0.3221 and 1.2969.

(9) finally obtain characterizing this cloth textured proper vector [1.37-0.2305 2.9258-0.32211.2969].

Embodiment 7

(1) obtain textile image, the image size still is the 64*64 pixel.

(2) ask for the average of each row of this image and the average of each row respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, adopt at box size sequence δ under the situation of 2～6 pixels, adopt meter box method to estimate this seasonal effect in time series box counting dimension, the result is 1.18.

(3) adopt one dimension FFT that the arbitrary capable gradation data of former figure is carried out computation of Period, obtain being listed as basic cycle P ₁=10 pixels.

(4) adopt one dimension FFT that arbitrary row gradation data of former figure is carried out computation of Period, obtain row basic cycle P ₂=17 pixels.

(5) former figure is implemented Sobel operator horizontal filtering, obtain the corresponding horizontal filtering image.

(6) former figure is implemented the vertical filtering of Sobel operator, obtain corresponding vertical filtering image.

(7) according to horizontal filtering image and P ₂And in conjunction with the kurtosis computing formula

$X\left(s\right)=\frac{E(s_i-\stackrel{\‾}{s})}{{\mathrm{\σ}}^4}$

Wherein,

Be all s _iAverage, σ is all s _iStandard deviation, the mathematical expectation of E (t) expression variable t,

Adopt the traversal method to calculate E ₁And E ₂, the rectangle subwindow slippage step-length in the traversal method is 3 pixels herein, the result is respectively 1.2589 and 11.1322.

(8) according to vertical filtering image and P ₁And in conjunction with the kurtosis computing formula

$X\left(s\right)=\frac{E(s_i-\stackrel{\‾}{s})}{{\mathrm{\σ}}^4}$

Wherein,

Be all s _iAverage, σ is all s _iStandard deviation, the mathematical expectation of E (t) expression variable t,

Adopt the traversal method to calculate E ₃And E ₄, the rectangle subwindow slippage step-length in the traversal method is 1 pixel herein, the result is respectively 1.0779 and 2.9019.

(9) finally obtain characterizing this cloth textured proper vector [1.181.258911.13221.07792.9019].

Claims (9)

1. one kind is used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, and it is characterized in that: described composite character vector is made of jointly a fractal general picture feature and four Sobel filtering minutias;

Fractal general picture feature extraction

At first cloth textured image is implemented vertical and horizontal projection pre-service, longitudinal projection refers to ask for respectively the gray scale accumulated value of each row pixel of image herein, transverse projection refers to ask for respectively the gray scale accumulated value of each row pixel of image, obtain two one dimension time serieses thus, with new one dimension time series of the end to end composition of these two sequences, calculate this seasonal effect in time series fractal dimension then as general picture feature of the present invention;

Sobel operator filtering minutia is extracted

At first adopt the one dimension Fast Fourier Transform (FFT) to obtain the horizontal and vertical basic cycle period size of textile image respectively, then textile image is carried out level and vertical Sobel operator Filtering Processing respectively; On this basis, calculate four extreme value gray-scale statistical amounts of cloth textured proper vector as minutia according to cloth textured basic cycle period and traversal method principle, described four extreme value gray-scale statistical measure features are respectively the very big statistic of transverse edge, the minimum statistic of transverse edge, the very big statistic of longitudinal edge and the minimum statistic of longitudinal edge, and four features of Ti Quing have characterized two extreme details statistical informations of cloth textured transverse edge and two extreme details statistical informations of longitudinal edge respectively thus;

At last an above-mentioned fractal general picture feature and four Sobel filtering minutias are formed the composite character vector, cloth textured in order to characterize;

The described leaching process that is used to characterize the cloth textured composite character vector of being made up of fractal general picture feature and sobel operator filtering minutia is as follows:

W carries out vertical and horizontal projection pre-service respectively to original fabrics image rectangular window, promptly ask for the gray scale accumulated value of each row of W and the gray scale accumulated value of each row of W respectively, obtain two one dimension time serieses, with new time series of the end to end composition of these two sequences, the fractal characteristic that calculates this sequence then is designated as Bd as characterizing cloth textured general picture feature;

W carries out Sobel operator horizontal filtering to original fabrics image rectangular window, is designated as W through filtered image _h, at filtered textile image rectangular window W _hIn set up rectangle subwindow W ₂, described rectangle subwindow W ₂Length equal filtered textile image rectangular window W _hLength, described rectangle subwindow W ₂Width to equal cloth textured vertically basic cycle period unit length be the line period unit length, with rectangle subwindow W ₂With each fixing step-length slippage vertically to travel through whole W _hThereby, corresponding several gray-scale statistical values of trying to achieve, note reckling and the maximum wherein is E respectively ₁And E ₂, E ₁Be the minimum statistic of transverse edge, E ₂Be the very big statistic of transverse edge;

W carries out the vertical filtering of Sobel operator to original fabrics image rectangular window, is designated as W through filtered image _v, at filtered textile image rectangular window W _vIn set up rectangle subwindow W ₁, described rectangle subwindow W ₁Length equal cloth textured laterally basic cycle period unit length and promptly be listed as the cycle unit length, described rectangle subwindow W ₁Width equal filtered textile image rectangular window W _vWidth, with rectangle subwindow W ₁With each fixing step-length slippage flatly to travel through whole W _vThereby, corresponding several gray-scale statistical values of trying to achieve, note reckling and the maximum wherein is E respectively ₃And E ₄, E ₃Be the minimum statistic of longitudinal edge, E ₄Be the very big statistic of longitudinal edge;

Finally obtain characterizing cloth textured composite character vector [Bd E ₁E ₂E ₃E ₄].

2. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 1 is characterized in that described fractal characteristic refers to adopt meter box method to estimate the box counting dimension that obtains.

3. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 2 is characterized in that used δ size sequence was 2～6 pixels when described box counting dimension was estimated.

4. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 1, it is characterized in that described statistic is that celestial agricultural entropy, logarithm energy entropy, p-norm entropy, gray average, standard deviation, the degree of bias or kurtosis are one of any.

5. according to claim 1 or 4 described a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method, it is characterized in that the preferred celestial agricultural entropy of described statistic.

6. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 1, it is characterized in that, described cloth textured image laterally consistent with weft direction, and cloth textured image is vertical consistent with warp thread direction.

7. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 1 is characterized in that described fixing step-length is 1～3 pixel.

8. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 1 is characterized in that described rectangle subwindow W ₁Horizontal sliding fixed step size and W ₂Vertical slippage fixed step size needn't be identical.

9. a kind of be used to characterize cloth textured fractal general picture and Sobel operator filtering details composite character vector extracting method according to claim 1 is characterized in that described fabric is a woven fabric.

Images