CN1075848C - Automatic fabric-crease grading method - Google Patents

Abstract

The present invention relates to a method for automatically grading the creases of a fabric. The present invention comprises the following steps: a. the creases of a sample fabric to be tested are scanned into an image by using a scanner; b. the crease image is displayed in a highlighting mode through image preprocessing, namely that the characteristic is extracted; c. the image data is quantified; d. the linearity and the deviation value characteristic of the image data are analyzed; e. the grade of the fabric image is judged by using a classification category equation; in the image preprocessing, image processing technologies, such as gray-level schedulization, ordering, equalization, obfuscation, etc. are used; In the image data quantification, the three-dimensional longitudinal section analysis, deviation value analysis technology and a statistical method are used. By using the method, the grading result of the crease appearance is more objective.

Description

The automatic ranking method of fabric-crease

The present invention relates to a kind of automatic ranking method of fabric-crease, particularly a kind ofly fabric-crease commented utmost point method automatically with the computer video.

For clothing industry, the wrinkle Zhe degree of fabric face to its moulding after the aesthetic property of clothes considerable influence is arranged, be very important through the detection of the wrinkle Zhe that presented after the washing.The grading of wrinkle Zhe outward appearance in the textiles test item generally must be employed three testing staff that were subjected to professional training, after respectively sample to be tested and standard comparison model being compared one by one, comments utmost point result average and obtain comparatively objective result by subjectivity.

Artificial comparison mode has some shortcomings, as:

(1) must employ that three manpowers are constantly substituted standard comparison model and sample to be tested compares, cause waste of manpower resource.

(2) must provide the grading chamber with good illumination, wasting space.

(3) may be because the influence of artificial carelessness or external environment (as the distance of testing staff's at that time mood, observation, the bright-dark degree etc. of light source on every side) cause the error of difference.

The object of the present invention is to provide a kind of video technology of grading automatically,, shorten the personnel training time to reduce the cost that detects the unit hardware cost.

For achieving the above object, the present invention takes following measure:

The automatic ranking method of fabric-crease of the present invention is characterized in that, comprises the steps:

A. utilize scanner that the fold scanning of sample to be tested fabric is the demonstration video;

B. highlight the fold video through the video preliminary treatment, promptly extract feature;

C. quantize image data;

D analyzes its linearity and deviation value feature:

E. utilize classification difference equation, judge the grade of fabric video.

Described method is characterized in that, described video preliminary treatment is to use video treatment technologies such as GTG degreeization, orderingization, equalization, obfuscation.

Described method is characterized in that, described image data quantizes to be to use three-dimensional vertical profile analysis and deviation value analytical technology and statistical method.

Reaching embodiment in conjunction with the accompanying drawings is described in detail as follows concrete technical characterictic of the present invention:

Description of drawings:

Fig. 1: apparatus schematic diagram of the present invention.

Fig. 2: the demonstration schematic diagram when utilizing method of the present invention to operate.

Fig. 3 A, 3B: the scan image comparison diagram that is respectively uniform light of the present invention and exposure status.

Fig. 4: equalization analysis result schematic diagram of the present invention.

Fig. 5: Fuzzy processing result schematic diagram of the present invention.

Fig. 6: fabric vertical section analysis result schematic diagram of the present invention.

Fig. 7: difference equation schematic diagram of the present invention.

Fig. 8: the flow chart of the automatic rating system of wrinkle Zhe video of the present invention.

Please consult Fig. 1 earlier, it is a device schematic diagram of the present invention, device involved in the present invention comprises one scan device 1, is connected in a host computer 2, the video of sample fabric 4 can be scanned on screen 3 and show, owing to utilize scanner 1 as the video input unit, so can control fixed light source effectively, improve the shortcoming of using video camera can not control the good environment light source.Before use, utilize master sample earlier, data are stored in the memory, and test sample book is to pass through the solid coloured cloth after washing as the comparison model.The scanning of its sample is carried out with about 15 ° of angles of inclination.

The operation of relevant software as shown in Figure 2, it is expressed as follows content:

(1) debated the fabric viewing area: after video is by scanner scans, be shown in " being debated the fabric viewing area ".

(2) graphic operation district: the small icons in the operation " graphic operation district ", can or dwindle fabric image enlargement to be measured.

(3) engineer's scale:, can control the size of fabric video to be measured via the adjustment button on the resize ratio chi.

(4) master sample viewing area: the master sample viewing area can be shown in this with the master sample video that corresponds to fabric grade to be measured.

(5) known thing display as a result by debating: the small icons on operation " is known the thing viewing area by debating ", can observe detailed quantized data.

Groundwork of the present invention is the mode that frame scan is handled via video with the video that gets will to be quantized it, then makes correct objective grading according to the data that quantize again; Therefore, in the application facet of correlation theory, mainly can be divided into two parts, promptly the correlation theory of " video quantification " and basis " grading " is obtained a result.

After obtaining video from scanner, wrinkle Zhe special little also be not very obviously, in addition because fabric itself has its unique fabric texture, also can therefore have influence on the correctness of quantized data, so before quantizing, must do preliminary treatment to video earlier, to highlight wrinkle Zhe feature and to remove unnecessary noise (as the texture of fabric); In the present invention, strengthen the feature of wrinkle Zhe with GTG degreeization (Histogram), orderingization (Posterization), equalization image analysis technology such as (Equalization), remove the interference that the texture of fabric causes with the technology of obfuscation (Blurring), the algorithm of looking (3D-Profile) with three-dimensional vertical profile again quantizes video at last.Detailed content is described as follows.

GTG degree: the distribution scenario of picture element GTG value among the whole Zhang Yingxiang of expression statistics.With the representation unit of GTG value,, draw the GTG degree figure of this Zhang Yingxiang with the picture element sum of each GTG value representation unit as Y-axis as X-axis.Generally speaking, be to avoid the unnecessary exposure that when scanning, caused, wish that all the Histogram that scan image presented that is used for experiment is perfect waveform image, rather than be covered with the crest of sawtooth.Because Histogram perfectly waveform to represent in the scanning process that light source is dispersed on the sample comparatively average, if but be covered with the waveform of sawtooth, represent then to expose in some corners of sample.Fig. 3 represents both comparison schematic diagrames.

Its GTG degreeization (shown in the figure of Fig. 3 A lower right corner) wherein, is not exposed the video (shown in the video of Fig. 3 A upper left corner) of influence, puts in order on the Zhang Yingxiang, so can present comparatively perfectly waveform because scanning ray is evenly distributed in.This type of video also can present scanned real surface simultaneously.Be exposed the video (shown in the video of Fig. 3 B upper left corner) of influence, because scanning ray is distributed on the whole Zhang Yingxiang unevenly, cause some corners of video to be subjected to light comparatively strong, remainder is subjected to light comparatively few, so its GTG degreeization (Histogram) (shown in the figure of Fig. 3 B lower right corner) can present jagged waveform.This type of video can't present scanned real surface.Therefore this type of video is because the light and shade contrast is too strong, and the bright-dark degree that causing being exposed influences surpasses the object wrinkle light and shade slightly that Zhe caused itself and contrasts.

Orderingization also is that the GTG value of video is divided into n section equably from 0～255, and each section is got the GTG value of a GTG value as this section.Its objective is progressive GTG value is removed, and directly show striking contrast.Because cloth wrinkle Zhe can present the gradually layer effect of brightness under the illumination of light, therefore utilize the method for orderingization to highlight the position of wrinkle Zhe.

Average turns to the subsequent analysis instrument of GTG degreeization, mainly is to set the most black GTG value in the original GTG degreeization for " Black ", and the whitest GTG value is set " White " for, and remaining median equivalent intersperses among the scope of GTG value 0～255 then.As shown in Figure 4, Fig. 4 B is the schematic diagram of GTG degreeization, and Fig. 4 A is a result of doing equalization after the GTG degreeization again.The main purpose of equalization in the present invention is that the feature with the video surface highlights, and equalization is handled the wrinkle Zhe degree of fabric is highlighted.

Obfuscation: sample to be tested is actual cloth mostly, and cloth is through being formed by weaving, so lines that the surface of cloth all can existence itself, like this, the lines on this cloth surface can produce suitable influence to the image analysis method after overscanning, brightness meeting as the surface presents many up-and-down crests or trough, and the linearity of this curve also can cause considerable influence when analyzing its 3D-Profile simultaneously.Yet the material of master sample is a paper, does not take place so do not have above-mentioned situation.Be to solve the influence that the cloth surface pattern is caused, the present invention is the obfuscation technology in the utilization video treatment technology, and the influence that lines caused on cloth surface is reduced to a minimum.Following equation has illustrated the computing situation of obfuscation.Wherein G represents the GTG value of this point.This equational main purpose is that (Mask) asks its average with one 3 * 3 mask with the GTG value of the every bit in the video. $G=\underset{i=1}{\overset{9}{\mathrm{\Σ}}}\frac{x_i}{n}$

Fig. 5 explanation is through the result of Fuzzy processing.We can the finding surface lines blured the video of the just surface wrinkle Zhe that stays.

3-D surface analysis (Surface Analysis) is to utilize each pixel (x, y, z) several right, x wherein, y represents the position of this pixel, and z represents the GTG value of this pixel, construction goes out a three-dimensional coordinate system, then with each pixel according to coordinate (x, y, z) plot 3-D Surface, utilize triangle gridding to simulate the degree of cloth wrinkle Zhe at last again; What yet the utilization triangle gridding was used as 3D modelling (Modeling) debates the knowledge feature, need expend many times of CPU on calculating and setting up on these triangle griddings and debated the surface of knowing thing with simulation.For the present invention's commercialized degree in the future adverse influence is arranged thus, so the analysis that utilizes easier method to carry out fabric face to be measured is imperative.

3D-distribute (3D-Profile) be used for substituting above-mentioned 3D modelling 3D-Surface Modeling) the signature analysis rule.3D distribution (3D-Profile) signature analysis is exactly to utilize to be debated the cross section of knowing body surface in fact, or the fluctuating quantity of vertical section is as the foundation of analyzing.That is to say, (x, y, three-dimensional spatial analysis z) be reduced to (x, z) or (wherein the z value still represents to be debated the intensity gray scale value of knowing body surface for y, dual space analysis z), and (x y) then is its abscissa and ordinate.Fig. 6 represents to be debated the result of 3D distribution (3D-Profile) vertical section analysis of knowing body surface.

The data that quantize through video are a pile numeral, must be through the process of arrangement with statistical analysis, these quantized data inducing classifications could be drawn the method and the rating result of grading: the present invention uses the standard that the linearity (Linearity) principle changes into quantized data grading, and set up the statistics of deviation value (variance) parameter as whole video, make correct grading with decision function (decision function) according to the linearity (linearity) and these two parameters of deviation value (variance).Detailed content is described as follows:

The linearity (Linearity) is debated at figure to knit in (Pattern Recognition) and is all extensively used always.Particularly about being debated when knowing body surface its linear characteristic being arranged, as debate (Optical Character Recognition:OCR) in the knowledge at optical character.We know that in literal or numeral usually having some strokes is a horizontal or heavily fortified point, as " 1 ", " one ".When these numerals of explanation, this feature of the linearity just can be used as the foundation of debating knowledge.The linearity of quite tangible horizontal, vertical stroke is than higher, and the linearity of other opposite strokes is then lower.And in the present invention,, know the degree foundation that thing surface wrinkle Zhe quantizes so the linearity just can be used as to be debated because we adopt the mode of 3D distribution (3D-profile) to analyze the surface of knowledge thing to be measured.The computing formula of the linearity is expressed as follows: ${\mathrm{\μ}}_x=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{x_i}{n},$ ${\mathrm{\μ}}_y=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{y}{n}$ $m_{11}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{(x_i-{\mathrm{\μ}}_x)\×(y_i-{\mathrm{\μ}}_y)}{n}$ $m_{20}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{{(x_i-{\mathrm{\μ}}_x)}^2}{n},$ $m_{02}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{{(y_i-{\mathrm{\μ}}_y)}^2}{n}$

Wherein: X _i, Y _iCoordinate for certain stack features value.

μ _x, μ _yThe average coordinate of all characteristic values.

M ₁₁Conversion constant.

m ₂₀, m ₀₂Mark and Building Y target variance the representation eigenvalue Building X respectively.

Deviation value (Variance) is the variance that is used for calculating whole video, makes picture element x in the video _iValue be the GTG value of this point, n is a picture element sum in the video, so variance

Be expressed as: $\mathrm{var}=\frac{\mathrm{n\&Sigma;}{x}_i^2-{\left(\mathrm{\&Sigma;}{x}_i\right)}^2}{{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C9810311600111.png,C9810311600161.png"\; state="\{\{state\}\}">n</figure-callout>}}^2}$

Decision function (Decition Function): each coordinate points is represented a stack features value among Fig. 7, and f ₁Value f ₂Then know the rule equation for debating, therefore can substitution f after the characteristic value of test sample book obtains ₁With f ₂In, judge classification under the test sample book according to the positive negative value of its gained again.Following equation then can be expressed as the form of table 1;

f ₁=a ₁V _ar+b ₁L+c ₁

f ₂=a ₂V _ar+b ₂L+c ₂

Wherein: a ₁, b ₁, c ₁, a ₂, b ₂, c ₂Be constant.

Table 1

f 1(x 1,x 2…x n)	+	-	+	-
					f 2(x 1,x 2…x n)	+	-	-	+
The result	Ⅱ	Ⅳ	Ⅲ	Ⅰ

x ₁, x ₂Represent characteristic parameter variance (V respectively _Ar) and the linearity (L).

According to above-mentioned principle, the flow chart of automatic ranking method of the present invention as shown in Figure 8, its step is to be scanned to extract feature via device by sample to be tested earlier, then carry out linearity signature analysis and variation features analysis respectively, according to the classification of discriminant equation formula, distinguished out the result of various different gradings again.

The present invention has following effect:

(1) can save human resources, shorten the personnel training time:

The method according to this invention, three professional testing staff that originally need to use can be reduced to one, this testing staff only need spend few time can learn method of operating of the present invention, detects training and needn't rely on long-term grading, therefore can shorten time and the cost of personnel training.

(2) save cost and save the space:

Hardware of the present invention only needs a computer, a scanner, and installing easily; Therefore, can lower equipment cost and the space waste of saving the grading chamber.

The accuracy that (three) can improve grading and rapid property

Utilize method of the present invention for the grading of wrinkle Zhe outward appearance compared with traditional manual mode ranking method, more objective, can not cause one sided rating result because of human factor.

Claims (3)

1, a kind of automatic ranking method of fabric-crease is characterized in that, comprises the steps:

A. utilize scanner that the fold scanning of sample to be tested fabric is the demonstration video;

B. highlight the fold video through the video preliminary treatment, promptly extract feature;

C. quantize image data;

D analyzes its linearity and deviation value feature;

E. utilize classification difference equation, judge the grade of fabric video.

2, method according to claim 1 is characterized in that, described video preliminary treatment is to use video treatment technologies such as GTG degreeization, orderingization, equalization, obfuscation.

3, method according to claim 1 is characterized in that, described image data quantizes to be to use three-dimensional vertical profile analysis and deviation value analytical technology and statistical method.

Images