CN101976441A - Method for extracting Sobel operator filtering profile for representing fabric texture and fractal detail mixed characteristic vector - Google Patents

Abstract

The invention relates to a method for extracting mixed feature vectors of Sobel operator filtering overview and fractal details for characterizing fabric texture. First, on the basis of horizontal and vertical Sobel operator filtering processing on the original fabric image respectively, a group of gray statistics with the same calculation method are used as the profile features; at the same time, each of the original images contains a horizontal The fractal dimension of the sub-window of the basic cycle or the vertical basic cycle, and finally select two fractal dimension extremes reflecting the horizontal detail information and two fractal dimension extremes reflecting the vertical detail information as the details of the fabric texture Features; the above two Sobel operator filter overview features and four fractal detail features form a mixed feature vector. The features of this mixed feature vector are highly complementary, taking into account the general information and detail information of the texture, as well as the horizontal and vertical information of the texture, and can comprehensively and meticulously describe the characteristics of the fabric texture.

Description

A mixed feature vector extraction method of Sobel operator filtering overview and fractal details for characterizing fabric texture

technical field

The invention belongs to the field of digital image processing and pattern recognition, in particular to a method for extracting feature vectors mixed with Sobel operator filtering overview and fractal details for characterizing fabric texture.

Background technique

With the help of fabric texture characterization technology, the purposes of fabric texture parameter estimation, texture classification, fabric appearance evaluation, and defect detection can be realized. Any fabric texture contains two important information, namely general information and detail information. The overview information provides the overall rough structure and grayscale impression for human eyes or machine vision, while the detail information provides the local fine structure and grayscale impression. Therefore, in order to fully and meticulously characterize the texture structure and reflect the texture characteristics to the maximum extent, both the general appearance and the detail information of the texture must be taken into account when extracting features. For ease of expression, the present application intends to refer to those features that mainly reflect general information as general features, and those that mainly reflect detailed information as detailed features. Obviously, the overview feature and the detail feature have their own emphases and are extremely complementary. The present invention aims at discussing a fabric texture characterization method based on Sobel operator filtering profile features and fractal detail features.

Compared with Euclidean geometry, fractal geometry can provide a better method for describing or generating natural or natural-like things with self-similarity, so it is widely used in many fields such as pattern recognition, image simulation and simulation. Self-similarity is one of the central concepts of fractal theory, which is closely related to the concept of dimension. The objects described by fractal geometry have statistical self-similarity, and fractal dimension is used to characterize self-similarity. Fractal dimension is one of the most commonly used characteristic parameters in image analysis with fractal theory. Fractal features, especially fractal dimension, can better describe texture roughness and complexity, so it is reasonable to use it as a measurement feature in the practice of texture classification and recognition. Among them, the box dimension has become the most commonly used fractal dimension because of its simple concept and easy calculation.

In order to illustrate the key points of the invention, it is necessary to briefly introduce the basic principles of box dimension and Sobel operator filtering.

为任意非空有界集，用N(δ，F)表示覆盖集F所需直径最大为δ的的集的最少数目，则F的盒维定义为set up

For any non-empty bounded set, use N(δ, F) to represent the minimum number of sets with the largest diameter δ required to cover the set F, then the box dimension of F is defined as

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\mathrm{<span\; class="notranslate">\; lim</span>}}\frac{\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; \&delta;</span>}}$

Note that the δ-coverage used in the definition is still a general set class. In this patent, the set F refers to the image obtained by accumulating the gray levels of pixels in each row and column and taking the average value when the fabric image is projected vertically and horizontally. Grayscale one-dimensional time series, that is, a curve representing the change in the average grayscale value of pixels in each row and column of the image, N(δ, F) represents the minimum number of squares with side length δ required to cover F, abbreviated as N( δ). D _B (F) is abbreviated as D.

When actually estimating the box dimension of a time series, since the sequence is a curve, the abscissa is the position of each point in the sequence, and the ordinate is the sequence value corresponding to each point, it is necessary to use a grid with a size of δ×δ to complete the calculation. Overlay the curve and count N(δ). From the definition of the box dimension, logN(δ)∞Dlog(1/δ), which shows that the asymptotes of several point pairs (log(1/δ), logN(δ)) when δ→0 are straight lines , whose slope is D. By changing the size of δ, multiple above-mentioned point pairs can be obtained, and then the corresponding straight line is fitted by the least square method. The slope of the line is the desired box dimension.

Considering that the woven fabric is made of warp and weft yarns interwoven vertically, its image is a typical texture image, so the fractal feature can be used to characterize the fabric texture. Conci et al. (1998) extracted the fractal dimension of fabric texture and its standard deviation as characteristic parameters to characterize fabric texture and detect fabric defects by using the differential box method. Xu Zengbo et al. (2000) based on the Wold model analysis of the fabric texture image, used the entire fractal characteristic curve in the process of obtaining the fractal dimension as the feature to characterize the fabric texture, and carried out fabric defect detection. Wen et al. (2002) used the Fourier frequency domain maximum likelihood estimation operator based on fractal Brownian motion to estimate the fractal parameter of the fabric image, the Hurst coefficient, and used it as a characteristic parameter to characterize the fabric texture to detect defects. Yang Yan et al. (2007) extracted a global fractal dimension feature from crepe fabric images to achieve an objective evaluation of fabric crepe effects. Bu Honggang et al. (2007) proposed a multi-fractal feature vector extraction method in order to overcome the limitations of single fractal features. This method has greatly improved the defect detection effect compared with previous related studies. The multiple extracted fractal feature vectors are overview features that reflect global information, so they are not suitable for detecting many local defects. In the patent "Texture Classification Method Based on Moments and Fractals" (2006), the researchers first calculate the second-order moment of the image to generate the moment feature image, then estimate the fractal dimension of the original image block and the moment feature image, and finally The fractal dimensions of the original image block and the six moment feature images form a feature vector, which is used as the input of the support vector machine for fabric texture classification.

Sobel operator is one of the operators in image processing, mainly used for edge detection. Technically, it is a discrete difference operator, which is used to approximate the gradient of the image brightness function. Using this operator at any point in the image will generate a corresponding gradient vector. There are two Sobel operators, one to detect horizontal edges and the other to detect vertical edges. The Sobel operator uses two 3×3 direction templates or convolution kernels to perform neighborhood convolution with each point in the image in the image space to complete edge detection. One of the two direction templates is enhanced by approximating the vertical direction gradient. The horizontal edge of the image, and the other enhances the vertical edge of the image by approximating the horizontal gradient. Sobel horizontal and vertical edge enhancement templates are

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\left(\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}& <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}& <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right)$

and $T_{\mathrm{the\; y}}=\left(\begin{array}{ccc}1& 0& -1\\ 2& 0& -2\\ 1& 0& -1\end{array}\right)$

In the field of Sobel operator characterization of fabric texture, no relevant domestic articles or patent reports have been seen. In order to detect fabric weft defects, Jasper and Potapalli in the United States used the Sobel horizontal filter operator to filter the fabric image, but only extracted the edge image profile after Sobel filtering, without further texture representation analysis, and only A Sobel filter operator is involved. In order to detect fabric defects, Lane proposed a texture representation method that combines Sobel operator and mathematical morphology in his national patent application. The method is to first perform Sobel horizontal and vertical filtering on the original image, and then The two filtered images are fused, and then the image is binarized, followed by the mathematical morphology of the binarized image, and finally the boundary points are used as the characteristics of the texture. The report did not consider the basic cycle length of the texture period as the basis for feature extraction, did not explain the selection method of the binarization threshold, and the single feature extracted only involved the number of border point pixels, and did not clearly define the meaning of the border points The distribution of boundary points in the image.

The fabric texture characterization methods involved in the above-mentioned existing literature or patents are limited to the extraction of global features for the characterization of fabric texture information, and fail to take into account both the general appearance and detailed information of fabric texture, so they cannot comprehensively and meticulously characterize the essential characteristics of fabric texture . In addition, the main feature of the above-mentioned Sobel operator texture representation method is that after the texture image is filtered by the Sobel operator, a certain threshold must be selected to realize the binarization of the image. This has two main disadvantages: one is that it is difficult to select the optimal threshold for different textures; the other is that after the image is binarized, a large amount of grayscale transition information is lost, leaving only black and full White binary information, and the texture image to be processed usually has 256 gray levels. Therefore, the above-mentioned processing method is relatively cumbersome and the features extracted on this basis cannot achieve a more adequate and appropriate representation of the texture. However, the above-mentioned documents or patents have the following disadvantages in the fractal feature representation of fabric texture: 1. The feature is directly extracted on the basis of a two-dimensional image, which requires a large amount of calculation; 2. The extracted fractal feature can only describe the global information of the texture, and cannot The detailed information of the fabric texture is meticulously and deeply represented; 3. The inherent warp and weft orientation characteristics of the fabric texture are not fully utilized in feature extraction to improve the stability of the feature; 4. The feature extraction does not fully utilize the inherent regular cycle characteristics of the fabric warp and weft yarns. Improve feature accuracy and stability.

Contents of the invention

The invention belongs to the field of digital image processing and pattern recognition, in particular to a method for extracting feature vectors mixed with Sobel operator filtering overview and fractal details for characterizing fabric texture. Firstly, on the basis of horizontal and vertical Sobel filter processing on the original fabric image respectively, a set of gray-scale statistics with the same calculation method from the two corresponding filtered images is used as the profile feature; at the same time, the original image is calculated according to the principle of traversal method The fractal dimension of each sub-window containing a horizontal basic cycle period and the fractal dimension of each sub-window containing a vertical basic cycle cycle, and finally select two extreme values of fractal dimension and two A fractal dimension extremum reflecting the vertical detail information is used as the detail feature to characterize the fabric texture; the above two Sobel operator filtering profile features and four fractal detail features form a mixed feature vector. The features of this mixed feature vector are highly complementary, taking into account the general information and detail information of the texture, as well as the horizontal and vertical information of the texture, and can comprehensively and meticulously describe the characteristics of the fabric texture.

A Sobel operator filtering general appearance and fractal details mixed feature vector extraction method for characterizing fabric texture of the present invention, in terms of general appearance feature extraction, the original fabric image is first synchronously processed by Sobel operator horizontal filtering and vertical filtering respectively to obtain two A corresponding filter image, and then extract a gray-level statistic with the same calculation method to form a feature vector representing the general appearance information of the fabric texture, a group of two; in terms of detail feature extraction, first use one-dimensional fast Fourier Transform to obtain the basic horizontal and vertical cycle size of the fabric texture image, and then calculate the fractal dimension of each sub-window containing a horizontal basic cycle in the image and each sub-window containing a vertical basic cycle in the image according to the principle of traversal method Finally, select two extreme values of fractal dimension that reflect horizontal detail information, that is, horizontal maximum fractal dimension and horizontal minimum fractal dimension, and two extreme values of fractal dimension that reflect vertical detail information, that is, vertical maximum fractal dimension Dimension and longitudinal minimum fractal dimension, as the detailed features of fabric texture. Two Sobel operator filter profile features and four fractal detail features together form the mixed feature vector of the present invention.

Wherein said sub-window comprising a horizontal basic cycle is a rectangular window whose length is long and the width of the fabric texture image is wide, and the fractal dimension of the sub-window of each horizontal basic cycle is It is calculated on the basis of the corresponding one-dimensional time series obtained by accumulating the gray value of the image pixel in the sub-window along the horizontal direction.

The described sub-window comprising a vertical basic cycle is a rectangular window whose length is long and the vertical basic cycle is wide, and the fractal dimension of the sub-window of each vertical basic cycle is It is calculated on the basis of the corresponding one-dimensional time series obtained by accumulating the gray value of the image pixel in the sub-window along the longitudinal direction.

The extraction process of the mixed feature vector composed of Sobel operator filter overview feature and fractal detail feature for characterizing fabric texture is as follows:

First collect the digitized fabric texture image, denoted as W, W is a rectangle, its size length × width is L ₁ ×L ₂ , that is, the horizontal and vertical lengths are L ₁ and L ₂ respectively, and its basic cycle along the horizontal direction is the column cycle is P ₁ pixels, and the basic cycle along the vertical direction, that is, the row cycle is P ₂ . Both the row cycle and the column cycle refer to the number of pixels after rounding, and P ₁ is obtained by calculating the basic cycle cycle of any row image pixel gray value set of W , P ₂ is obtained by calculating the basic cycle period of any set of image pixel gray values of W.

The original image is synchronously implemented with Sobel operator horizontal filtering and vertical filtering respectively, record the image after horizontal filtering of Sobel operator as W _h , and the image after vertical filtering of Sobel operator as W _v ;

Select a grayscale statistic, and then directly calculate the grayscale statistic of W _h , as the grayscale statistic feature of the horizontal edge texture profile, denoted as _Sh ;

Select the grayscale statistics consistent with the calculation of _Wh , and directly calculate the grayscale statistics of _Wv , as the grayscale statistics feature of the vertical edge texture profile, denoted as _Sv ;

In the fabric texture image W, select a horizontal basic cycle period P ₁ as long and the width L ₂ of the fabric texture image as a wide rectangular window as a sub-window containing a horizontal basic cycle period, denoted as W ₁ ; choose a fabric texture The length _L of the image is long, and the longitudinal basic cycle period _P2 is a wide rectangular window as a sub-window containing a vertical basic cycle period, denoted as W ₂ ;

For a certain W ₁ , calculate its image pixel grayscale projection along the row direction, that is, superimpose the image pixel grayscale values of each row of the sub-window along the horizontal direction to obtain a one-dimensional time series, from which a fractal can be calculated dimension, and then slide W ₁ horizontally with a fixed step to traverse the entire W, there are a total of L ₁ -P ₁ +1 W ₁ , so that L ₁ -P ₁ +1 fractal dimensions can be obtained accordingly, respectively Note that the smallest and largest ones are E ₁ and E ₂ , which are the minimum horizontal fractal dimension and the maximum horizontal fractal dimension, which reflect the horizontal extreme detail information of the texture;

For a certain W ₂ , calculate its image pixel gray value projection along the column direction, that is, superimpose the image pixel gray value of each column of the sub-window along the vertical direction to obtain a one-dimensional time series, from which it can be calculated A fractal dimension, and then slide W ₂ vertically with a fixed step to traverse the entire W, there are a total of L ₂ -P ₂ +1 W ₂ , so that L ₂ -P ₂ +1 fractal dimensions can be obtained accordingly , record the minimum and maximum among them as E ₃ and E ₄ , which are the minimum vertical fractal dimension and the maximum vertical fractal dimension, which reflect the vertical extreme detail information of the texture;

Finally, the mixed feature vector [S _h S _v E ₁ E ₂ E ₃ E ₄ ] representing the fabric texture is obtained.

Wherein, the above-mentioned method for extracting feature vectors by combining Sobel filter overview and fractal details for characterizing fabric texture, the fabric is woven fabric.

As described above, a Sobel operator filter overview and fractal details mixed feature vector extraction method for characterizing fabric texture, the horizontal direction of the image of the fabric is consistent with the weft direction, and the vertical direction of the image of the fabric is consistent with the warp direction.

As described above, the Sobel operator filters the profile feature, and the gray level statistic can be Shannon entropy, gray level mean or gray level standard deviation.

As mentioned above, a method for extracting feature vectors using Sobel operator filtering overview and fractal details for characterizing fabric texture, the fractal dimension refers to the box dimension, and its specific calculation method refers to the relevant introduction in the background technology.

According to the above-mentioned method for extracting feature vectors by combining Sobel filter overview and fractal details for characterizing fabric texture, the δ size sequence used for box dimension estimation is 2 to 6 pixels.

As mentioned above, a method for extracting feature vectors by combining Sobel filter overview and fractal details for characterizing fabric texture, the fixed step size refers to 1 to 3 pixels.

As mentioned above, a Sobel operator filter overview and fractal details mixed feature vector extraction method for characterizing fabric _texture , the horizontal sliding fixed step of the rectangular sub-window W ₁ each time is the same as the vertical The sliding fixed steps do not have to be the same.

A kind of Sobel operator filtering overview and fractal details mixed feature vector extraction method for characterizing fabric texture as above, wherein the calculation of above-mentioned basic cycle period P ₁ and P ₂ is realized by means of one-dimensional fast Fourier transform (FFT) . For an N-point sequence x(n), its FFT transform pair is defined as

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; nk</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}$

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; nk</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}$

称为旋转因子。in,

called the twiddle factor.

一般情况下，规定f_s＝1，因此，从而第k点对应的实际周期

The X(k) sequence obtained after the real number sequence x(n) is processed by FFT is a complex number sequence. The first value of the complex number sequence corresponds to a frequency of 0, which has no practical meaning. It is directly removed, and the remaining sequence is a structure For a symmetrical sequence, you only need to take the first N/2 data for spectrum analysis. The modulus of X(k) is called the magnitude, and the square of the magnitude is called the power, denoted as W. The frequency corresponding to the maximum power is the main frequency of the sequence x(n), and the reciprocal of the main frequency is the basic period of the sequence. Suppose the sampling frequency of the sequence x(n) is f _s (Hz), then the kth point is the actual frequency corresponding to X(k)

Generally, it is stipulated that f _s =1, therefore, Thus the actual period corresponding to the kth point

As mentioned above, a Sobel operator filter general appearance and fractal details mixed feature vector extraction method for characterizing fabric texture, Shannon entropy is the first choice for the gray level statistics. Shannon entropy is an important measure of the degree of uncertainty of a signal. For an image, the larger the amount of information and the more regular the gray distribution, the smaller the corresponding Shannon entropy, and vice versa. Shannon entropy is defined as follows:

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>$

It is conventionally agreed that 0log ₂ (0)=0, and _si is the gray value of the image pixel. Instead of the Shannon entropy statistic described above, other kinds of statistic can be used directly.

Beneficial effect

A mixed feature vector for characterizing fabric texture extracted by a Sobel operator filter overview and fractal details mixed feature vector extraction method for characterizing fabric texture:

(1) Each feature has a high degree of complementarity, taking into account the general information and detail information of the texture, as well as the horizontal and vertical information of the texture, and can comprehensively and meticulously describe the characteristics of the fabric texture;

(2) In addition to the complementarity between overview information and detail information, fractal features and Sobel operator filter features are two different texture representation methods, each with its own emphasis, so there are still features caused by method differences. effect complementary effect;

(3) The calculation of Sobel operator filtering profile features does not need to implement binarization processing on the filtered image, so more useful transition information is retained in the filtered image, not just simple binary information, and therefore no need Spending effort to optimize the binarization threshold makes the calculation process easier and faster;

(4) In the process of extracting fractal detail features, the fabric texture has the characteristics of warp and weft orientations, so that its main information is concentrated in the warp and weft. The basis of the vertical or horizontal one-dimensional projection sequence of the image sub-window It is calculated based on the above rather than two-dimensional images, which not only retains most of the useful information, but also greatly reduces the amount of calculation;

(5) The calculation of fractal detail features makes full use of the unique cycle rules of fabric texture, so the calculated features are more stable and close to reality.

Previous related research does not have the above five advantages.

Description of drawings

Fig. 1 is a schematic diagram of fabric texture mixed feature vector extraction of the present invention

Fig. 2 is the fabric image of a 64*64 pixel size of embodiment 1

Fig. 3 is the image after the original image of embodiment 1 is filtered horizontally by the Sobel operator

Fig. 4 is the image after the original image of embodiment 1 is vertically filtered by Sobel operator

Fig. 5 is the fabric image of a 64*64 pixel size of embodiment 2

Fig. 6 is the image after the horizontal filtering of the original image by the Sobel operator in embodiment 2

Fig. 7 is the image after the original image of embodiment 2 is vertically filtered by the Sobel operator

Fig. 8 is the fabric image of a 64*64 pixel size of embodiment 3

Fig. 9 is the image after the horizontal filtering of the original image by the Sobel operator in embodiment 3

Fig. 10 is the image after the original image of embodiment 3 is vertically filtered by the Sobel operator

Detailed ways

The present invention will be further described below in combination with specific embodiments. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the teachings of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

A Sobel operator filtering general appearance and fractal details mixed feature vector extraction method for characterizing fabric texture of the present invention, in terms of general appearance feature extraction, the original fabric image is first synchronously processed by Sobel operator horizontal filtering and vertical filtering respectively to obtain two A corresponding filter image, and then extract a gray-level statistic with the same calculation method to form a feature vector representing the general appearance information of the fabric texture, a group of two; in terms of detail feature extraction, first use one-dimensional fast Fourier Transform to obtain the basic horizontal and vertical cycle size of the fabric texture image, and then calculate the fractal dimension of each sub-window containing a horizontal basic cycle in the image and each sub-window containing a vertical basic cycle in the image according to the principle of traversal method Finally, two fractal dimension extremums reflecting the horizontal detail information (that is, the horizontal maximum fractal dimension and the horizontal minimum fractal dimension) and two fractal dimension extreme values reflecting the vertical detail information (that is, the vertical fractal dimension The maximum fractal dimension and the longitudinal minimum fractal dimension) are used as the detailed features to characterize the fabric texture. Two Sobel operator filter profile features and four fractal detail features together form the mixed feature vector of the present invention.

Wherein said sub-window comprising a horizontal basic cycle is a rectangular window whose length is long and the width of the fabric texture image is wide, and the fractal dimension of the sub-window of each horizontal basic cycle is It is calculated on the basis of the corresponding one-dimensional time series obtained by accumulating the gray value of the image pixel in the sub-window along the horizontal direction.

The described sub-window comprising a vertical basic cycle is a rectangular window whose length is long and the vertical basic cycle is wide, and the fractal dimension of the sub-window of each vertical basic cycle is It is calculated on the basis of the corresponding one-dimensional time series obtained by accumulating the gray value of the image pixel in the sub-window along the longitudinal direction.

The extraction process of the mixed feature vector composed of Sobel operator filter overview feature and fractal detail feature for characterizing fabric texture is as follows:

First collect the digitized fabric texture image, denoted as W, W is a rectangle, its size length × width is L ₁ ×L ₂ , that is, the horizontal and vertical lengths are L ₁ and L ₂ respectively, and its basic cycle along the horizontal direction is the column cycle is P ₁ pixels, and the basic cycle along the vertical direction, that is, the row cycle is P ₂ . Both the row cycle and the column cycle refer to the number of pixels after rounding, and P ₁ is obtained by calculating the basic cycle cycle of any row image pixel gray value set of W , P ₂ is obtained by calculating the basic cycle period of any set of image pixel gray values of W.

The original image is synchronously implemented with Sobel operator horizontal filtering and vertical filtering respectively, record the image after horizontal filtering of Sobel operator as W _h , and the image after vertical filtering of Sobel operator as W _v ;

Select a grayscale statistic, and then directly calculate the grayscale statistic of W _h , as the grayscale statistic feature of the horizontal edge texture profile, denoted as _Sh ;

Select the grayscale statistics consistent with the calculation of _Wh , and directly calculate the grayscale statistics of _Wv , as the grayscale statistics feature of the vertical edge texture profile, denoted as _Sv ;

In the fabric texture image W, select a horizontal basic cycle period P ₁ as long and the width L ₂ of the fabric texture image as a wide rectangular window as a sub-window containing a horizontal basic cycle period, denoted as W ₁ ; choose a fabric texture The length _L of the image is long, and the longitudinal basic cycle period _P2 is a wide rectangular window as a sub-window containing a vertical basic cycle period, denoted as W ₂ ;

For a certain W ₁ , calculate its image pixel grayscale projection along the row direction, that is, superimpose the image pixel grayscale values of each row of the sub-window along the horizontal direction to obtain a one-dimensional time series, from which a fractal can be calculated dimension, and then slide W ₁ horizontally with a fixed step to traverse the entire W, there are a total of L ₁ -P ₁ +1 W ₁ , so that L ₁ -P ₁ +1 fractal dimensions can be obtained accordingly, respectively Note that the smallest and largest ones are E ₁ and E ₂ , which are the minimum horizontal fractal dimension and the maximum horizontal fractal dimension, which reflect the horizontal extreme detail information of the texture;

For a certain W ₂ , calculate its image pixel gray value projection along the column direction, that is, superimpose the image pixel gray value of each column of the sub-window along the vertical direction to obtain a one-dimensional time series, from which it can be calculated A fractal dimension, and then slide W ₂ vertically with a fixed step to traverse the entire W, there are a total of L ₂ -P ₂ +1 W ₂ , so that L ₂ -P ₂ +1 fractal dimensions can be obtained accordingly , record the minimum and maximum among them as E ₃ and E ₄ , which are the minimum vertical fractal dimension and the maximum vertical fractal dimension, which reflect the vertical extreme detail information of the texture;

Finally, the mixed feature vector [S _h S _v E ₁ E ₂ E ₃ E ₄ ] representing the fabric texture is obtained.

Wherein the above-mentioned one is used for characterizing fabric texture and is used for characterizing fabric texture, and described fabric is woven fabric.

As described above, a Sobel operator filter overview and fractal details mixed feature vector extraction method for characterizing fabric texture, the horizontal direction of the image of the fabric is consistent with the weft direction, and the vertical direction of the image of the fabric is consistent with the warp direction.

As described above, the Sobel operator filters the profile feature, and the gray level statistic can be Shannon entropy, gray level mean or gray level standard deviation.

As mentioned above, a method for extracting feature vectors using Sobel operator filtering overview and fractal details for characterizing fabric texture, the fractal dimension refers to the box dimension, and its specific calculation method refers to the relevant introduction in the background technology.

According to the above-mentioned method for extracting feature vectors by combining Sobel filter overview and fractal details for characterizing fabric texture, the δ size sequence used for box dimension estimation is 2 to 6 pixels.

As mentioned above, a method for extracting feature vectors by combining Sobel filter overview and fractal details for characterizing fabric texture, the fixed step size refers to 1 to 3 pixels.

As mentioned above, a Sobel operator filter overview and fractal details mixed feature vector extraction method for characterizing fabric texture, the horizontal sliding step of the rectangular sub-window W ₁ each time and the vertical sliding step of W ₂ each time The step lengths do not have to be the same.

A kind of Sobel operator filtering overview and fractal details mixed feature vector extraction method for characterizing fabric texture as above, wherein the calculation of above-mentioned basic cycle period P ₁ and P ₂ is realized by means of one-dimensional fast Fourier transform (FFT) .

As mentioned above, a Sobel operator filter general appearance and fractal details mixed feature vector extraction method for characterizing fabric texture, Shannon entropy is the first choice for the gray level statistics.

Instead of the Shannon entropy statistic described above, other kinds of statistic can be used directly.

Further explanation is made below in conjunction with accompanying drawing:

(1) Acquire digitized fabric texture image, denoted as W, as shown in the rectangle ABCD in Fig. 1, its size is L ₁ × L ₂ pixels, that is, the horizontal (AD) and vertical (AB) lengths are L ₁ pixels and L ₂ pixels;

(2) Carry out the Sobel operator horizontal filtering and vertical filtering processing on the original image synchronously, record the image after the Sobel operator horizontal filtering as W _h , and the image after the Sobel operator vertical filtering as W _v ;

(3) Select a grayscale statistic from Shannon entropy, gray mean value, and grayscale variance, Shannon entropy is preferred, and then directly calculate the grayscale statistic of W _h as the grayscale statistics of the horizontal edge texture profile Quantitative characteristics, denoted as _Sh ;

(4) Select the same grayscale statistic as when calculating _Wh , directly calculate the grayscale statistic of _Wv , as the grayscale statistic feature of the vertical edge texture profile, denoted as _Sv ;

(5) Extract any row of pixel gray value sets of the fabric texture image to be analyzed, and obtain its basic period along the horizontal direction, namely the column basic period P ₁ , by means of one-dimensional fast Fourier transform;

(6) extract any column pixel gray value set of the fabric texture image to be analyzed, and obtain its basic period along the longitudinal direction, that is, the row basic period P ₂ by means of one-dimensional fast Fourier transform;

(7) Choose any sub-window in W as shown in the rectangle A ₁ B ₁ C ₁ D ₁ , abbreviated as W ₁ , the sub-window has a horizontal length of P ₁ and a vertical length of L ₂ , for this W ₁ , calculate the gray value projection of the image pixel along the row direction, that is, superimpose the gray value of the image pixel in each row of the sub-window along the horizontal direction to obtain a one-dimensional time series, under the premise that the δ size sequence is 2 to 6 pixels , a box dimension feature can be calculated from the time series, and then W ₁ is horizontally slid at a fixed step size of 1 to 3 pixels each time to traverse the entire W, and there are a total of L ₁ -P ₁ +1 W ₁ , so that L ₁ -P ₁ +1 box dimension features can be obtained correspondingly, and the smallest and largest among them are respectively recorded as E ₁ and E ₂ ;

(8) Choose any sub-window in W as shown by the rectangle A ₂ B ₂ C ₂ D ₂ , abbreviated as W ₂ , the horizontal length of the sub-window is L ₁ and the vertical length is P ₂ , for this W ₂ , calculate the projection of the gray value of the image pixel along the column direction, that is, superimpose the gray value of the image pixel of each column of the sub-window along the vertical direction to obtain a one-dimensional time series, on the premise that the δ size sequence is 2 to 6 pixels Next, a box dimension feature can be calculated from the time series, and then W ₂ is vertically slid at a fixed step size of 1 to 3 pixels each time to traverse the entire W, and there are a total of L ₂ -P ₂ +1 W ₂ , so that L ₂ -P ₂ +1 box dimension features can be obtained correspondingly, and the minimum and maximum among them are respectively recorded as E ₃ and E ₄ ;

(9) Obtain the feature vector [S _h S _v E ₁ E ₂ E ₃ E ₄ ] representing the fabric texture.

Example 1

(1) Acquire the fabric image W, the size of which is 64×64 pixels, as shown in Figure 2.

(2) Implement Sobel operator horizontal filtering on W to obtain the image shown in Figure 3, denoted as W _h .

(3) Implement Sobel operator vertical filtering on W to obtain the image shown in Figure 4, which is denoted as W _v ;.

(4) Shannon entropy is selected as the gray scale statistic, and the calculation formula of Shannon entropy is as follows:

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>$

(5) Calculate the Shannon entropy of W _h , that is _Sh , as the gray level statistical feature of the horizontal edge texture profile, and the result is -5.23×10 ⁷ .

(6) Calculate the Shannon entropy of W _v , that is, S _v , as the gray statistical feature of the vertical edge texture profile, and the result is -9.51×10 ⁷ .

(7) Use one-dimensional FFT to perform period calculation on any row of grayscale data in the original image, and obtain a column basic period P ₁ =6 pixels.

(8) Perform period calculation on any column of grayscale data in the original image by using one-dimensional FFT, and obtain a row basic period P ₂ =4 pixels.

(9) In W, take the sub-window W ₁ shown in the rectangle A ₁ B ₁ C ₁ D ₁ in Fig. 1, the horizontal length of the sub-window is 6 pixels and the vertical length is 64 pixels, and the image pixels of each row of the sub-window are gray The degree value can be projected along the horizontal direction to obtain a one-dimensional time series with a length of 64 pixels. Then, according to the principle of the traversal method, when the sub-window sliding fixed step is selected as 1 pixel and the δ size sequence is 2 to 6 pixels, it can be Based on a series of one-dimensional time series, a series of corresponding box dimensions are obtained, and the minimum and maximum are 1.42 and 1.48, respectively.

(10) In W, take the sub-window W ₂ shown in the rectangle A ₂ B ₂ C ₂ D ₂ in Figure 1, the horizontal length of the sub-window is 64 pixels and the vertical length is 4 pixels, and the image pixels of each row of the sub-window are gray The degree value can be projected along the longitudinal direction to obtain a one-dimensional time series with a length of 64 pixels, and then according to the principle of the traversal method, when the sub-window sliding fixed step is selected as 1 pixel and the δ size sequence is 2 to 6 pixels, it can be Based on a series of one-dimensional time series, a series of corresponding box dimensions are obtained, and the minimum and maximum are 1.44 and 1.47, respectively.

(11) Finally, the mixed feature vector [-5.23×10 ⁷ -9.51×10 ⁷ 1.421.481.441.47] representing the fabric texture is obtained.

Example 2

(1) Acquire the fabric image W, the size of which is 64×64 pixels, as shown in Figure 5.

(2) Implement Sobel operator horizontal filtering on W to obtain the image shown in Fig. 6, denoted as W _h .

(3) Implement Sobel operator vertical filtering on W to obtain the image shown in Figure 7, which is denoted as W _v ;.

(4) Select the gray-scale mean as the gray-scale statistic, and the calculation formula of the gray-scale mean is as follows:

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

(5) Calculate the mean value of the gray value of W _h , that is _Sh , as the gray statistical feature of the horizontal edge texture profile, and the result is 86.84.

(6) Calculate the mean gray value of W _v , namely S _v , as the gray statistical feature of the vertical edge texture profile, and the result is 88.69.

(7) Perform period calculation on any row of grayscale data in the original image by using one-dimensional FFT to obtain a column basic period P ₁ =20 pixels.

(8) Perform cycle calculation on any column of grayscale data in the original image by using one-dimensional FFT, and obtain a row basic cycle P ₂ =11 pixels.

(9) In W, take the sub-window W ₁ shown in the rectangle A ₁ B ₁ C ₁ D ₁ in Figure 1, the horizontal length of the sub-window is 20 pixels and the vertical length is 64 pixels, and the image pixels of each row of the sub-window are gray The degree value can be projected along the horizontal direction to obtain a one-dimensional time series with a length of 64 pixels. Then, according to the principle of the traversal method, when the sub-window sliding fixed step is selected as 2 pixels and the δ size sequence is 2 to 6 pixels, it can be Based on a series of one-dimensional time series, a series of corresponding box dimensions are obtained, and the minimum and maximum are 1.33 and 1.45, respectively.

(10) In W, take the sub-window W ₂ shown in the rectangle A ₂ B ₂ C ₂ D ₂ in Figure 1, the horizontal length of the sub-window is 64 pixels and the vertical length is 11 pixels, and the image pixels of each row of the sub-window are gray The degree value can be projected along the longitudinal direction to obtain a one-dimensional time series with a length of 64 pixels, and then according to the principle of the traversal method, when the sub-window sliding fixed step is selected as 1 pixel and the δ size sequence is 2 to 6 pixels, it can be Based on a series of one-dimensional time series, a series of corresponding box dimensions are obtained, and the minimum and maximum are 1.30 and 1.47, respectively.

(11) Finally, the mixed feature vector [86.8488.691.331.451.301.47] representing the fabric texture is obtained.

Example 3

(1) Acquire the fabric image W, the size of which is 64×64 pixels, as shown in Figure 8.

(2) Implement Sobel operator horizontal filtering on W to obtain the image shown in Fig. 9, denoted as W _h .

(3) Implement Sobel operator vertical filtering on W to obtain the image shown in Figure 10, which is denoted as W _v ;.

(4) Select the grayscale standard deviation as the grayscale statistic, and the formula for calculating the grayscale standard deviation is as follows:

standard deviation $x\left(\mathrm{the\; s}\right)=\sqrt{\frac{1}{\mathrm{no}-1}\underset{i=1}{\overset{\mathrm{no}}{\mathrm{\&Sigma;}}}{({\mathrm{the\; s}}_i-\stackrel{\&OverBar;}{\mathrm{the\; s}})}^2},$ in, $\stackrel{\&OverBar;}{\mathrm{the\; s}}=\frac{1}{\mathrm{no}}\underset{i=1}{\overset{\mathrm{no}}{\mathrm{\&Sigma;}}}{\mathrm{the\; s}}_i$

(5) Calculate the gray standard deviation of W _h , that is _Sh , as the gray statistical feature of the horizontal edge texture profile, and the result is 110.8.

(6) Calculate the gray standard deviation of W _v , namely S _v , as the gray statistical feature of the vertical edge texture profile, and the result is 115.9.

(7) Using one-dimensional FFT to calculate the period of any row of grayscale data in the original image, and obtain the column basic period P ₁ =8 pixels.

(8) Perform period calculation on any column of grayscale data in the original image by using one-dimensional FFT, and obtain a row basic period P ₂ =15 pixels.

(9) In W, take the sub-window W ₁ shown in the rectangle A ₁ B ₁ C ₁ D ₁ in Fig. 1, the horizontal length of the sub-window is 8 pixels and the vertical length is 64 pixels, and the image pixels of each row of the sub-window are gray The degree value can be projected along the horizontal direction to obtain a one-dimensional time series with a length of 64 pixels. Then, according to the principle of the traversal method, when the sub-window sliding fixed step is selected as 2 pixels and the δ size sequence is 2 to 6 pixels, it can be Based on a series of one-dimensional time series, a series of corresponding box dimensions are obtained, and the minimum and maximum are 1.21 and 1.29, respectively.

(10) In W, take the sub-window W ₂ shown in the rectangle A ₂ B ₂ C ₂ D ₂ in Figure 1, the horizontal length of the sub-window is 64 pixels and the vertical length is 15 pixels, and the image pixels of each row of the sub-window are gray The degree value can be projected along the longitudinal direction to obtain a one-dimensional time series with a length of 64 pixels. Then, according to the principle of traversal method, when the sub-window sliding fixed step is selected as 3 pixels and the δ size sequence is 2 to 6 pixels, it can be Based on a series of one-dimensional time series, a series of corresponding box dimensions are obtained, and the minimum and maximum are 1.25 and 1.33, respectively.

(11) Finally, the mixed feature vector [110.8115.91.211.291.251.33] representing the fabric texture is obtained.

Claims (10)

1. A Sobel operator filtering general picture and fractal detail mixed feature vector extraction method for representing fabric texture is characterized in that: the mixed feature vector is composed of two Sobel operator filtering profile features and four fractal detail features;

profile feature extraction: the method comprises the steps that an original fabric image is firstly subjected to Sobel operator horizontal filtering and vertical filtering synchronously and respectively to obtain two corresponding filtering images, and then a characteristic vector representing fabric texture profile information is formed by extracting one gray scale statistic with a consistent calculation mode from the two corresponding filtering images;

extracting detail features: firstly, the basic transverse and longitudinal cycle period sizes of a fabric texture image are solved by adopting one-dimensional fast Fourier transform, then the fractal dimension of each sub-window containing one transverse basic cycle period and the fractal dimension of each sub-window containing one longitudinal basic cycle period in the image are calculated according to the traversing method principle, and finally two fractal dimension extrema which reflect transverse detail information, namely transverse maximum fractal dimension and transverse minimum fractal dimension, and two fractal dimension extrema which reflect longitudinal detail information, namely longitudinal maximum fractal dimension and longitudinal minimum fractal dimension are selected from the image and are used as detail characteristics for representing fabric textures;

wherein

The sub-window containing one transverse basic cycle period is a rectangular window which takes one transverse basic cycle period as a length and takes the width of the fabric texture image as a width, and the fractal dimension of the sub-window of each transverse basic cycle period is calculated on the basis of a corresponding one-dimensional time sequence formed by transversely accumulating the gray values of the image pixels in the sub-window;

the sub-window containing a longitudinal basic cycle period is a rectangular window which is long in the length of a fabric texture image and wide in the longitudinal basic cycle period, and the fractal dimension of the sub-window of each longitudinal basic cycle period is calculated on the basis of a corresponding one-dimensional time sequence formed by longitudinally accumulating the gray values of image pixels in the sub-window;

the extraction process of the mixed feature vector for representing the fabric texture and composed of the Sobel operator filtering profile features and the fractal detail features is as follows:

firstly, acquiring a digital fabric texture image, wherein the image is marked as W, the W is a rectangle, and the length and the width of the size are L₁×L₂I.e. having transverse and longitudinal lengths L, respectively₁And L₂And its basic period in the lateral direction, i.e., column period, is P₁A pixel having a fundamental period in the longitudinal direction, i.e., a line period, of P₂The individual pixel, row period and column period all refer to the number of pixels, P, after rounding₁Basic cycle of any line image pixel gray value set by calculating WIs obtained in phase P₂The method is obtained by calculating the basic cycle period of any column of image pixel gray value set of W;

synchronously and respectively carrying out horizontal filtering and vertical filtering processing of a Sobel operator on the original image, and recording the image subjected to the horizontal filtering of the Sobel operator as W_hThe image after vertical filtering by the Sobel operator is W_v；

Selecting a gray scale statistic, and directly calculating W_hThe gray scale statistic of (2) is taken as the gray scale statistic characteristic of the horizontal edge texture profile and is marked as S_h；

Selecting and calculating W_hThe time-consistent gray scale statistic amount, directly calculates W_vThe gray scale statistic of (2) is taken as the gray scale statistic characteristic of the vertical edge texture profile and is marked as S_v；

In the fabric texture image W, a transverse basic cycle period P is selected₁Is the length and width L of the fabric texture image₂The wide rectangular window is taken as a sub-window containing one transverse basic cycle period and is marked as W₁(ii) a Selecting a length L of a fabric texture image₁Is a long and longitudinal basic cycle period P₂The wide rectangular window is taken as a sub-window containing one longitudinal basic cycle period and is marked as W₂；

For a certain W₁Calculating the gray projection of the image pixels along the row direction, namely superposing the gray values of the image pixels of the rows of the sub-window along the transverse direction to obtain a one-dimensional time sequence, calculating a fractal dimension from the time sequence, and then calculating the W₁Horizontally sliding in fixed steps to traverse the entire W, with a total of L₁-P₁+ 1W₁Thus, L can be obtained accordingly₁-P₁+1 fractal dimensions, respectively noting the minimum and maximum as E₁And E₂The minimum fractal dimension and the maximum fractal dimension reflect the transverse extreme detail information of the texture;

for a certain W₂Calculating the projection of the gray value of the image pixel along the column direction, i.e. superposing the gray value of the image pixel of each column of the sub-window along the longitudinal direction to obtain a one-dimensional imageTime series from which a fractal dimension can be calculated and then W₂Slipping vertically with a fixed step size to traverse the entire W, for a total of L₂-P₂+ 1W₂Thus, L can be obtained accordingly₂-P₂+1 fractal dimensions, respectively noting the minimum and maximum as E₃And E₄The minimum fractal dimension and the maximum fractal dimension in the longitudinal direction are obtained, and the minimum fractal dimension and the maximum fractal dimension in the longitudinal direction reflect the extreme detail information in the longitudinal direction of the texture;

finally, a Sobel operator filtering general picture and fractal detail mixed characteristic vector [ S ] for representing fabric texture is obtained_hS_vE₁E₂E₃E₄]。

2. The method for extracting the hybrid feature vector of the Sobel operator filtering profile and the fractal detail for characterizing the fabric texture as claimed in claim 1, wherein the fabric is woven fabric.

3. The method for extracting the hybrid feature vector of the Sobel operator filtering profile and the fractal detail for characterizing the fabric texture as claimed in claim 1, wherein the transverse direction of the image of the fabric is consistent with the weft direction, and the longitudinal direction of the image of the fabric is consistent with the warp direction.

4. The method for extracting the hybrid feature vector of the Sobel operator filtering profile and the fractal detail for characterizing the fabric texture as claimed in claim 1, wherein the gray scale statistic is the Shannon entropy, the gray scale mean or the gray scale standard deviation.

5. The method for extracting the hybrid feature vector of the Sobel operator filtering profile and the fractal detail for characterizing the fabric texture as claimed in claim 1 or 4, wherein the gray scale statistic is preferably the Xiannon entropy.

6. The method as claimed in claim 1, wherein the fractal dimension is a box dimension for extracting feature vectors of hybrid Sobel operator filtered profiles and fractal details for characterizing fabric texture.

7. The method for extracting the hybrid feature vector of the Sobel operator filtering profile and the fractal detail for characterizing the fabric texture as claimed in claim 6, wherein the delta size sequence used for the box-dimension estimation is 2-6 pixels.

8. The method for extracting the hybrid eigenvector of the Sobel operator filtering profile and the fractal detail for representing the fabric texture as claimed in claim 1, wherein the fixed step length refers to 1-3 pixels.

9. The method for extracting Sobel operator filtering profile and fractal detail mixed feature vector for characterizing fabric texture as claimed in claim 1, wherein the rectangular sub-window W₁Each time of horizontal slip fixed step length and W₂The vertical slip fixed step size need not be the same for each pass.

10. The method for extracting the hybrid feature vector of the Sobel operator filtering profile and the fractal detail for characterizing the fabric texture as claimed in claim 1, wherein the basic cycle period P is₁And P₂The calculation of (c) is performed by means of a one-dimensional fast fourier transform.

Images