CN105373798A - K neighbor image matting and mathematical morphology-based calligraphy character extracting method - Google Patents

Abstract

The invention discloses a K neighbor image matting and mathematical morphology-based calligraphy character extracting method. According to the method, a reference image is established through K neighbor image matting; guided filtering of different window sizes is performed on the interior and edge of a character respectively through using a mathematical morphology-based edge extraction method according to the characteristic that the change of the shade of the ink of a calligraphy image is complex; and more morphological and expressional information is searched through utilizing image fusion technologies. With the K neighbor image matting and mathematical morphology-based calligraphy character extracting method of the invention adopted, expressional information about the emotion and personality of a calligrapher, which is carried by calligraphy itself can be extracted more accurately; Chinese character information can be extracted more accurately for ancient calligraphy works in China which are complex in the change of the shade of ink and are fuzzy in the edge of typefaces; and the effects of the extraction of the expressional information of Chinese characters are significant.

Description

Calligraphy character extraction method based on K neighbor matting and mathematical morphology

Technical Field

The invention belongs to the technical field of image processing, and relates to a calligraphy art information extraction method based on K neighbor matting and digital morphology in the fields of Chinese calligraphy art research and historical cultural heritage protection, which is used for extracting the expression information of Chinese characters in calligraphy work images.

Background

In the fields of Chinese calligraphy and art research and historical cultural heritage protection, image preprocessing and image segmentation are adopted to more completely and accurately extract the shape and quality and expression information of Chinese characters from ancient Chinese calligraphy works. At present, the method for extracting Chinese character information mainly adopts a method of combining image denoising, edge detection, image segmentation and the like to extract the shape and quality information of Chinese characters.

A method for detecting pictures and characters is disclosed in the patent application of Beijing university (publication No. CN101122952, grant date: 2008, 2 and 13 months, and application date: 2007, 9 and 21 months) of the patent application. The method comprises the steps of firstly, combining edge images of an original image on each color component to obtain an accumulated edge image; setting the edge points in the accumulated edge graph as the corresponding colors of the edge points in the original graph, and decomposing the accumulated edge graph into a plurality of sub-edge graphs by a clustering method according to the difference of the colors of the edge points; and finally, performing horizontal and vertical projection for multiple times in each sub-edge image, performing region segmentation in the vertical direction and the horizontal direction according to the projection image, and positioning character regions in the image. The method can accurately obtain the regional information of the characters in the image and the shape and quality information of the characters, but the key for detecting the character information in the method mainly depends on the edge information of the image, namely the shape and quality information of the characters is mainly concerned, and the spiritual information of the characters is not considered, so that the detected character information is incomplete, and the adverse effect is brought to the research work of the calligraphy works in the later period.

XiaoqingLu et al, in the literature "XiaoqingLu, ZhiTang, Yanliu, Liangcaigao, TingWang, Zhipeng Wang. 'Stroke-based character segmentation of Low-quality ImagesonPinnecticChinese tablet' [ C ],201312 the International conference amplification and recognition", 2013 proposed a method for extracting Chinese characters from low-quality ancient tablet images based on Stroke. The method comprises the following specific steps: (1) carrying out denoising pretreatment on the original stele image; (2) applying a mapping-based segmentation method to the denoised image to obtain an initial segmentation result; (3) setting a minimum intensity threshold value by using a self-adaptive Otsu method to obtain a Stroke filtering mask, and performing filtering processing on the de-noised image by using the mask to obtain strength information of Stroke; (4) the segmentation result obtained in the step 2 and the filtering mask obtained in the step 3 are used for selecting the components with higher Stroke strength as initial seeds; (5) based on the guiding information in the seed window, an iterative process is used for extracting Chinese character information contained in the tablet image; (6) and after the iteration is finished, the obtained segmentation result is the extracted Chinese character information. The method can better process the influence of cracks in the tombstone image on Chinese character extraction, and can more completely extract Chinese character information in the tombstone image. However, the method only focuses on extracting the shape and quality information of the Chinese characters, and is difficult to be applied to extracting the magical information of the Chinese characters in the paste image; in the implementation process of the method, more parameters are obtained according to experience. Therefore, the method has a large limitation.

Disclosure of Invention

Aiming at the problems in the prior art, the invention aims to provide a method for extracting the artistic information of the calligraphy work based on K neighbor matting and mathematical morphology, so as to overcome the defect that the detail information of the Chinese character extraction method adopted in the technology for extracting the artistic information in the calligraphy work is seriously lost, and improve the accuracy of extracting the artistic information of the calligraphy work and the artistic appreciation value.

In order to realize the task, the invention adopts the following technical scheme:

a calligraphy character extraction method based on K neighbor matting and mathematical morphology comprises the following steps:

reading a color image to be processed by using software in a computer;

converting the color image to be processed from the RGB color space to the CTE-Lab color space to obtain an L channel;

extracting Chinese character information in the image by using a K neighbor matting algorithm to obtain a gray image serving as a reference image for extracting calligraphy characters;

after the reference image is binarized, corroding the binarized reference image by using mathematical morphology, and subtracting the corroded reference image from the reference image obtained in the third step to obtain an edge image of the calligraphy character;

step five, performing small-window guide filtering processing on the edge image, and selecting a window with a proper size to perform guide filtering processing on the calligraphy character internal image;

and sixthly, performing pixel-level image fusion on the filtering result of the edge image and the filtering result of the image inside the calligraphy character to finish the process.

Further, the specific process of the third step includes:

step S30, extracting feature vectors using the color and position information of the image, the formula is as follows:

X(i)＝(cos(h),sin(h),s,v,x,y)_i

in the above formula, x (i) is a feature vector, h, s, v are three components of the space of HSV color, respectively, and x, y are position coordinates of a pixel;

step S31, defining a kernel function:

k(i,j)＝1-||X(i)-X(j)||/C

in the above formula, k (i, j) is a kernel function, x (i) and x (j) are different feature vectors, and C is a weight adjustment coefficient for ensuring that k (i, j) belongs to (0, 1);

obtaining a Laplace matrix from a kernel function:

L＝D-A

in the above formula, D is a diagonal matrix, and the elements on the diagonal of DA is a similarity matrix;

step S32, adding user constraint information to obtain a closed solution:

α＝(L+λM)^-1(λV)

in the above formula, M is a diagonal matrix representing the user's label of a known pixel, V is a vector representing the user's label of a foreground region, λ is a constraint coefficient, and L is the brightness of the color image in the Lab color space;

step S33, substituting the value of the closed solution α into the following formula to obtain a reference image:

R＝αf+(1-α)b

in the above formula, R is a reference image, f is an unknown foreground layer, and b is an unknown background layer.

Further, the specific process of the fourth step includes:

step S40, binarizing the reference image

The average pixel value of the reference image is calculated using the following formula:

$u=\frac{\underset{x=1,y=1}{\overset{x=M,y=N}{\Σ}}f(x,y)}{M\×N}$

in the above formula, u represents an average pixel value of the L channel, f (x, y) represents a pixel value of a pixel at coordinates (x, y) in the image, and M and N represent the length and width of the image, respectively;

step S41, setting the optimal threshold value for binarization processing of the reference image as T, and counting the proportion w of pixels with pixel values larger than T in the L channel to the image₁And the proportion w of pixels with pixel values less than or equal to T in the L channel to the image₂And calculating the average pixel value u of the pixels with the pixel values larger than T in the L channel₁And an average pixel value u of pixels having a pixel value of T or less in the L channel₂；

Step S42, traversing each possible value of T, and calculating the inter-class difference value using the following formula:

G＝w₁×(u₁-u)×(u₁-u)+w₂×(u₂-u)×(u₂-u)

in the above formula, G represents a difference value between the target portion and the background portion during the binarization processing, when G reaches a maximum value, an optimal threshold T for binarization can be obtained, and then the reference image is subjected to binarization processing using the following formula:

step S43, corroding the binarized reference image by using a matrix with 3 x 3 structural elements to reduce the edge of the calligraphy character by one pixel, and subtracting the corroded reference image from the reference image to obtain the edge of the calligraphy character; wherein:

the erosion operation is defined as:

RΘB^s＝{Z,B_z∈R}

the mathematical form edge extraction operator is as follows:

ED(R)＝R-(RΘB)

in the above two formulas, B is a structural element of 3 x 3, B_ZAs a result of the translation of the structuring element by Z units, B^SFor a set of structural elements that are symmetric about the origin, ed (R) is the edge of the calligraphic word in the reference image R.

Compared with the prior art, the invention has the following advantages:

1. the method uses the K-nearest neighbor image matting and the guiding filter, well solves the problems that the traditional method lacks gray level details and is difficult to extract fuzzy edges of calligraphy characters, can better extract the form and quality and the expression information of the Chinese characters in the calligraphy works, and improves the integrity of Chinese character information extraction.

2. The invention uses the edge extraction method based on mathematical morphology, and well solves the problem that the traditional method can not keep sharp edges of Chinese characters. The method ensures that the calligraphy characters keep smooth edges and improves the accuracy of Chinese character information extraction.

3. According to the method, the reference image is established through K-neighbor matting, according to the characteristic that the calligraphy image is complex in ink shade change, guiding filtering of different window sizes is respectively carried out on the inner part and the edge of a font by using an edge extraction method based on mathematical morphology, and more form and expression information is searched by using an image fusion technology, so that the problem that the detail information of a Chinese character extraction method adopted in the art information extraction technology in the existing calligraphy works is seriously lost is solved, and the accuracy of calligraphy work art information extraction and the art appreciation value are improved.

Drawings

FIG. 1 is an overall flow diagram of the process of the present invention;

FIG. 2 is a partial image of a calligraphy image "Songfeng Geshi";

FIG. 3 shows the result after color space conversion and K-nearest neighbor matting, where FIG. 3(a) is the L-channel of the original image and FIG. 3(b) is the reference image;

FIG. 4 shows the result after edge extraction of a reference image, wherein FIG. 4(a) is an internal image and FIG. 4(b) is an edge image;

FIG. 5 shows the result of different window size guided filtering for an intra image and an edge image, where FIG. 5(a) is the intra image and FIG. 5(b) is the edge image;

FIG. 6 is a graph of the results of image fusion;

FIG. 7 is a graph comparing the experimental results of the first set of two image works in simulation experiment 2;

FIG. 8 is a graph comparing the results of the second set of two image works in simulation experiment 2;

FIG. 9 is a graph comparing the results of the simulation experiment 2 for the third set of two image works;

Detailed Description

First, detailed steps

The flow chart of the invention is shown in fig. 1, and the specific process is as follows:

a calligraphy character extraction method based on K neighbor matting and mathematical morphology comprises the following steps:

reading a color image to be processed by using software in a computer; the software used in the computer can adopt Matlab software;

step two, converting the color image to be processed from the RGB color space to the CTE-Lab color space to obtain an L channel, wherein the specific conversion formula is as follows:

X＝0.412453×R+0.357580×G+0.180423×B

Y＝0.212671×R+0.715160×G+0.072169×B

Z＝0.019334×R+0.119193×G+0.950227×B

X₁＝X/(255×0.950456)

Y₁＝Y/255

Z₁＝Z/(255×1.088754)

wherein X, Y and Z respectively represent tristimulus values of CIE1931 standard chromaticity observer spectrum, R, G and B respectively represent red, green and blue channels of the color image in RGB color space, and X₁,Y₁,Z₁Respectively representing the values of X, Y and Z in linear normalization;

if Y >0.008856, then:

f(X₁)＝X₁^(1/3)

f(Y₁)＝Y₁^(1/3)

f(Z₁)＝Z₁^(1/3)

L＝116×f(Y₁)-16

if Y <0.008856, then:

f(X₁)＝7.787×X₁+16/116

f(Y₁)＝7.787×Y₁+16/116

f(Z₁)＝7.787×Z₁+16/116

L＝903.3×Y₁

a＝500×(f(X₁)-f(Y₁))+128

b＝200×(f(Y₁)-f(Z₁))+128

where f (·) is the correction function, L represents the luminance of the color image in the Lab color space, a, b represents the color of the color image in the Lab color space, the positive half axis of a represents red, the negative half axis represents green, the positive half axis of b represents yellow, and the negative half axis represents blue.

Extracting Chinese character information in the image by using a K neighbor matting algorithm to obtain a gray image serving as a reference image for extracting calligraphy characters; the specific process is as follows:

step S30, extracting feature vectors using the color and position information of the image, the formula is as follows:

X(i)＝(cos(h),sin(h),s,v,x,y)_i

in the above formula, x (i) is a feature vector, h, s, v are three components of the space of HSV color, respectively, and x, y are position coordinates of a pixel;

step S31, defining a kernel function:

k(i,j)＝1-||X(i)-X(j)||/C

in the above formula, k (i, j) is a kernel function, x (i) and x (j) are different feature vectors, and C is a weight adjustment coefficient for ensuring that k (i, j) belongs to (0, 1);

obtaining a Laplace matrix from a kernel function:

L＝D-A

in the above formula, D is a diagonal matrix, and the elements on the diagonal of DA is a similarity matrix;

step S32, adding user constraint information to obtain a closed solution:

α＝(L+λM)^-1(λV)

in the above formula, M is a diagonal matrix representing the user's label of a known pixel, V is a vector representing the user's label of a foreground region, λ is a constraint coefficient, and L is the brightness of the color image in the Lab color space;

step S33, substituting the value of the closed solution α into the following formula to obtain a reference image:

R＝αf+(1-α)b

in the above formula, R is a reference image, f is an unknown foreground layer, and b is an unknown background layer.

After the reference image is binarized, corroding the binarized reference image by using mathematical morphology, and subtracting the corroded reference image from the reference image obtained in the third step to obtain an edge image of the calligraphy character; the specific process is as follows:

step S40, binarizing the reference image by using OTSU

The average pixel value of the reference image is calculated using the following formula:

$u=\frac{\underset{x=1,y=1}{\overset{x=M,y=N}{\&Sigma;}}f(x,y)}{M\&times;N}$

in the above formula, u represents an average pixel value of the L channel, f (x, y) represents a pixel value of a pixel at coordinates (x, y) in the image, and M and N represent the length and width of the image, respectively;

step S41, setting the optimal threshold value for binarization processing of the reference image as T, and counting the proportion w of pixels with pixel values larger than T in the L channel to the image₁And the proportion w of pixels with pixel values less than or equal to T in the L channel to the image₂And calculating the average pixel value u of the pixels with the pixel values larger than T in the L channel₁And an average pixel value u of pixels having a pixel value of T or less in the L channel₂(ii) a The correlation formula is:

$w_1=\frac{W_1}{M\&times;N}$

$w_2=\frac{W_2}{M\&times;N}$

$\begin{array}{cc}{u}_1=\frac{\&Sigma;i\&times;n\left(i\right)}{W_1}& i>T\end{array}$

$\begin{array}{cc}{u}_2=\frac{\&Sigma;i\&times;n\left(i\right)}{W_2}& i\&le;T\end{array}$

wherein, W₁And W₂Respectively representing the number of pixels of which the pixel value is greater than T and the number of pixels of which the pixel value is less than or equal to T in an L channel, i represents the pixel value of a pixel in an image, and n (i) represents the number of pixels of which the pixel value is equal to i;

step S42, traversing each possible value of T, and calculating the inter-class difference value using the following formula:

G＝w₁×(u₁-u)×(u₁-u)+w₂×(u₂-u)×(u₂-u)

in the above formula, G represents a difference value between the target portion and the background portion during the binarization processing, when G reaches a maximum value, an optimal threshold T for binarization can be obtained, and then the reference image is subjected to binarization processing using the following formula:

in step S43, in the mathematical morphology calculation, the erosion has an effect of eliminating the object boundary. Corroding the binarized reference image by using a matrix with 3 x 3 structural elements to reduce the edge of the calligraphy character by one pixel, and subtracting the corroded reference image from the reference image to obtain the edge of the calligraphy character; wherein:

the erosion operation is defined as:

RΘB^s＝{Z,B_z∈R}

the mathematical form edge extraction operator is as follows:

ED(R)＝R-(RΘB)

in the above two formulas, B is a structural element of 3 x 3, B_ZAs a result of the translation of the structuring element by Z units, B^SFor a set of structural elements that are symmetric about the origin, ed (R) is the edge of the calligraphic word in the reference image R.

Step five, performing small-window guide filtering processing on the edge image, and selecting a window with a proper size to perform guide filtering processing on the calligraphy character internal image;

in step S50, a small-window guiding filter process is first performed on the edge image ED:

taking the edge image ED as the input image I and the image L channel as the guide image Ig, the guide filter is a linear transformation to the guide image, that is:

$I_o(x,y)=a_k{I}_g(x,y)+b_k,\&ForAll;(x,y)\&Element;{\mathrm{\&omega;}}_k$

wherein, I_o(x, y) is the pixel value at the coordinate position in the filtered output image, a_kAnd b_kIs a linear coefficient, I_g(x, y) is the pixel value at coordinate position (x, y) in the guide image, ω_kA local window centered at a pixel point and having a radius r. In guiding the filtering process to the edge image, in order to keep the edge sharp, the filter is guided by a small window, i.e. a local window ω of r2 is used₂；

In order to make the input image I and the output image I_oThe difference between them is minimal, i.e. it needs to be in the window omega₂To minimize the function of:

E＝∑((I_o(x,y)-I(x,y))²+a_k ²)

＝∑((a_kI_g(x,y)+b_k-I(x,y))²+a_k ²)

where I (x, y) is a pixel value at a coordinate position of (x, y) in the input image, and E is I_oThe difference between (x, y) and I (x, y) is a regularization parameter that prevents excessive values. When E reaches the minimum a_kAnd b_kRespectively as follows:

$a_k={({{\mathrm{\&sigma;}}_k}^2+\mathrm{\&epsiv;})}^{-1}\left(\frac{1}{\left|\mathrm{\&omega;}\right|}\underset{(x,y)\&Element;{\mathrm{\&omega;}}_k}{\&Sigma;}{I}_g\right(x,y)I(x,y)-{\mathrm{\&mu;}}_k{\stackrel{\&OverBar;}{I}}_k)$

$b_k={\stackrel{\&OverBar;}{I}}_k-a_k{\mathrm{\&mu;}}_k$

wherein σ_k ²And mu_kAre respectively at the window omega₂Inner I_gThe mean and variance of (x, y),is I (x, y)) At window omega₂Inner mean, | ω₂Is the window omega₂The number of internal pixel points.

Since a pixel may be covered by multiple windows, the parameter a can be calculated based on_kAnd b_kThe filter output I is calculated by the following formula_o(x,y)：

$I_o(x,y)={\stackrel{\&OverBar;}{a}}_{xy}{I}_g(x,y)+{\stackrel{\&OverBar;}{b}}_{xy}$

Wherein,andis the average of all window coefficients covering pixel (x, y).

From I_o(x, y) leads to a guided filtering result ED' of the edge image ED.

Step S51, performing a guide filtering process on the calligraphy character internal image:

taking the internal image IE as the input image I, taking the image L channel as the guide image Ig, and using a larger filtering window, i.e. using a local window omega of r-8, to extract more gray information₈(ii) a The other processes are the same as step S50, and the guided filtering result IE' of the intra image is obtained.

And sixthly, performing pixel-level image fusion on the filtering result of the edge image and the filtering result of the image inside the calligraphy character to finish the process.

In order to enable the final extraction result to simultaneously contain rich gray information in the calligraphy character and sharp edges of strokes, image fusion is carried out on ED 'and IE' to obtain the complete and accurate shape, quality and verve of the calligraphy character.

The pixel value at (x, y) for ED 'and IE' is taken to be the minimum, namely:

g(x,y)＝min(ED′(x,y),IE′(x,y))

where ED '(x, y) is the pixel value of ED' at coordinate (x, y), IE '(x, y) is the pixel value of IE' at coordinate (x, y), g (x, y) is the pixel value of the image fusion result at coordinate (x, y), and min (·) is the minimum value.

Second, simulation experiment

Simulation experiment 1: the method for extracting the Chinese character shape and character and the expression information in the calligraphy image is simulated.

The simulation condition of simulation 1 was performed under MATLABR2013a software, and the parameter of the pilot filtering was 0.1¹⁰The parameter lambda of the K neighbor matting algorithm is 100, the level is 0.5, and l is 1.

Referring to fig. 2 to 6, a simulation experiment was performed on a part of the "songfeng poem" of the calligraphy image huang court jia. The calligraphy image is well stored, the reference image contains main shape and quality information of Chinese characters, then the reference image is subjected to edge extraction to obtain an internal image and an edge image, the internal image and the edge image are respectively subjected to filtering processing by guide filters with different window sizes, gray level details and sharp edges can be obtained, and filtering results are subjected to image fusion to extract complete expression information, so that the calligraphy image can truly reflect the concentration change of pen and ink and the trend of pen points, and the edge condition of calligraphy characters can be accurately restored. In the result image, the calligraphy owners can better show the true and false of strokes and the sudden change and gradual change of the stroke tips when writing Chinese characters.

Simulation experiment 2, the comparative analysis simulation is performed on the method of the present invention.

The simulation conditions of simulation experiment 2 were performed under MATLABR2013a software, and the parameters of the pilot filtering were 0.1¹⁰The parameter lambda of the K neighbor matting algorithm is 100, the level is 0.5, and l is 1. The method is mainly compared and analyzed with OTSU, FastFuzzyC-means (FFCM) and MultiChanneland guided filters (MCGF), so that the method has obvious advantages in the aspect of extracting the expression information of Chinese characters in the calligraphy works. The comparison and analysis of the experimental results are described below:

referring to fig. 7, fig. 8 and fig. 9 (two calligraphy image works are selected for each group), for the calligraphy images, complete and accurate shape and quality information and spiritual information need to be extracted simultaneously. Firstly, in the aspect of extracting the shape and quality information of the Chinese characters, all the methods can accurately extract the shape and quality information of the Chinese characters. But the shape and quality information of the extraction result of the method of the invention is more complete, as shown in fig. 7 and fig. 9. In the extraction of the artifact information of the chinese characters, for the area with fuzzy edges in the calligraphic characters, the Otsu, FFCM and MCGF extraction results have a large loss of detail, as shown in fig. 8 and 9. For strokes written by a dry pen, the difficulty of extraction is large, and most of the information is lost by OTSU, FFCM and MCGF, as shown in FIG. 7, FIG. 8 and FIG. 9. However, compared with the method, the method has higher accuracy in the aspect of extracting the shape and character information and the expression information, can keep rich gray level details for both the whitening area and the dry stroke, and has clear and complete edges of the strokes.

Claims (3)

1. A calligraphy character extraction method based on K neighbor matting and mathematical morphology is characterized by comprising the following steps:

reading a color image to be processed by using software in a computer;

converting the color image to be processed from the RGB color space to the CTE-Lab color space to obtain an L channel;

extracting Chinese character information in the image by using a K neighbor matting algorithm to obtain a gray image serving as a reference image for extracting calligraphy characters;

after the reference image is binarized, corroding the binarized reference image by using mathematical morphology, and subtracting the corroded reference image from the reference image obtained in the third step to obtain an edge image of the calligraphy character;

step five, performing small-window guide filtering processing on the edge image, and selecting a window with a proper size to perform guide filtering processing on the calligraphy character internal image;

and sixthly, performing pixel-level image fusion on the filtering result of the edge image and the filtering result of the image inside the calligraphy character to finish the process.

2. The method for extracting calligraphy characters based on K-neighbor matting and mathematical morphology as claimed in claim 1, wherein the specific process of the third step comprises:

step S30, extracting feature vectors using the color and position information of the image, the formula is as follows:

X(i)＝(cos(h),sin(h),s,v,x,y)_i

in the above formula, x (i) is a feature vector, h, s, v are three components of the space of HSV color, respectively, and x, y are position coordinates of a pixel;

step S31, defining a kernel function:

k(i,j)＝1-||X(i)-X(j)||/C

in the above formula, k (i, j) is a kernel function, x (i) and x (j) are different feature vectors, and C is a weight adjustment coefficient for ensuring that k (i, j) belongs to (0, 1);

obtaining a Laplace matrix from a kernel function:

L＝D-A

in the above formula, D is a diagonal matrix, and the elements on the diagonal of DA is a similarity matrix;

step S32, adding user constraint information to obtain a closed solution:

α＝(L+λM)^-1(λV)

in the above formula, M is a diagonal matrix representing the user's label of a known pixel, V is a vector representing the user's label of a foreground region, λ is a constraint coefficient, and L is the brightness of the color image in the Lab color space;

step S33, substituting the value of the closed solution α into the following formula to obtain a reference image:

R＝αf+(1-α)b

in the above formula, R is a reference image, f is an unknown foreground layer, and b is an unknown background layer. .

3. The method for extracting calligraphy characters based on K-neighbor matting and mathematical morphology as claimed in claim 1, wherein the specific process of the fourth step comprises:

step S40, binarizing the reference image

The average pixel value of the reference image is calculated using the following formula:

$u=\frac{\underset{x=1,y=1}{\overset{x=M,y=N}{\&Sigma;}}f(x,y)}{M\&times;N}$

in the above formula, u represents an average pixel value of the L channel, f (x, y) represents a pixel value of a pixel at coordinates (x, y) in the image, and M and N represent the length and width of the image, respectively;

step S41, if the optimal threshold for binarization processing of the reference image is T, then the ratio of pixels with pixel values greater than T in the L channel to the image is countedExample w₁And the proportion w of pixels with pixel values less than or equal to T in the L channel to the image₂And calculating the average pixel value u of the pixels with the pixel values larger than T in the L channel₁And an average pixel value u of pixels having a pixel value of T or less in the L channel₂；

Step S42, traversing each possible value of T, and calculating the inter-class difference value using the following formula:

G＝w₁×(u₁-u)×(u₁-u)+w₂×(u₂-u)×(u₂-u)

in the above formula, G represents a difference value between the target portion and the background portion during the binarization processing, when G reaches a maximum value, an optimal threshold T for binarization can be obtained, and then the reference image is subjected to binarization processing using the following formula:

step S43, corroding the binarized reference image by using a matrix with 3 x 3 structural elements to reduce the edge of the calligraphy character by one pixel, and subtracting the corroded reference image from the reference image to obtain the edge of the calligraphy character; wherein:

the erosion operation is defined as:

RΘB^s＝{Z,B_z∈R}

the mathematical form edge extraction operator is as follows:

ED(R)＝R-(RΘB)

in the above two formulas, B is a structural element of 3 x 3, B_ZAs a result of the translation of the structuring element by Z units, B^SFor a set of structural elements that are symmetric about the origin, ed (R) is the edge of the calligraphic word in the reference image R.