Seal recognition system based on machine learning

Abstract

The invention relates to the technical field of seal recognition, in particular to a seal recognition system based on machine learning, which comprises an image acquisition module used for acquiring a pot bottom image of a dark-red enameled pottery; the image analysis module is used for analyzing the image parameter characteristics and the extraction gradient; the image processing module is used for carrying out region extraction on the kettle bottom image according to the extraction gradient so as to obtain a seal region, and is also used for analyzing region change characteristics and seal shape characteristics; the feature analysis module is used for analyzing image gray features; the feature storage module is used for analyzing the seal associated features and storing the seal associated features; the recognition acquisition module is used for acquiring a to-be-recognized image; the recognition analysis module is used for analyzing the to-be-recognized gray features and the to-be-recognized shape features according to the to-be-recognized image; and the seal identification module is used for identifying the seal. According to the invention, accurate identification of the dark-red enameled pottery seal image is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of seal recognition, and in particular to a seal recognition system based on machine learning. Background Art

[0002] As a traditional Chinese handicraft, the seal on the purple clay teapot is an important basis for identifying the authenticity of the work and tracing the author's information. However, due to the different fonts and writing methods of the seals, and the influence of factors such as the material of the purple clay teapot and wear during use, seal recognition has become a challenging task. Traditional image processing methods are difficult to effectively deal with these complex situations, so a more intelligent and efficient recognition algorithm is needed.

[0003] Chinese Patent Publication No.: CN106570475A discloses a method for searching for purple clay teapot seals. First, the purple clay teapot seal image obtained by taking a photo with a mobile phone is cropped and scaled, and then the brightness and contrast of the image are changed by adjusting the mean and mean square error of the image to obtain a standardized image. The edge features and shape area features of the purple clay teapot seal are extracted from the standardized image, and after feature fusion, the seal is roughly divided into four categories: circle, triangle, rectangle, and irregular shape using a trained SVM classifier. The SIFT features of the purple clay teapot seal image are extracted, geometric constraints are introduced, and an improved matching method based on belief propagation is used to match the image in the template library; at the same time, a clustering method is used to optimize the time required for retrieval to complete the seal retrieval. The invention realizes the extraction and classification of the physical features of the purple clay teapot image, but does not realize the comprehensive analysis of the gradient of the grayscale value of the seal area in the purple clay teapot image and the morphological changes of the seal area. There are problems such as low efficiency in analyzing the purple clay teapot image data and inaccurate recognition of the purple clay teapot seal. Summary of the invention

[0004] The object of the present invention is to provide a seal recognition system based on machine learning to solve at least one of the problems existing in the prior art.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A seal recognition system based on machine learning, comprising:

[0007] An image acquisition module is used to acquire an image of the bottom of a purple clay teapot;

[0008] An image analysis module, used to analyze image parameter features according to the pot bottom image, and also used to analyze the extracted gradient according to the image parameter features;

[0009] An image processing module is used to extract the region of the pot bottom image according to the extracted gradient to obtain the seal region, and is also used to analyze the regional change characteristics and seal shape characteristics according to the seal region;

[0010] A feature analysis module is used to analyze the grayscale features of the image according to the stamp area and the extracted gradient, and is also used to adjust the analysis process of the grayscale features of the image according to the image parameter features and the regional change features;

[0011] A feature storage module is used to analyze seal-related features according to image grayscale features and seal shape features, and store the seal-related features;

[0012] An identification and acquisition module is used to acquire an image to be identified;

[0013] A recognition and analysis module is used to analyze the grayscale features to be recognized and the shape features to be recognized according to the image to be recognized;

[0014] The seal recognition module is used to recognize the seal according to the grayscale features to be recognized, the shape features to be recognized and the seal-related features.

[0015] Furthermore, the image analysis module is provided with a grayscale analysis unit, which is used to calculate the image parameter characteristics according to the grayscale value of the pot bottom image through a grayscale analysis formula. The grayscale analysis unit is provided with a grayscale analysis formula as follows:

[0016]

[0017] Where A represents the image parameter feature, L(x,y) represents the grayscale value of each pixel in the pot bottom image, a represents the grayscale mean of the pot bottom image, X represents the number of pixels in the x-axis direction of the pot bottom image, and Y represents the number of pixels in the y-axis direction of the pot bottom image.

[0018] Furthermore, the image analysis module is also provided with a gradient analysis unit, which is used to calculate and extract the gradient according to the first mean value, the second mean value and the image feature parameter through a gradient analysis formula, and the gradient analysis unit is provided with a gradient analysis formula as follows:

[0019] Q=A×{[(Na1+Na2) / (X×Y)+Na2]+[(Na1+Na2) / (X×Y)+Na1]} / 2;

[0020] Among them, Q represents the extracted gradient, Na1 represents the first mean number, and Na2 represents the second mean number.

[0021] Furthermore, the image processing module is provided with a binarization processing unit, which is used to perform binarization on the pot bottom image according to the extracted gradient. When L(x, y)>aA×α, the binarization processing unit sets the grayscale value of the current analysis pixel to 255; when L(x, y)≤aA×α, the binarization processing unit sets the grayscale value of the current analysis pixel to 0, wherein α represents the extraction parameter, α∈Z, -5≤α≤5.

[0022] Furthermore, the image processing module is also provided with an image extraction unit, which is used to count the number of pixels with grayscale values ​​of 255 and 0 in the bottom image of the pot after binarization processing as the first pixel number and the second pixel number respectively, and analyze the seal area according to the first pixel number and the second pixel number. When NL1(α)>NL2(α), the image extraction unit takes the pixel points with grayscale values ​​of 0 in the bottom image of the pot after binarization processing as the pending area; when NL1(α)≤NL2(α), the image extraction unit takes the pixel points with grayscale values ​​of 255 in the bottom image of the pot after binarization processing as the pending area.

[0023] Furthermore, the image processing module is also provided with a regional analysis unit, which is used to analyze the seal area according to the regional change characteristics and the pending area. When R(α)≤r, the regional analysis unit uses the current analysis pending area as the combined area; when R(α)>r, the regional analysis unit uses the current analysis pending area as the non-combined area; the regional analysis unit extracts pixel points of the combined area in the seal image as the seal area.

[0024] Further, the image processing module is further provided with a shape analysis unit, which is used to analyze the shape feature of the seal according to the seal area and the rotation parameter, and the shape analysis unit extracts the number of pixels with the smallest horizontal coordinate in the seal area after the rotation operation as the minimum horizontal number, recorded as NX1(θ), extracts the number of pixels with the largest horizontal coordinate in the seal area after the rotation operation as the maximum horizontal number, recorded as NX2(θ), extracts the number of pixels with the smallest vertical coordinate in the seal area after the rotation operation as the minimum vertical number, recorded as NY1(θ), and extracts the number of pixels with the largest vertical coordinate in the seal area after the rotation operation as the maximum vertical number, recorded as NY2(θ), wherein θ represents the rotation parameter, θ={0,45,90,xy}, and xy represents the mirror parameter;

[0025] The shape analysis unit calculates the shape feature of the seal according to the minimum horizontal number, the maximum horizontal number, the minimum vertical number and the maximum vertical number through a shape analysis formula. The shape analysis unit is provided with a shape analysis formula as follows:

[0026]

[0027] NXY1(θ)=N×1(θ)+N×2(θ)+NY1(θ)+NY2(θ),

[0028] NXY2(θ)=N×1(θ)×N×2(θ)×NY1(θ)×NY2(θ),

[0029] Among them, H represents the shape characteristics of the seal, H1 represents the quantity fluctuation parameter, N×Y1(θ) represents the sum of the horizontal and vertical quantities, and N×Y2(θ) represents the product of the horizontal and vertical quantities.

[0030] Furthermore, the feature analysis module is further provided with a feature analysis unit, which is used to calculate the image grayscale feature according to the seal area and the extracted gradient through a feature analysis formula, and the feature analysis unit is provided with a feature analysis formula as follows:

[0031]

[0032] Among them, G represents the grayscale feature of the image, L(x1,y1) represents the grayscale value of the pixel points in the seal area, NU1 represents the number of pixels in the seal area, and g represents the grayscale mean of the seal area.

[0033] Furthermore, the feature analysis module is further provided with a parameter analysis unit, which is used to adjust the analysis process of the image grayscale feature according to the image parameter feature. When A / a≤β, the parameter analysis unit determines that the image parameter feature meets the threshold value and does not adjust the analysis process of the image grayscale feature; when A / a>β, the parameter analysis unit determines that the image parameter feature does not meet the threshold value and adjusts the analysis process of the image grayscale feature. The adjusted image grayscale feature is G1, where β represents the parameter comparison threshold value;

[0034] The feature analysis module is further provided with a change analysis unit, which is used to optimize the adjustment process of the image grayscale feature according to the regional change feature. When W≤w, the change analysis unit determines that the regional change feature meets the threshold and does not optimize the adjustment process of the image grayscale feature; when W>w, the change analysis unit determines that the regional change feature does not meet the threshold and optimizes the adjustment process of the image grayscale feature. The optimized image grayscale feature is G2.

[0035] Wherein, W represents the change parameter, and w represents the change comparison threshold.

[0036] Furthermore, the feature storage module sets the seal-related feature to F(G)=H according to the image grayscale feature and the seal shape feature;

[0037] The seal recognition module extracts the seal-related features that satisfy d1≤G / D1≤d2 from the seal-related features as the recognition matching features, and recognizes the seal according to the recognition matching features and the shape features to be recognized. When F(i) min ≤D2≤F(i) max When d1≤∑F(i) / (N i×D2)≤d2, the seal recognition module extracts the extraction area in the image to be recognized as the seal in the image to be recognized; when D2<when F(i) min or D2>F(i) max or ∑F(i) / (Ni×D2)<d1 or ∑F(i) / (Ni×D2)>d2, the seal recognition module determines that the extracted area in the image to be recognized is not a seal; wherein F(i) represents the recognition matching feature, i represents the image grayscale feature of the seal-related feature that satisfies d1≤G / D1≤d2, and F(i) min Indicates the minimum value of the corresponding seal shape feature in the recognition matching feature, F(i) max represents the maximum value of the seal shape feature corresponding to the recognition matching feature, Ni represents the number of recognition matching features, d1 represents the first recognition comparison threshold, and d2 represents the second recognition comparison threshold.

[0038] The beneficial effects of the present invention are as follows: the image of the bottom of the purple clay teapot is acquired by the image acquisition module, and the bottom of the teapot is analyzed by other modules to analyze the seal-related features of the bottom of the teapot, and the image features of the image to be identified are acquired and analyzed for matching and comparison with the seal-related features to realize the recognition of the seal in the image to be identified, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 Schematic diagram of the structure of the seal recognition system based on machine learning in this embodiment.

[0041] Figure 2 Schematic diagram of the structure of the image analysis module of this embodiment.

[0042] Figure 3 FIG. 4 is a schematic diagram of the structure of the image processing module of this embodiment.

[0043] Figure 4Schematic diagram of the structure of the feature analysis module of this embodiment. DETAILED DESCRIPTION

[0044] In order to more clearly illustrate the present invention, the present invention is further described below in conjunction with preferred embodiments and accompanying drawings. Similar components in the accompanying drawings are represented by the same reference numerals. It should be understood by those skilled in the art that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the scope of protection of the present invention.

[0045] It should be noted that, although the terms first, second, third, etc. may be used to describe in the embodiments of the present application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of the present application, the first may also be referred to as the second, and similarly, the second may also be referred to as the first.

[0046] See also Figure 1 As shown, it is a seal recognition system based on machine learning in this embodiment, including:

[0047] An image acquisition module is used to acquire the bottom image of the purple clay teapot, wherein the bottom image of the teapot is acquired by user interactive input;

[0048] The image analysis module is used to analyze the image parameter characteristics according to the pot bottom image, and is also used to analyze the extracted gradient according to the image parameter characteristics. The image analysis module is connected to the image acquisition module.

[0049] Specifically, in this embodiment, when analyzing the image of the bottom of the pot, the pixel point in the lower left corner of the bottom of the pot is taken as the coordinate origin, and the two sides adjacent to the coordinate origin are taken as the x-axis and y-axis. The x-axis increases from left to right, and the y-axis increases from bottom to top. The unit length is 1 pixel. A plane rectangular coordinate system is established, and the coordinate points are used to represent the position of each pixel in the bottom of the pot image.

[0050] See also Figure 2 As shown, the image analysis module includes:

[0051] Grayscale analysis unit, used to analyze image parameter features, use image parameter features to represent the change of grayscale value of each pixel in the image, and improve the accuracy of the system's analysis of image parameters such as image size and resolution;

[0052] The grayscale analysis unit calculates the image parameter feature according to the grayscale value of the pot bottom image through a grayscale analysis formula. The grayscale analysis unit is provided with a grayscale analysis formula as follows:

[0053]

[0054] Wherein, A represents the image parameter feature, L(x,y) represents the grayscale value of each pixel in the pot bottom image, a represents the grayscale mean value of the pot bottom image, X represents the number of pixels in the x-axis direction of the pot bottom image, and Y represents the number of pixels in the y-axis direction of the pot bottom image; the grayscale analysis unit analyzes the grayscale value of the pot bottom image to analyze the image feature parameters, thereby realizing the analysis of the grayscale features of the pot bottom image, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition

[0055] Please continue reading Figure 2 As shown, the image analysis module also includes:

[0056] A gradient analysis unit, used for analyzing and extracting gradients, and realizing characteristic analysis of the gradient of grayscale changes of each pixel in the pot bottom image, the gradient analysis unit being connected to the grayscale analysis unit;

[0057] The gradient analysis unit extracts the number of pixels in the pot bottom image that satisfy L(x, y)>a as the first mean number, and extracts the number of pixels in the pot bottom image that satisfy L(x, y)≤a as the second mean number.

[0058] The gradient analysis unit calculates and extracts the gradient according to the first mean value, the second mean value and the image feature parameter through a gradient analysis formula. The gradient analysis unit is provided with a gradient analysis formula as follows:

[0059] Q=A×{[(Na1+Na2) / (X×Y)+Na2]+[(Na1+Na2) / (X×Y)+Na1]} / 2;

[0060] Among them, Q represents the extracted gradient, Na1 represents the first mean number, and Na2 represents the second mean number.

[0061] The gradient analysis unit compares the grayscale values ​​of the pixels in the pot bottom image to extract the first mean number and the second mean number, classifies the pixels in the pot bottom image according to the grayscale value, and analyzes the extracted gradient according to the number of pixels with different grayscale in the grayscale image, thereby realizing the analysis of the gradient of the grayscale value changes of the pixels in the pot bottom image, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition.

[0062] Please continue reading Figure 1 As shown, the seal recognition system based on machine learning also includes:

[0063] The image processing module is used to extract the area of ​​the pot bottom image according to the extracted gradient to obtain the seal area, and is also used to analyze the regional change characteristics and seal shape characteristics according to the seal area. The image processing module is connected to the image analysis module.

[0064] See also Figure 3 As shown, the image processing module includes:

[0065] A binarization processing unit, used for binarizing the pot bottom image;

[0066] The binarization processing unit performs binarization processing on the pot bottom image according to the extracted gradient. When L(x,y)>aA×α, the binarization processing unit sets the grayscale value of the current analysis pixel to 255. When L(x,y)≤aA×α, the binarization processing unit sets the grayscale value of the current analysis pixel to 0, wherein α represents the extraction parameter, α∈Z, -5≤α≤5. In this embodiment, the binarization processing unit analyzes the extracted gradient to perform binarization processing on the seal image, binarizes the grayscale value of each pixel in the image, increases the difference between the image analysis data, and improves the accuracy of the image analysis, thereby improving the efficiency of the system in analyzing the purple clay teapot image data and improving the accuracy of the purple clay teapot seal recognition.

[0067] Please continue reading Figure 3 As shown, the image processing module also includes:

[0068] An image extraction unit is used to extract the undetermined area in the pot bottom image after the binarization processing, increase the number of samples analyzed by the system, and distinguish the seal area in the undetermined area. The image extraction unit is connected to the binarization processing unit;

[0069] The image extraction unit counts the number of pixels with grayscale values ​​of 255 and 0 in the bottom image of the pot after binarization processing as the first pixel number and the second pixel number, respectively, and analyzes the seal area according to the first pixel number and the second pixel number. When NL1(α)>NL2(α), the image extraction unit takes the pixel points with grayscale values ​​of 0 in the bottom image of the pot after binarization processing as the pending area; when NL1(α)≤NL2(α), the image extraction unit takes the pixel points with grayscale values ​​of 255 in the bottom image of the pot after binarization processing as the pending area; wherein NL1(α) represents the first pixel number, and NL2(α) represents the second pixel number. The image extraction unit analyzes the number of pixels in the bottom image of the pot after binarization processing to extract the pending area in the bottom image of the pot, and takes the pending area as the seal area to be analyzed, thereby increasing the diversity of system analysis, thereby improving the system's analysis efficiency of purple clay teapot image data, and improving the accuracy of purple clay teapot seal recognition.

[0070] Please continue reading Figure 3 As shown, the image processing module also includes:

[0071] A region analysis unit, used to determine the seal region in the to-be-determined region, and to classify and determine the pixel point change characteristics of the to-be-determined region, the region analysis unit being connected to the image extraction unit;

[0072] The regional analysis unit calculates the regional change feature according to the to-be-determined region, the first number of pixels and the second number of pixels through a regional analysis formula. The regional analysis unit is provided with a regional analysis formula as follows:

[0073] R(α)=|NL1(α)-NL1(α-1)| / [NL1(α)+NL2(α)], α>-5;

[0074] Among them, R(α) represents the regional variation characteristics.

[0075] The regional analysis unit analyzes the seal area according to the regional change characteristics and the pending area. When R(α)≤r, the regional analysis unit uses the current analyzed pending area as a combined area; when R(α)>r, the regional analysis unit uses the current analyzed pending area as a non-combined area; the regional analysis unit extracts the pixel points of the combined area in the seal image as the seal area, wherein r represents the regional comparison threshold, 0.2≤r≤0.4; the regional analysis unit analyzes the change in the number of different grayscale pixels in the pending area to analyze the regional change characteristics, thereby realizing the judgment of whether the pixel points extracted according to the extracted gradient are the seal area, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition; it can be understood that in this embodiment, the value of the regional comparison threshold is not specifically limited, and technicians in this field can set it freely, as long as it meets the extraction of the seal area. The optimal value of the regional comparison threshold is: r=0.3.

[0076] Please continue reading Figure 3 As shown, the image processing module also includes:

[0077] A shape analysis unit, used to analyze the shape features of the seal, and to use the shape features of the seal to represent the position features of the edge pixels of the seal area, so as to analyze the morphology of the seal area. The shape analysis unit is connected to the area analysis unit;

[0078] The shape analysis unit analyzes the shape feature of the seal according to the seal area and the rotation parameter. The shape analysis unit extracts the number of pixels with the smallest horizontal coordinate in the seal area after the rotation operation as the minimum horizontal number, recorded as NX1(θ), extracts the number of pixels with the largest horizontal coordinate in the seal area after the rotation operation as the maximum horizontal number, recorded as NX2(θ), extracts the number of pixels with the smallest vertical coordinate in the seal area after the rotation operation as the minimum vertical number, recorded as NY1(θ), and extracts the number of pixels with the largest vertical coordinate in the seal area after the rotation operation as the maximum vertical number, recorded as NY2(θ), wherein θ represents the rotation parameter, θ={0,45,90,xy}, and xy represents the mirror parameter;

[0079] The shape analysis unit calculates the shape feature of the seal according to the minimum horizontal number, the maximum horizontal number, the minimum vertical number and the maximum vertical number through a shape analysis formula. The shape analysis unit is provided with a shape analysis formula as follows:

[0080]

[0081] NXY1(θ)=NX1(θ)+NX2(θ)+NY1(θ)+NY2(θ),

[0082] NXY2(θ)=NX1(θ)×NX2(θ)×NY1(θ)×NY2(θ),

[0083] Among them, H represents the shape feature of the seal, H1 represents the quantity fluctuation parameter, N×Y1(θ) represents the sum of the horizontal and vertical quantities, and N×Y2(θ) represents the product of the horizontal and vertical quantities; the shape analysis unit rotates the seal area and extracts and compares the edge pixels to extract multiple groups of pixel points in the seal area as pixel points for shape feature analysis, thereby increasing the diversity of sample data, thereby improving the system's analysis efficiency of purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition.

[0084] Specifically, the seal area after the rotation operation in this embodiment is the seal area obtained by rotating the seal area according to the rotation parameter. The rotation operation is to rotate the seal area clockwise. When θ=0, the seal area is not rotated clockwise. When θ=45, the seal area is rotated 45 degrees clockwise. When θ=90, the seal area is rotated 90 degrees clockwise. When θ=xy, the seal area is rotated clockwise by x=(x min +x max ) / 2 is the axis to flip the stamp area left and right.

[0085] Please continue reading Figure 1 As shown, the seal recognition system based on machine learning also includes:

[0086] A feature analysis module, used for analyzing the grayscale features of the image according to the seal area and the extracted gradient, and for adjusting the analysis process of the grayscale features of the image according to the image parameter features and the regional change features, the feature analysis module being connected to the image processing module;

[0087] See also Figure 4 As shown, the feature analysis module includes:

[0088] A feature analysis unit is used to analyze the grayscale features of the image, use the image grayscale features to represent the grayscale difference features of the pixels in the seal area, and set the order of magnitude of the analyzed image to improve the accuracy of data analysis;

[0089] The feature analysis unit calculates the image grayscale feature through a feature analysis formula according to the seal area and the extracted gradient. The feature analysis unit is provided with a feature analysis formula as follows:

[0090]

[0091] Among them, G represents the grayscale feature of the image, L(x1, y1) represents the grayscale value of the pixel points in the seal area, NU1 represents the number of pixels in the seal area, and g represents the grayscale mean of the seal area. The feature analysis unit analyzes the seal area and extracts the gradient to analyze the image grayscale feature, and constructs the order of magnitude difference between the grayscale features of different images to ensure the number of image features stored in the system, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition.

[0092] Please continue reading Figure 4 As shown, the feature analysis module also includes:

[0093] A parameter analysis unit, used for adjusting the analysis process of the image grayscale feature, so that the adjusted image grayscale feature is related to the acquired parameter data of the image, thereby increasing the analysis accuracy of the system on the image feature, and the parameter analysis unit is connected to the feature analysis unit;

[0094] The parameter analysis unit adjusts the analysis process of the image grayscale feature according to the image parameter feature. When A / a≤β, the parameter analysis unit determines that the image parameter feature meets the threshold and does not adjust the analysis process of the image grayscale feature. When A / a>β, the parameter analysis unit determines that the image parameter feature does not meet the threshold and adjusts the analysis process of the image grayscale feature. The adjusted image grayscale feature is G1, and G1=G / (1+A / a) is set, where β represents the parameter comparison threshold, 0.05≤β≤0.1. The parameter analysis unit analyzes the image parameter characteristics to adjust the analysis process of the image grayscale characteristics, reduce the influence of the grayscale change amplitude of the pixels in the seal image on the image grayscale characteristic analysis, and improve the accuracy of the system's image grayscale characteristic analysis, thereby improving the system's analysis efficiency of the purple clay teapot image data and the accuracy of purple clay teapot seal recognition. It can be understood that in this embodiment, the value of the parameter comparison threshold is not specifically limited, and technicians in this field can set it freely, as long as the adjustment of the image grayscale characteristic analysis process is satisfied. The optimal value of the parameter comparison threshold is: β=0.08.

[0095] Please continue reading Figure 4 As shown, the feature analysis module also includes:

[0096] A change analysis unit is used to optimize the adjustment process of the image grayscale feature, so that the optimized image grayscale feature is related to the change data of the pixel points in the seal extraction process, thereby increasing the system's analysis accuracy of the image feature. The change analysis unit is connected to the parameter analysis unit;

[0097] The change analysis unit optimizes the adjustment process of the image grayscale feature according to the regional change feature. When W≤w, the change analysis unit determines that the regional change feature meets the threshold and does not optimize the adjustment process of the image grayscale feature. When W>w, the change analysis unit determines that the regional change feature does not meet the threshold and optimizes the adjustment process of the image grayscale feature. The optimized image grayscale feature is G2, and G2=G1×e 1-W ; Where W represents the variable parameter, set:

[0098] w represents the change comparison threshold, 0.7≤w≤0.9. It is understandable that the value of the change comparison threshold is not specifically limited in this embodiment, and those skilled in the art can freely set it as long as the optimization of the image grayscale feature adjustment process is satisfied. The best value of the change comparison threshold is: w=0.8.

[0099] Please continue reading Figure 1 As shown, the seal recognition system based on machine learning also includes:

[0100] The feature storage module is used to analyze the seal-related features according to the image grayscale features and the seal shape features, and store the seal-related features. The feature storage module is connected to the feature analysis module.

[0101] The feature storage module sets the seal association feature as F(G)=H according to the image grayscale feature and the seal shape feature. The feature storage module analyzes the image grayscale feature and the seal shape feature to analyze the seal association feature. The seal association feature is used to represent the functional relationship between the grayscale feature and the shape feature of the seal area in the pot bottom image. The data are matched one by one to facilitate data storage, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition.

[0102] Please continue reading Figure 1 As shown, the seal recognition system based on machine learning also includes:

[0103] An identification and acquisition module is used to acquire an image to be identified, wherein the image to be identified is acquired by user interactive input;

[0104] A recognition and analysis module, used for analyzing the extraction area, the grayscale features to be recognized and the shape features to be recognized according to the image to be recognized, and the recognition and analysis module is connected to the recognition acquisition module;

[0105] The analysis process of the extraction area of ​​the image to be identified by the recognition and analysis module is the same as the analysis process of the seal area of ​​the pot bottom image, which will not be elaborated in detail in this embodiment. The analysis process of the grayscale features to be identified by the recognition and analysis module is the same as the analysis process of the image features of the pot bottom image, which will not be elaborated in detail in this embodiment. The analysis process of the shape features to be identified by the image to be identified is the same as the analysis process of the seal shape features of the pot bottom image, which will not be elaborated in detail in this embodiment.

[0106] Please continue reading Figure 1 As shown, the seal recognition system based on machine learning also includes:

[0107] The seal recognition module is used to recognize the seal according to the extracted area, the grayscale features to be recognized, the shape features to be recognized and the seal-related features. The seal recognition module is connected with the feature storage module and the recognition and analysis module.

[0108] The seal recognition module extracts the seal-related features that satisfy d1≤G / D1≤d2 from the seal-related features as the recognition matching features, and recognizes the seal according to the recognition matching features and the shape features to be recognized. When F(i) min ≤D2≤F(i) maxWhen d1≤∑F(i) / (N i×D2)≤d2, the seal recognition module extracts the extraction area in the image to be recognized as the seal in the image to be recognized; when D2<when F(i) min or D2>F(i) max or ∑F(i) / (N i×D2)<d1 or ∑F(i) / (N i×D2)>d2, the seal recognition module determines that the extracted area in the image to be recognized is not a seal; wherein F(i) represents the recognition matching feature, i represents the image grayscale feature of the seal-related feature satisfying d1≤G / D1≤d2, and F(i) min Indicates the minimum value of the corresponding seal shape feature in the recognition matching feature, F(i) max Represents the maximum value of the corresponding seal shape feature in the identification matching feature, Ni represents the number of identification matching features, d1 represents the first identification comparison threshold, 0.9≤d1<1, d2 represents the second identification comparison threshold, 1<d1≤1.1, the seal recognition module compares the grayscale features to be identified, the shape features to be identified and the seal-related features to identify the seal in the image to be identified, and realizes the recognition of the seal by comparing the features of the image to be identified with the features of the stored seal image, thereby improving the system's analysis efficiency of the purple clay teapot image data and improving the accuracy of purple clay teapot seal recognition. It can be understood that in this embodiment, the value of the identification comparison threshold is not specifically limited, and technical personnel in this field can set it freely, as long as it meets the recognition of the seal. The optimal value of the identification comparison feature is: d1=0.95, d2=1.05.

[0109] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the embodiments here. All obvious changes or modifications derived from the technical solution of the present invention are still within the protection scope of the present invention.

Claims

1. A seal recognition system based on machine learning, characterized in that: include: An image acquisition module is used to acquire an image of the bottom of a purple clay teapot; An image analysis module, used to analyze image parameter features according to the pot bottom image, and also used to analyze the extracted gradient according to the image parameter features; An image processing module is used to extract the region of the pot bottom image according to the extracted gradient to obtain the seal region, and is also used to analyze the regional change characteristics and seal shape characteristics according to the seal region; A feature analysis module is used to analyze the grayscale features of the image according to the stamp area and the extracted gradient, and is also used to adjust the analysis process of the grayscale features of the image according to the image parameter features and the regional change features; A feature storage module is used to analyze seal-related features according to image grayscale features and seal shape features, and store the seal-related features; An identification and acquisition module is used to acquire an image to be identified; The recognition and analysis module is used to analyze the extraction area, the grayscale features to be recognized and the shape features to be recognized according to the image to be recognized; The seal recognition module is used to recognize the seal according to the extracted area, the grayscale features to be recognized, the shape features to be recognized and the seal-related features.

2. The seal recognition system based on machine learning according to claim 1, characterized in that: The image analysis module is provided with a grayscale analysis unit, which is used to calculate the image parameter characteristics according to the grayscale value of the pot bottom image through a grayscale analysis formula. The grayscale analysis unit is provided with a grayscale analysis formula as follows: Where A represents the image parameter feature, L(x,y) represents the grayscale value of each pixel in the pot bottom image, a represents the grayscale mean of the pot bottom image, X represents the number of pixels in the x-axis direction of the pot bottom image, and Y represents the number of pixels in the y-axis direction of the pot bottom image.

3. The seal recognition system based on machine learning according to claim 2 is characterized in that: The image analysis module is further provided with a gradient analysis unit, which is used to calculate and extract the gradient according to the first mean value, the second mean value and the image feature parameter through a gradient analysis formula. The gradient analysis unit is provided with a gradient analysis formula as follows: Q=A×{[(Na1+Na2) / (X×Y)+Na2]+[(Na1+Na2) / (X×Y)+Na1]} / 2; Among them, Q represents the extracted gradient, Na1 represents the first mean number, and Na2 represents the second mean number.

4. The seal recognition system based on machine learning according to claim 3 is characterized in that: The image processing module is provided with a binarization processing unit, which is used to binarize the pot bottom image according to the extracted gradient. If L(x, y)>aA×α, the binarization processing unit sets the grayscale value of the current analysis pixel to 255; if L(x, y)≤aA×α, the binarization processing unit sets the grayscale value of the current analysis pixel to 0, wherein α represents the extraction parameter, α∈Z, -5≤α≤5.

5. The seal recognition system based on machine learning according to claim 4 is characterized in that: The image processing module is also provided with an image extraction unit, which is used to count the number of pixels with grayscale values ​​of 255 and 0 in the bottom image of the pot after binarization processing as the first pixel number and the second pixel number respectively, and analyze the seal area according to the first pixel number and the second pixel number. If NL1(α)>NL2(α), the image extraction unit takes the pixel points with grayscale values ​​of 0 in the bottom image of the pot after binarization processing as the pending area; if NL1(α)≤NL2(α), the image extraction unit takes the pixel points with grayscale values ​​of 255 in the bottom image of the pot after binarization processing as the pending area.

6. The seal recognition system based on machine learning according to claim 5, characterized in that: The image processing module is further provided with a region analysis unit, which is used to analyze the seal region according to the regional change characteristics and the pending region. If R(α)≤r, the region analysis unit uses the currently analyzed pending region as the combined region; If R(α)>r, the region analysis unit takes the current analysis pending region as a non-combined region; and the region analysis unit extracts pixel points of the combined region in the seal image as the seal region.

7. The seal recognition system based on machine learning according to claim 6 is characterized in that: The image processing module is further provided with a shape analysis unit, which is used to analyze the shape feature of the seal according to the seal area and the rotation parameter. The shape analysis unit extracts the number of pixels with the smallest horizontal coordinate in the seal area after the rotation operation as the minimum horizontal number, recorded as NX1(θ), extracts the number of pixels with the largest horizontal coordinate in the seal area after the rotation operation as the maximum horizontal number, recorded as NX2(θ), extracts the number of pixels with the smallest vertical coordinate in the seal area after the rotation operation as the minimum vertical number, recorded as NY1(θ), and extracts the number of pixels with the largest vertical coordinate in the seal area after the rotation operation as the maximum vertical number, recorded as NY2(θ), wherein θ represents the rotation parameter, θ={0,45,90,xy}, and xy represents the mirror parameter; The shape analysis unit calculates the shape feature of the seal according to the minimum horizontal number, the maximum horizontal number, the minimum vertical number and the maximum vertical number through a shape analysis formula. The shape analysis unit is provided with a shape analysis formula as follows: NXY1(θ)=NX1(θ)+NX2(θ)+NY1(θ)+NY2(θ), NXY2(θ)=NX1(θ)×NX2(θ)×NY1(θ)×NY2(θ), Among them, H represents the shape feature of the seal, H1 represents the quantity fluctuation parameter, NXY1(θ) represents the sum of the horizontal and vertical quantities, and NXY2(θ) represents the product of the horizontal and vertical quantities.

8. The seal recognition system based on machine learning according to claim 7, characterized in that: The feature analysis module is further provided with a feature analysis unit, which is used to calculate the image grayscale feature according to the seal area and the extracted gradient through a feature analysis formula. The feature analysis unit is provided with a feature analysis formula as follows: Among them, G represents the grayscale feature of the image, L(x1, y1) represents the grayscale value of the pixel points in the seal area, NU1 represents the number of pixels in the seal area, and g represents the grayscale mean of the seal area.

9. The seal recognition system based on machine learning according to claim 8, characterized in that: The feature analysis module is further provided with a parameter analysis unit, which is used to adjust the analysis process of the image grayscale feature according to the image parameter feature. If A / a≤β, the parameter analysis unit determines that the image parameter feature meets the threshold value, and does not adjust the analysis process of the image grayscale feature; if A / a>β, the parameter analysis unit determines that the image parameter feature does not meet the threshold value, and adjusts the analysis process of the image grayscale feature. The adjusted image grayscale feature is G1, where β represents the parameter comparison threshold value; The feature analysis module is further provided with a change analysis unit, which is used to optimize the adjustment process of the image grayscale feature according to the regional change feature. If W≤w, the change analysis unit determines that the regional change feature meets the threshold and does not optimize the adjustment process of the image grayscale feature; if W>w, the change analysis unit determines that the regional change feature does not meet the threshold and optimizes the adjustment process of the image grayscale feature. The optimized image grayscale feature is G2. Wherein, W represents the change parameter, and w represents the change comparison threshold.

10. The seal recognition system based on machine learning according to claim 9, characterized in that: The feature storage module sets the seal-related feature to F(G)=H according to the image grayscale feature and the seal shape feature; The seal recognition module extracts the seal-related features that satisfy d1≤G / D1≤d2 from the seal-related features as the recognition matching features, and recognizes the seal according to the recognition matching features and the shape features to be recognized. If F(i) min ≤D2≤F(i) max and d1≤∑F(i) / (Ni×D2)≤d2, the seal recognition module extracts the extraction area in the image to be recognized as the seal in the image to be recognized; if D2<when F(i) min or D2>F(i) max Or ∑F(i) / (Ni×D2)<d1 or ∑F(i) / (Ni×D2)>d2, the seal recognition module determines that the extracted area in the image to be recognized is not a seal; wherein F(i) represents the recognition matching feature, i represents the image grayscale feature of the seal-related feature that satisfies d1≤G / D1≤d2, and F(i) min Indicates the minimum value of the corresponding seal shape feature in the recognition matching feature, F(i) max It represents the maximum value of the seal shape feature corresponding to the recognition matching feature, Ni represents the number of recognition matching features, d1 represents the first recognition comparison threshold, and d2 represents the second recognition comparison threshold.

Images